Next Article in Journal
Wear and Corrosion Resistance of Thermally Formed Decorative Oxide Layers on Austenitic Steel
Previous Article in Journal
Experimental Investigation of the Interplay Between Al-, B-, and Ti-Nitrides in Microalloyed Steel and Thermodynamic Analysis
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

Research Progress on Process Optimization of Metal Materials in Wire Electrical Discharge Machining

1
College of Water Conservancy Engineering, Yellow River Conservancy Technical University, Kaifeng 475000, China
2
Henan Key Lab of Intelligent Manufacturing of Mechanical Equipment, Zhengzhou University of Light Industry, Zhengzhou 450002, China
3
Guangdong Provincial Key Laboratory of Digital Manufacturing Equipment, Guangdong HUST Industrial Technology Research Institute, Huazhong University of Science and Technology, Dongguan 523808, China
4
Hubei Water Resources and Hydropower Science and Technology Promotion, Hubei Water Resources Research Institute, Wuhan 430070, China
*
Authors to whom correspondence should be addressed.
Metals 2025, 15(7), 706; https://doi.org/10.3390/met15070706
Submission received: 13 May 2025 / Revised: 18 June 2025 / Accepted: 20 June 2025 / Published: 25 June 2025

Abstract

Wire electrical discharge machining (WEDM), as a significant branch of non-traditional machining technologies, is widely applied in fields such as mold manufacturing and aerospace due to its high-precision machining capabilities for hard and complex materials. This paper systematically reviews the research progress in WEDM process optimization from two main perspectives: traditional optimization methods and artificial intelligence (AI) techniques. Firstly, it discusses in detail the applications and limitations of traditional optimization methods—such as statistical approaches (Taguchi method and response surface methodology), Adaptive Neuro-Fuzzy Inference Systems, and regression analysis—in parameter control, surface quality improvement, and material removal-rate optimization for cutting metal materials in WEDM. Subsequently, this paper reviews AI-based approaches, traditional machine-learning methods (e.g., neural networks, support vector machines, and random forests), and deep-learning models (e.g., convolutional neural networks and deep neural networks) in aspects such as state recognition, process prediction, multi-objective optimization, and intelligent control. The review systematically compares the advantages and disadvantages of traditional methods and AI models in terms of nonlinear modeling capabilities, adaptability, and generalization. It highlights that the integration of AI by optimization algorithms (such as Genetic Algorithms, particle swarm optimization, and manta ray foraging optimization) offers an effective path toward the intelligent evolution of WEDM processes. Finally, this investigation looks ahead to the key application scenarios and development trends of AI techniques in the WEDM field for cutting metal materials.

1. Introduction

Wire electrical discharge machining (WEDM) is a non-traditional machining technology that utilizes the principle of electrical discharge to process materials. As shown in Figure 1, it generates high temperatures through pulsed discharges between a wire electrode and the workpiece, causing localized melting or vaporization of the material to achieve cutting [1,2,3]. Specifically, when a pulsed voltage is applied between the wire electrode and the workpiece, the dielectric medium in the gap is broken down, forming a discharge channel. The instantaneous high temperature rapidly melts or vaporizes the material surface. After the discharge ends, the flow of the dielectric fluid and the expansion of bubbles remove the melted or vaporized material particles, achieving gradual material removal and shaping. WEDM is widely used in fields such as mold manufacturing, aerospace, and medical devices due to its characteristics of high precision, excellent surface quality, and capability to machine complex shapes [2,4,5,6,7]. Traditional WEDM processes rely heavily on experienced operators and complex adjustments of processing parameters, making machining efficiency and accuracy highly susceptible to human factors [8,9,10]. With the rapid development of machine learning (ML), and especially deep learning (DL) technologies, the automation and intelligence level of WEDM has significantly improved [11,12]. ML and DL techniques, through learning from large volumes of historical data, can automatically optimize process parameters, improve machining efficiency and accuracy, and reduce human intervention, thereby significantly enhancing the stability and reliability of the WEDM process [13,14,15]. The importance of WEDM lies not only in its high precision and surface quality, but also in its ability to machine materials with complex shapes and hard materials with minimal tool wear. Although its machining efficiency is generally lower than that of conventional methods, such as milling or turning, WEDM remains advantageous in scenarios where traditional techniques face difficulties due to severe tool wear or geometric constraints [16,17,18,19,20,21]. WEDM technology effectively addresses these issues [22]. Furthermore, WEDM can machine complex 3D shapes, a process which is particularly important in mold making and aerospace applications [23,24,25]. With the advancement of Industry 4.0 and smart manufacturing, the demand for automation and intelligence in WEDM is increasing [26,27,28]. The integration of ML and DL technologies allows for more efficient parameter optimization and more precise machining control in WEDM processes, further enhancing its application value. Furthermore, the synergistic development of electrical discharge machining-process optimization and hydraulic machinery manufacturing essentially reflects a breakthrough in the adaptability of modern manufacturing technologies to complex operational conditions. In the field of hydraulic equipment, core components, such as turbines and pump valves, operate under high water pressure, strong corrosion, and long service cycles, which impose multidimensional and composite demands on machining processes.
Although WEDM technology has been widely applied in various fields, its process optimization still faces numerous challenges. Traditional WEDM process optimization mainly relies on empirical formulas and trial-and-error methods, which are not only time-consuming and labor-intensive but also difficult to adapt to the complex and ever-changing processing conditions [29,30,31]. Furthermore, traditional methods often exhibit limitations when dealing with nonlinear, multivariable coupled process parameters, making it difficult to achieve a global optimal solution. The application of ML and DL technologies in WEDM process optimization has, to some extent, addressed the shortcomings of traditional methods, but some issues still remain. For example, ML models require a large amount of high-quality data for training, and obtaining this data in actual production is often costly and time-consuming [32,33,34]. While DL models have powerful nonlinear fitting capabilities, their “black-box” nature makes them difficult to interpret, limiting their effective application in actual production [32,35,36]. In addition, both ML and DL models have limited generalization capabilities, and when processing conditions change, the model’s predictive accuracy may significantly decrease [32,37]. In practical applications, WEDM process optimization also faces other challenges. For example, electrode wire wear and breakage during processing not only affect machining accuracy but also increase production costs [38,39,40]. Additionally, the thermal affected zone and residual stress issues generated during processing can adversely impact the mechanical properties and surface quality of the workpiece [41,42]. These issues are difficult to effectively address in traditional WEDM processes, and while data analysis and model optimization in ML and DL applications can partially alleviate these problems, further research and improvements are still needed. However, from the perspective of process optimization, the introduction of intelligent control systems is driving a transformation of traditional machining paradigms. Dynamic parameter adjustment techniques based on real-time operational feedback enable an automatic balance between machining accuracy and efficiency according to the structural characteristics of hydraulic components, such as curved flow channels and non-standard hole arrays, thereby reducing the risk of localized overheating and micro-defect formation. Innovations in hybrid machining strategies have also gradually emerged. Notably, in turbine-machinery manufacturing, WEDM has been widely applied for the high-precision machining of critical structural features, such as the fir-tree slots on turbine disks. Owing to its excellent contour control and high machining consistency, WEDM demonstrates clear superiority in the precise cutting of such complex curved-slot geometries.
WEDM process optimization can be broadly categorized into three types: traditional optimization methods, ML-based optimization methods, and DL-based optimization methods. Traditional optimization methods primarily include empirical formulas, trial-and-error methods, regression analysis, and response surface methodology (RSM). These methods optimize parameters by establishing mathematical models that relate process parameters to machining outcomes [43,44,45]. However, traditional methods often exhibit limitations when dealing with complex nonlinear problems [46]. ML-based optimization methods, through the introduction of algorithms such as Support Vector Machine (SVM) and random forest (RF), are better equipped to handle multivariable and nonlinear problems [34,47,48]. By learning from historical data, ML methods can automatically identify the complex relationships between process parameters and machining outcomes, thereby achieving more precise process optimization. DL-based optimization methods take process optimization to a higher level of intelligence [49]. DL models, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), can handle even more complex nonlinear relationships and exhibit stronger generalization capabilities [13,50,51]. However, training DL models requires significant computational resources and data, and their lack of interpretability limits their application in real-world production to some extent [29,52,53].

2. Traditional Methods

Process optimization is the core driving force behind the development of WEDM technology (in terms of methodology). The goal is to achieve a comprehensive improvement in processing efficiency, accuracy, and surface quality through multidimensional collaborative control. The performance of WEDM is influenced by dynamic coupling factors, such as pulse parameters (discharge energy) and wire feed (WF) speed, and requires a systematic approach to balance key indicators, such as material-removal rate (MRR), surface roughness (SR), and electrode wear. Traditional optimization methods use design of experiments (DOE) and modeling techniques (such as response surface methodology and grey relational analysis (GRA)) to establish the mapping relationship between process parameters and performance responses, and integrate intelligent algorithms (such as Genetic Algorithms (GAs) and particle swarm optimization (PSO)) for multi-objective optimization. Additionally, the introduction of innovative technologies like ultrasonic vibration assistance and nanofluid media further expands the boundaries of process optimization. This section systematically describes the application of traditional optimization strategies, such as statistical methods and fuzzy algorithms, in WEDM, analyzing their technical principles and adaptability, and providing a theoretical framework and practical guidance for process parameter decision-making under complex working conditions.

2.1. Statistical Methods

WEDM, as a high-precision machining technology, relies heavily on process parameter optimization to enhance processing efficiency and quality. In recent years, many researchers have conducted in-depth studies on WEDM process parameters using experimental design and optimization methods, achieving significant results. Sharma et al. [54] Using Inconel 706, a newly developed nickel-based superalloy specifically designed for aerospace applications, as the experimental material, a hybrid optimization strategy integrating the Taguchi method, GRA, and Principal Component Analysis was employed to simultaneously enhance the MRR and reduce SR. analysis of variance (ANOVA) was conducted to identify the control parameters with statistically significant influence on machining performance. The results demonstrated that the optimal combination of process parameters determined by the Taguchi–GRA–Principal Component Analysis hybrid approach was a pulse-on time of 105 μs, a pulse-off time of 27 μs, a servo voltage of 32 V, and a wire feed rate of 4 m/min. The multi-performance characteristic index improved by 0.0428, validating the reliability of the experimental outcomes. The experimental error was 3.27%, which is within an acceptable range. Ikram et al. [55] applied the Taguchi DOE method to study the effects and optimization of eight process parameters on SR, kerf width (KW), and MRR. They performed ANOVA and signal-to-noise ratio to identify key factors. The experimental results demonstrated that the optimal parameters improved the signal-to-noise ratios of SR, KW, and MRR by 2.972 dB, 1.04 dB, and 0.804 dB, respectively. Subsequently, Bobbili et al. [56] utilized Taguchi’s orthogonal array (L27) to study the effects of process parameters on MRR and Ra and obtained the best process parameter combination through ANOVA. Thankachan et al. [57] combined the Taguchi method with GRA for multi-objective optimization and evaluated the processing characteristics of WEDM. The results indicated that this method could achieve the optimization objectives of maximum MRR (20.19 mm3/min) and minimum Ra (3.01 µm). Kavimani et al. [58] utilized the Taguchi method combined with GRA for multi-objective optimization, selecting control parameters to balance the best MRR with the minimum Ra. The results demonstrated that when P-ON (pulse-on time) = 40 µs, P-OFF (pulse-off time) = 23 µs, WFD = 2 m/min, wt.% = 0.2, and D% = 10, ideal results were achieved. Xiao et al. [59] developed a robust method based on the wolf-pack algorithm (WPA) to optimize WEDM cutting parameters. They determined WPA parameters using the Taguchi orthogonal array, and the experimental results depicted that, compared to the signal-to-noise (S/N) statistical method, SR and MRR were improved by 12% and 3.99%, respectively. Zaman et al. [60] used Taguchi L9 orthogonal DOE to optimize MRR in AISI 1045 medium carbon steel, and they identified key control parameters through ANOVA and S/N ratio. The experimental results showed that, under optimized conditions (current (I) = 16 A, voltage (V) = 50 V, and P-ON = 100 µs), the MRR reached a maximum of 0.7112 mm3/min, as shown in Figure 2a. Also, ANOVA indicated that current had the greatest influence on MRR, with a coefficient of determination (R2) value of 70.23%. Additionally, the application of techniques like the Taguchi method by [61,62,63,64] has also made significant contributions to process-parameter optimization and has played a vital role in enhancing machining performance.
At the same time, advanced optimization methods such as RSM have also been widely applied in this field, providing important support for precision machining of complex materials. Kung and Chiang [44] applied Central Composite Design and RSM to establish a mathematical model to study the effects of peak current (IP), T-ON, duty cycle, and WF rate on the MRR and Ra during the WEDM process of alumina-based ceramics (Al2O3 + Ti C). The experimental results revealed that the model’s predicted values were highly consistent with the experimental values, with an error of 95%. Sarkar et al. [65] adopted RSM to establish a second-order mathematical model based on processing parameters to predict SR, dimensional deviation (DD), and cutting speed (CS). The experimental results demonstrated that the average prediction errors for Ra, DD, and CS were 5.72%, 2.74%, and 7.08%, respectively. Sivaprakasam et al. [66] utilized RSM to model and analyze the Micro-WEDM processing characteristics, optimizing the processing parameters through second-order mathematical models and ANOVA. The results showed that the optimized MRR, KW, and Ra were 0.259943 mm3/min, 87 µm, and 0.97 µm, respectively. Soundararajan et al. [67] employed the central composite rotatable design of RSM to systematically study the effects of WEDM parameters (such as P-ON and P-OFF), and IP) on MRR and Ra. They also adopted the expected function analysis method for multi-objective optimization of process variables. The results depicted (as shown in Figure 2b) that the average error rates for MRR and Ra predicted by RSM were 7.30% and 3.0%, respectively. Wang et al. [10] proposed a method to optimize processing parameters aimed at minimizing corner errors and maximizing processing speed. Through RSM, they analyzed the influence of each parameter and determined the optimal parameters using data and algorithms. After optimization, the corner errors at 30°, 60°, and 90° were reduced to 0.056, 0.024, and 0.011 mm, with prediction accuracies of 99%, 91%, and 90%, respectively. T et al. [68] adopted RSM to optimize process parameters and analyze their impact on machining performance, determining the contribution of each parameter through ANOVA. The experimental results demonstrated that the optimal parameter combination was T-ON = 53 μs, T-OFF = 28 μs, I = 2.65 A, VG = 185 V, and Df = 1.5 LPM, achieving a desirability value of 0.624. The maximum taper angle error was only 3.9%. These studies not only deepened the understanding of the effects of WEDM processing parameters but also improved machining precision and efficiency through multi-objective optimization and ANOVA methods, providing strong theoretical support and practical guidance for real-world production.
Figure 2. Statistical methods: (a) Proposed method for selecting the optimal process parameter settings, experimental physical diagram, and main effect plot of the S/N ratio (Reprinted from Ref. [60]). (b) Overall framework, experimental physical diagram, and comparison between experimental groups and RSM predictions of Ra of A413 workpiece (Reprinted with permission from Ref. [67]. Copyright 2016, Elsevier).
Figure 2. Statistical methods: (a) Proposed method for selecting the optimal process parameter settings, experimental physical diagram, and main effect plot of the S/N ratio (Reprinted from Ref. [60]). (b) Overall framework, experimental physical diagram, and comparison between experimental groups and RSM predictions of Ra of A413 workpiece (Reprinted with permission from Ref. [67]. Copyright 2016, Elsevier).
Metals 15 00706 g002

2.2. Fuzzy Algorithm

In the WEDM machining field, traditional optimization methods often struggle to balance machining accuracy, efficiency, and stability. In recent years, fuzzy logic and intelligent optimization algorithms have gradually become important tools for optimizing WEDM processes due to their good adaptability and nonlinear mapping capabilities. Many researchers have applied methods such as fuzzy control and Adaptive Neuro-Fuzzy Inference Systems (ANFISs) to optimize key machining parameters, improving machining quality and stability. Yan and Liao [69] developed a fuzzy logic-based WEDM monitoring and Adaptive Control Optimization system, which adjusts feed speed and P-OFF in real-time, keeping parameters in an ideal state for stable machining. The results demonstrated (as shown in Figure 3a) that, during finishing, the anomaly rate was adjusted from 25% to the optimal value of 40%, without oscillations, within 150 s. Since the MRR in WEDM is much lower than the servo response speed of the motion system, adjusting the feed rate and pulse interval too quickly during the transient machining stage may lead to an unstable machining process, which may cause problems such as deterioration of the gap conditions, damage to the surface of the workpiece, or even wire breakage. Therefore, the rise time (150 s) shown in the graph is more appropriate for stabilizing and controlling the machining process. When the machining state reaches the set stable operating conditions, the two control parameters will be moderately fine-tuned to achieve the desired machining performance. During rough machining, feed speed was adjusted, P-OFF was stabilized at 19.2 µs, and machining speed was maintained at 100 mm2/min. Lin et al. [70] adopted a fuzzy logic-based control strategy to improve WEDM corner-machining accuracy. The experimental results showed that corner-machining errors could be reduced to below 50% of the normal value during rough machining, with only a maximum increase of 10% in machining time. Fard et al. [71] utilized ANFIS to establish the mapping relationship between process inputs and responses and combined it with the Artificial Bee Colony (ABC) algorithm to optimize process parameters. The results confirmed that the predictions of CS and SR using ANFIS-ABC were highly consistent with the measured values (CS: 4.9929 vs. 5.18; Ra: 3.1931 vs. 3.15). Maher et al. [72] applied ANFIS and the Taguchi method to study the effects of processing parameters such as IP, P-ON, and wire tension (WT). The results showed (as shown in Figure 3b) that low pulse width (0.15 µs), low IP (16 A), and high WT (400 g) could reduce SR and heat-affected zone within a specific testing range. Goyal et al. [73] proposed an intelligent method combining ANFIS and Nondominated Sorting Genetic Algorithm-II (NSGA-II) for modeling and optimizing the WEDM process. The experiments revealed that the ANFIS model predicted MRR and wire-wear rate with average errors of 5.35% and 5.13%, respectively, while the NSGA-II optimization results were consistent with the experiments, with maximum absolute errors of 6.784% for MRR and 7.589% for wire-wear rate. Additionally, studies by [25,74,75], and others have also achieved significant success using fuzzy control methods.

2.3. Other

In addition to the above two methods, other approaches, such as regression analysis, have also made significant contributions to the optimization of WEDM processes. For example, Tosun [76] established a model for the relationship between CS and SR with changes in parameters through regression analysis and, based on this model, optimized the multiple cutting performance of WEDM. The results showed that the computed values were highly consistent with the experimental values, with a difference of no more than 10%. Yuan et al. [77] proposed a Gaussian Process Regression (GPR)-based prediction reliability optimization and optional decision method for optimizing the WEDM-HS process. The results indicated that the mean squared errors (MSEs) for predicting MRR and Ra were 1.2505 and 0.0229, respectively. Pasam et al. [78] developed a mathematical model for control parameters and surface quality through linear regression analysis, and they optimized the SR prediction model using a GA. The results showed that the regression coefficient was 0.943, and after optimization, the best control parameters achieved an SR of 1.85 µm in the WEDM processing of Ti6Al4V alloy. Ikram et al. [55] established linear regression and additive models for SR, KW, and MRR. The results showed that the average prediction errors for SR and KW were 1.57% and 1.26%, respectively, and the average prediction error for MRR was 2.24%, with the experimental and predicted values being highly consistent. Our team, Ming et al. [79], combined GPR and NSGA-II to develop a multi-objective optimization model for YG15. The experimental results demonstrated that the relative errors of surface peak-to-peak value (SZ), surface root mean square value (SQ), and MRR were all within 20%, which is within a reasonable range. Nguyen and Duong [80] proposed a multi-response optimization method, developing a Kriging model based on machining factors such as Ton, current, voltage, and wire speed (WS), and used an Adaptive Memetic Genetic Algorithm to predict optimal values. The results showed that the R2 values for the Kriging models of ASR, AKW, and CAR were 0.9894, 0.9818, and 0.9822, respectively. Xu et al. [81] established a multiple linear regression (MLR) model to predict CS and KW under different process parameters, and they optimized the process-parameter combination using the Bat Algorithm (BA). The results indicated that the proposed method had prediction errors of 0.66% and 0.16% for CS and KW, respectively. Observation using confocal laser scanning microscopy showed that, under high IP, SR was 4.922 µm, which was reduced to 4.447 µm after the MLR optimization. AbouHawa and Eissa [82] determined the correlation between process parameters and angle error using a second-order polynomial regression model, and they conducted multi-response optimization using a composite expectation function and generalized simplified gradient method, finding the optimal process combination for carbon fiber-reinforced polymer (CFRP) composites of varying thicknesses. After optimization, the angular accuracy of CFRP at thicknesses of 0.5, 1.0, 1.5, and 2.0 mm improved by 24.66%, 1.24%, 30%, and 31.27%, respectively. Additionally, Singh et al. [83] stated that the selection of traditional process parameters was conducted through time-consuming trial-and-error experiments. The above studies further demonstrate the effectiveness of traditional methods in process optimization.

2.4. Summary

This section analyzes the citation trends of WEDM process-optimization research, with a focus on the methods used, as shown in Figure 4. As depicted, most scholars focus on statistical methods (such as Taguchi’s method and response surface methodology), as well as fuzzy control methods. This research preference primarily stems from the need to adapt to the complexity of WEDM processes: Taguchi’s method efficiently optimizes multivariable systems through orthogonal experimental design, response surface methodology constructs nonlinear models to balance multi-objective conflicts (such as machining efficiency and surface quality), and fuzzy control addresses uncertainty (such as discharge gap fluctuations) through fuzzy logic. These methods’ comprehensive advantages in nonlinear modeling, multi-objective optimization, and real-time adjustments make them the mainstream choices for process optimization. Moreover, in wire electrical-discharge machining, the dynamic coupling of process parameters (such as discharge energy and electrode wear) demands higher robustness for control models. Notably, compared to traditional methods such as linear regression, the citation frequency of the above-mentioned methods is significantly higher. This trend can be attributed to the fact that linear regression relies on strict linear assumptions and static analytical frameworks, thus making it difficult to capture the complex nonlinear relationships and time-varying characteristics between parameters in WEDM (such as accumulated thermal deformation during the machining process). In contrast, Taguchi’s method, response surface methodology, and fuzzy control, through experimental design, higher-order modeling, rule reasoning, and dynamic feedback mechanisms, are better suited to the strong nonlinearity and real-time requirements of the process, thus dominating both theory and practice.
According to the literature reviewed in this Section, Table 1 provides a detailed overview of various metallic materials used in the WEDM process and summarizes the research on process optimization for them. The materials involved include high-temperature alloys, tool steels, carbon steels, and alloy steels. Table 1 presents the key parameters of interest during the machining process, such as MRR and Ra. To optimize this machining performance, a wide range of optimization methods have been applied in research, such as Taguchi’s method combined with multi-response signal-to-noise ratio, RSM, and linear regression analysis combined with GA, and ANFIS, among others. In the study of Inconel 718, for example, Ramakrishnan and Karunamoorthy [54] applied Taguchi’s method with multi-response signal-to-noise ratio optimization, which significantly improved both MRR and Ra. The study reported a 1.11-fold increase in MRR and a reduction of Ra by 1.09 µm. In the research on AZ31 magnesium alloy, Kavimani et al. [58] adopted Taguchi’s method combined with GRA to optimize machining parameters, achieving ideal results under conditions such as P-ON = 40 μs. These methods assist researchers in making precise parameter adjustments based on the material characteristics and processing conditions, thus improving machining efficiency and surface quality. Each study meticulously documents the performance changes in different materials during the machining process, particularly the improvements in MRR and surface quality after optimization. These results show that the composition, hardness, conductivity, and other physical properties of the material significantly affect the machining process. Therefore, these factors must be considered when developing machining plans. It is especially important to note that there is often a trade-off between speed and accuracy, which requires the selection of appropriate machining parameters and optimization methods when processing different materials to achieve the best machining outcomes. Overall, these research findings provide valuable references for the application of WEDM technology in machining various metallic materials, emphasizing the importance of tailoring parameters to the specific material type and machining requirements. This not only helps improve production efficiency but also ensures high-quality machining results.
Table 1. The key literature on traditional methods in workpiece metal materials and optimization techniques.
Table 1. The key literature on traditional methods in workpiece metal materials and optimization techniques.
Metal (Material)Author(s)/YearOptimization MethodComments
High-temperature alloysInconel 706Sharma et al. (2018) [54]Taguchi method, GRA, and Principal Component AnalysisBy applying the Taguchi-GRA-PCA hybrid optimization method, simultaneous optimization of MRR and SR was achieved in wire-cutting processing of the novel aerospace high-temperature alloy Inconel 706, resulting in high productivity and excellent surface integrity.
γ-Titanium aluminideSarkar et al. (2008) [65]RSMThe method improves the machining efficiency of γ-TiAl alloys while maintaining the required surface finish and geometric accuracy.
Ti6Al4V alloysPasam et al. (2010) [78]Linear regression analysis combined with GA In the WEDM of Ti6Al4V alloy, a surface roughness of 1.85 µm can be achieved using optimized control parameters.
Tool steelsTool steel D2Ikram et al. (2013) [55]Taguchi method with variance analysis and signal-to-noise ratio The machining challenges of die materials are addressed in the WEDM of D2 tool steel, enabling effective processing of this high-hardness, high-performance material through parameter optimization.
SKD61Yan and Liao. (1998) [69]Fuzzy logic-based WEDM monitoring and ACO system Using SKD61 as a representative hard-to-machine material, the fuzzy control adaptive system is validated for its ability to enhance machining stability, reduce wire breakage, optimize material-removal rate, and improve surface finish in WEDM.
SKD-11Lin et al. (2001) [70]Fuzzy logic control strategy Fuzzy logic control is validated using SKD-11, effectively addressing the critical issue of corner accuracy in machining high-hardness materials via WEDM.
Chromium alloy (Cr12)Yuan et al. (2008) [77]GPR This material validated the effectiveness of the GPR model and optimization methods for hard-to-machine materials, solving core issues related to efficiency, surface quality, and machining stability.
Carbon steels and alloy steelsAISI 1045 medium carbon steelZaman et al. (2022) [60]Taguchi method L9OA The timing sprocket made of this material is selected to achieve maximum MRR through WEDM.
AISI 1050 carbon steel blockMaher et al. (2015) [72]ANFIS combined with Taguchi method Experiments on this material resolve the inherent trade-off between efficiency and surface quality in WEDM of high-hardness materials, while optimizing the heat-affected zone and energy consumption.
AISI 4140 steelTosun (2003) [76]Regression analysis This study not only demonstrates the applicability of WEDM to hard-to-machine materials, but also resolves the efficiency–surface quality conflict through systematic experiments and statistical analysis.
Non-ferrous metals and alloysHigh-purity copper (99.98%)Thankachan et al. (2018) [57]Taguchi method with GRA This material overcomes the decline in electrical conductivity of traditional composites caused by ceramic particle addition, while WEDM parameter optimization enables the development of a copper-based surface composite with high conductivity, wear resistance, and machinability.
AZ31 magnesium alloyKavimani et al. (2019) [58]Taguchi method combined with GRA AZ31 magnesium alloy is selected as the matrix, with graphene and SiC reinforcements introduced to address the limitations in mechanical properties and machining difficulties of magnesium alloys.
A413 aluminum alloySoundararajan et al. (2016) [67]RSM The study on extrusion casting and WEDM parameters of A413 alloy demonstrates its favorable casting performance and machining quality, suggesting its suitability for industrial applications, particularly in the automotive and aerospace sectors.
Aluminum matrix composite (A413-9% B4C)Sivaprakasam et al. (2013) [66]RSM Research on A413–9% B4C composites not only addresses the challenges of machining hard materials, but also achieves high-efficiency and high-precision machining using Micro-WEDM and parameter optimization.
75%Al 6061–25%SiCFard et al. (2013) [71]ANFIS combined with ABC Using 75% Al 6061–25% SiC as the research subject, the study resolves issues such as tool wear, high cost, and environmental pollution in traditional machining of hard composites, while enhancing machining efficiency and surface quality through intelligent optimization methods.
Special materialsα–β-Type titanium alloysGoyal et al. (2021) [73]ANFIS combined with NSGA-II The use of Ti–6Al–4V titanium alloy not only confirms the efficiency of WEDM for hard-to-machine materials but also addresses the complexity of parameter optimization through intelligent algorithms.
Tungsten steel YG15Ming et al. (2014) [79]GPR combined with NSGA-II Using YG15, the advantages of WEDM in machining ultra-hard materials are validated, and the contradiction between high MRR and superior surface quality is effectively resolved through multi-objective optimization.

3. Machine Learning

Machine-learning technologies have become a key driving force in the optimization of WEDM processes, particularly at the level of technological innovation. Their core value lies in enabling data-driven modeling and intelligent decision-making to address nonlinear, high-dimensional, and coupled process challenges that traditional methods often struggle to handle. WEDM performance is dynamically influenced by multiple variables, including discharge parameters, material properties, and environmental factors. Machine learning, with its powerful nonlinear fitting capabilities (e.g., neural networks), generalization from small datasets (e.g., SVM), and feature-importance interpretation (e.g., RF), can accurately model the relationship between process parameters and machining outcomes such as SR and MRR. In recent years, algorithms such as artificial neural networks (ANNs) and SVM, when combined with optimization strategies like GA and PSO, have demonstrated significant advantages in collaborative parameter optimization, wire-breakage prediction, and real-time control. These developments have accelerated the transition of WEDM toward intelligent and adaptive manufacturing. This section systematically elaborates on the application progress of machine-learning methods—such as neural networks and SVM—in WEDM, analyzing their technical principles and adaptability, and providing methodological support for achieving high-precision, high-efficiency intelligent machining.

3.1. Neural Network

In recent years, significant progress has been made in the optimization of WEDM processes. ANNs, various intelligent algorithms, and machine-learning technologies have been widely applied for parameter optimization and fault prediction. For instance, Ming et al. [84] proposed a novel method combining ANN with a leader wolf-pack algorithm (LWPA) to optimize WEDM cutting parameters. The single-objective optimization results showed that processing time, cost, and Ra were optimized to 164.1852 min, 239.5442 CNY, and 1.0223 μm, respectively. For SR, the ANN-LWPA model reduced the value by 0.1337 μm (11.6%), 0.3377 μm (24.8%), and 0.105 μm (10.3%) compared to experimental data, regression model, and ANN model, respectively. Abhilash and Chakradhar [85] adopted an offline classification model to predict machining faults and developed a multi-class classification model using ANN. The model demonstrated fast computation, ease of use, and an overall accuracy of 90%, reaching 95% prediction accuracy in confirmation tests. Shandilya et al. [86] developed an artificial neural network (ANN) model to predict the material-removal rate (MRR) during the wire electrical discharge machining (WEDM) of SiCp/6061 aluminum matrix composites. The model uses servo voltage, pulse-on time, pulse-off time, and wire feed rate as input parameters, and was trained using 29 sets of experimental data. The results showed that the model achieved a high prediction accuracy, with a correlation coefficient (R) of 0.9968, indicating a strong agreement between the predicted and experimental MRR values. Chou et al. [87] applied ML to analyze the relationship between machining parameters and wire breakage probability. Results showed that the model achieved over 85% accuracy in predicting wire breakage 10 s in advance. Paturi et al. [88] built an ANN model to estimate WEDM performance. The 8-5-5-4 ANN architecture exhibited the lowest MSE and absolute error on both training and test datasets. The R-value between ANN predictions and WEDM test values was 0.9995, indicating strong robustness in modeling the relationship between WEDM process factors and machinability parameters. Ishfaq et al. [26] applied three artificial intelligence (AI) modeling techniques—ANN, SVM, and Extreme Learning Machine (ELM)—and trained them using hyperparameter tuning. Results (as shown in Figure 5a) revealed that the ANN model achieved an R2 of 0.90, with RMSE and mean absolute error (MAE) both at 0.03 mm/s, showing excellent generalization capability. The R2 value of ANN was 59.9% and 2.19% higher than those of SVM and ELM, respectively. After parameter optimization, WEDM CS increased by 27.3%. These studies highlight the remarkable advantages of ANN and its improved algorithms in WEDM process parameter optimization and performance prediction, especially in enhancing prediction accuracy, improving machining efficiency, and reducing SR.
Meanwhile, with the continuous advancement of WEDM technology, researchers have widely adopted improved ANN models—such as Backpropagation Neural Networks (BPNNs)—along with intelligent optimization algorithms (e.g., GA and Simulated Annealing Algorithm (SAA)) to enhance the precision of process parameter optimization and prediction performance. Saha et al. [89] developed second-order multivariable regression and BPNN models to relate input process parameters (e.g., P-ON, P-OFF, IP, and capacitance) with CS and Ra. Results showed that the optimal neural network architecture was 4-11-2, yielding an average overall prediction error of 3.29% for CS and Ra. Chen et al. [90] proposed a hybrid approach combining BPNN with a SAA to optimize WEDM process parameters. The BPNN model achieved MSE of 1.68 × 10−2 for CS, 6.75 × 10−10 for Ra, and 9.30 × 10−8 for Rt, with prediction errors kept within 3%. Zhang et al. [91] conducted a central composite design experiment based on the RSM to optimize parameters, and modeled MRR and Ra using a BPNN-GA approach. Results indicated that the relative residuals of RSM for Ra and MRR were 33.3% and 53.28%, respectively, whereas the BPNN-GA method achieved much lower errors of 1.2% and 5.24%, demonstrating superior predictive accuracy. Zhang et al. [92] applied both BPNN-GA and NSGA-II optimization methods to obtain optimal solutions. Testing showed that the 4-6-5-1 structure of the BPNN provided the most accurate results. The BPNN-GA yielded relative errors of 18.6%, 12.5%, and 3.77% for Ra, white layer thickness, and surface crack density, respectively—all within 20%, confirming the effectiveness of the method. The NSGA-II Pareto-optimal solutions had relative errors below 30%, with both methods offering distinct advantages. Yusoff et al. [93] proposed an ANN-based prediction method for multi-performance evaluation in WEDM of Inconel 718. The study found that a cascade-forward backpropagation neural network was the optimal network type. The 5-14-4 cascade-forward backpropagation neural network with a single hidden layer demonstrated high accuracy and generalizability, achieving an average error of 5.16%, in strong agreement with experimental results. Singh and Misra [94] modeled and optimized the WEDM process for nickel-based alloy 263 using BBD and ANN, and analyzed the effects on Ra and Rz via Surfcom SR measurements. As shown in Figure 5b, BPNN was an effective tool for predicting responses, achieving a 99% match between predicted and experimental values. The BPNN model exhibited lower Root Mean Square Error (RMSE) and Prediction Error Percentage values, outperforming the RSM model in accuracy. In summary, these studies provide strong evidence for the significant advantages of BPNN and its improved algorithms in WEDM process optimization, especially in enhancing prediction accuracy and controlling multiple performance parameters.
Figure 5. Neural networks: (a) Methodology applied in the WEDM case study, including the WEDM setup and stainless-clad steel material (Reprinted from Ref. [26]). (b) Tasks performed in the experimental study and comparison between experimental values and predicted values from RSM and ANN (Reprinted with permission from Ref. [94]. Copyright 2019, Elsevier).
Figure 5. Neural networks: (a) Methodology applied in the WEDM case study, including the WEDM setup and stainless-clad steel material (Reprinted from Ref. [26]). (b) Tasks performed in the experimental study and comparison between experimental values and predicted values from RSM and ANN (Reprinted with permission from Ref. [94]. Copyright 2019, Elsevier).
Metals 15 00706 g005

3.2. Support Vector Machine and Random Forest

SVM, as an efficient machine-learning algorithm, has been widely applied in recent years for the optimization of WEDM processes. When combined with adaptive control and multi-objective optimization techniques, studies have achieved significant results in enhancing machining accuracy, reducing errors, and optimizing parameters. For example, Huang et al. [95] proposed an SVM-based online workpiece height-estimation method integrated into a novel reciprocating WEDM system. The system collected current and voltage signals from the discharge gap through a sampling circuit and used an adaptive control unit to adjust machining parameters. The results revealed that the method was effective, with an estimation error of less than 2 mm and a machining time reduction of over 30%. Nain et al. [96] applied the SVM algorithm (using normalized polynomial and radial basis kernels) to model the WEDM process and employed the GRA method to optimize process variable combinations for achieving ideal outcomes in CS, wire-wear ratio, and DD. The optimal parameter combination was P-ON—0.4; P-OFF—13; IP—130; WT—1500, servo voltage (SV)—36; and WF—10. The copper content was 8.66% at the highest CS and 7.05% at the lowest. Bijeta Nayak and Sankar Mahapatra [97] developed a Support Vector Regression (SVR) model to predict the angular error in WEDM taper cutting. The results confirmed that, for training data, the model achieved a RMSE of 0.0037, a Nash–Sutcliffe efficiency coefficient (E) of 0.9998, and a R2 of 0.9999; and for testing data, RMSE was 0.2721, E was 0.8644, and R2 was 0.96037. Paturi et al. [98] modeled the WEDM process and predicted SR using SVM, ANN, and RSM models, with GA employed to optimize the process parameters. As shown in Figure 6a, the SVM model yielded the most accurate predictions, with a Mean Absolute Percentage Error (MAPE) of 0.0347% and an R-value of 99.9985%, indicating high consistency between predicted and experimental results. GA optimization improved SR by 61.31%. Sharma et al. [99] developed GPR and SVM regression models using the WEKA platform to predict the SR of Ti-6Al-7Nb titanium alloy. The GPR model achieved a maximum RMSE of 0.0844 mm and R2 of 0.985 for training data, and RMSE of 0.0445 and R2 of 0.991 for testing data. The SVM model achieved an RMSE of 0.001 mm and R2 of 0.995 for training data, and an RMSE of 0.0176 and R2 of 0.996 for testing data. SVM outperformed GPR in both R and RMSE. These studies collectively demonstrate that SVM and its improved models exhibit high accuracy and stability in WEDM process optimization, performance prediction, and error control, significantly improving machining precision and reducing processing time.
Meanwhile, RF, with its outstanding predictive performance and parameter optimization capabilities, offers an effective approach to enhancing machining accuracy, stability, and efficiency. For instance, Zhang et al. [100] proposed a novel online monitoring method for WEDM-MS, integrating digital image processing and machine learning to monitor discharge pulses. Experimental results demonstrated that the system achieved an accuracy of 93.13% across various electrical parameters and materials, with the SVM-RF approach outperforming other recognition methods, such as LVQ neural networks, standalone SVM, and RF. Nain et al. [101] designed experiments using the Taguchi L27 orthogonal array and modeled the WEDM process using RF and M5P tree methods. Sensitivity analysis and PSO identified the optimal parameter combination as P-ON—0.7460 (μs); P-OFF—9 (μs); IP—130 (A); WT—1500 (Gm); SV—36 (V); WF—6 (m/min); MRR—26.2193 (mm2/min); Ra—2.1499 (μm); and DD—148.541 (μm). Upadhyay et al. [102] employed logistic regression, RF, decision trees, and Gaussian Naive Bayes to predict wire breakage in the WEDM of nickel-based alloy Nimonic 263. The results revealed that the RF model achieved the highest prediction accuracy on the test data, reaching 100%. Jai Rajesh et al. [103] combined regression analysis with metaheuristic algorithms to determine the optimal WEDM parameters for aerospace-grade Ti6Al4V alloy. As shown in Figure 6b, the RF regression model performed best, achieving R2 values of 0.8555 for MRR and 0.8887 for Ra, highlighting its ability to capture complex causal relationships. Metaheuristic algorithms such as PSO and the Firefly Algorithm also provided optimized parameter combinations for MRR and Ra. Saha et al. [104] applied machine learning-based metaheuristic algorithms to improve kerf quality during the WEDM of woven CFRP. RF and Adaptive Boosting were used to model KW, and optimization was carried out using techniques like the moth–flame optimizer. The RF model showed strong predictive performance with an R2 of 0.8272, RMSE of 3.7129, and MAE of 2.6498. The optimal parameters determined by the moth–flame optimizer were P-ON = 26 μs, P-OFF = 50 μs, I = 7 A, and V = 70 V, resulting in a minimum KW of 250 μm and an improvement of approximately 5.6% after optimization. These studies collectively demonstrate the strong potential of RF in improving prediction accuracy, optimizing parameter combinations, and enhancing process stability, offering valuable insights for the intelligent control of WEDM processes.
Figure 6. SVM and RF: (a) Overview of the SVM regression nonlinear model and the correlation between experimental results and SVM model predictions (Reprinted with permission from Ref. [98]. Copyright 2021, Elsevier). (b) RF structural diagram and ranking analysis of the regression model (Reprinted from Ref. [103]).
Figure 6. SVM and RF: (a) Overview of the SVM regression nonlinear model and the correlation between experimental results and SVM model predictions (Reprinted with permission from Ref. [98]. Copyright 2021, Elsevier). (b) RF structural diagram and ranking analysis of the regression model (Reprinted from Ref. [103]).
Metals 15 00706 g006

3.3. Other

In addition to the ANN, SVM, and RF methods, many other advanced machine-learning and optimization techniques have been applied to optimize the WEDM process. For example, Konda [105] applied three decision tree algorithms (C5.0, CART, and CHAID) to predict MRR by classifying the target into 2 to 5 categories, using three discretization methods, generating 36 combination models and evaluating their prediction accuracy. The results confirmed that 21 models were suitable for WEDM, with prediction accuracy ranging from 71.43% to 100%. Wang et al. [106] proposed a new tolerance-monitoring method based on unsupervised machine learning, using ionization time distribution as the monitoring variable and employing K-means clustering and hierarchical clustering to avoid time-consuming experiments. The results indicated that regions of clusters 1 and 2 were 100% related to short circuits, while regions of clusters 3 and 5 were 100% within the ±15 μm tolerance band. Ulas et al. [107] successfully predicted the SR of WEDM-processed aluminum alloys using ELM, Weighted Extreme Learning Machine (W-ELM), SVR, and Q-SVR machine-learning models. The R2 values of ELM, W-ELM, SVR, and Q-SVR were 0.9411, 0.9720, 0.8824, and 0.9613, respectively, with W-ELM performing the best, achieving an R2 of 0.9720 and an RMSE of 0.0364. The machine-learning methods predicted the SR with an error rate as low as 2.8%, demonstrating high feasibility. Trung-Thanh and Khoa [108] used a radial basis function to construct a prediction model, combined with Nondominated Sorting Particle Swarm Optimization (NSPSO) to obtain the optimal WEDM processing parameters. The results indicated that the radial basis function model significantly improved the prediction accuracy of WEDM outputs. The optimal parameters were a current of 5.0 A, pulse duration of 6.0 microseconds, voltage of 33.0 V, and WS of 4 m per minute. After optimization, the root mean square roughness (RMSR) and recast layer depth were reduced by 60.98% and 15.55%, respectively, while the CS increased by 8.90%. Abhilash and Chakradhar [109] utilized a kernel function-based naive bayes algorithm to perform binomial classification of processing results and predict process continuity, with labels for wire breakage and continuous processing. The results demonstrated that the naive bayes model had the highest accuracy for wire breakage classification at 96.7%, outperforming five alternative classifiers. Vijayakumar and Chandradass [110] combined the Taguchi method–MCDM–supervised machine-learning technique to optimize WEDM process parameters to enhance MRR, profile error, and SR of 20MnCr5 steel. The results showed that the decision tree had an accuracy rate of 96%, with an ACU of 0.99, demonstrating excellent performance. Overall, these studies demonstrate the broad application of machine-learning technologies in WEDM processing, particularly in improving processing accuracy and optimizing processing parameters, which hold significant academic and practical value.

3.4. Summary

In the optimization process of the WEDM process, research primarily focuses on processing parameter optimization, surface quality prediction and optimization, MRR and energy efficiency optimization, and multi-objective optimization. This section compiles the citation trends of the key literature (2008–2024), categorized by research area and the machine-learning algorithms employed. As shown in Figure 7, neural networks have become the most widely used method due to their advantages in nonlinear system modeling, with improved models playing an important role in optimization. Additionally, methods such as SVM, RF, and W-ELM have also been applied, but they are cited less frequently due to limitations in applicability and ease of use. WEDM, as a high-precision machining method, has its efficiency and quality directly influenced by parameter optimization. Surface-quality optimization can reduce defects, MRR and energy-efficiency optimization balance quality and energy consumption, and multi-objective optimization can comprehensively enhance process performance. Neural networks dominate in modeling complex nonlinear issues in WEDM, such as the effects of discharge energy and pulse intervals, due to their adaptive learning ability and mature tool support (such as MATLAB (R2024b) and TensorFlow (2.19.0)). In contrast, SVM performs well with high-dimensional small-sample data, but WEDM datasets are usually large, and hyperparameter selection is more complex, limiting its application. RF is robust in feature selection and noise handling but is weaker in handling continuous variables. W-ELM has fast computation speed but limited generalization ability, leading to its less frequent application. Overall, ANN, due to its powerful nonlinear modeling ability and well-established software support, dominates in WEDM research, while other machine-learning methods, due to their respective limitations and applicability, are cited less frequently.
Research on the optimization of the WEDM process has received extensive attention in recent years, particularly with the application of machine-learning techniques to various metallic materials, significantly enhancing machining efficiency and precision. Table 2 summarizes the research achievements involving diverse metal materials and optimization techniques, incorporating a range of machine-learning methods, such as ANN, SVM combined with GA, RF, M5P model trees integrated with sensitivity analysis, and PSO. For instance, Inconel 718, a high-temperature alloy known for its exceptional thermal resistance and corrosion resistance, is widely used in the aerospace and energy sectors. Paturi et al. [98] reported that the combination of SVM and GA significantly improved both the surface roughness and machining accuracy of Inconel 718, achieving a prediction accuracy of 99.9985%. Similarly, Udimet L605, another high-performance alloy extensively applied in manufacturing high-temperature gas turbine blades due to its excellent oxidation resistance and thermal stability, demonstrated enhanced surface quality and a significantly increased MRR after optimization using RF and PSO. These cases illustrate the substantial potential of machine-learning techniques in optimizing the machining of complex, high-performance materials via WEDM. Likewise, SKD11 tool steel, known for its outstanding hardness and wear resistance and commonly adopted in die and cutting-tool manufacturing, has shown promising optimization results. Zhang et al. [91] employed a BPNN-GA to optimize its WEDM process, resulting in relative residual errors of only 1.2% for Ra and 5.24% for MRR, indicating the high efficiency of machine learning in processing hard materials. Moreover, AI7075-T6 aluminum alloy, favored in aerospace and automotive industries for its lightweight and high-strength properties, has also benefited from machine learning-based optimization in WEDM. The optimized process yielded significant improvements in both surface quality and material-removal efficiency. These examples collectively demonstrate that machine learning-based optimization methods can markedly enhance the performance of WEDM across a wide range of metallic materials. Not only do they improve machining efficiency and precision, but they also offer novel strategies and solutions for processing challenging materials. Therefore, the continued development of machine-learning techniques in metal machining is expected to provide strong technical support for the efficient and precise manufacturing of high-performance metals.
Table 2. The key literature on the application of machine-learning methods in metallic materials and their optimization techniques during the process of WEDM.
Table 2. The key literature on the application of machine-learning methods in metallic materials and their optimization techniques during the process of WEDM.
Metal (Material)Author(s)/YearOptimization MethodComments
High-temperature alloysInconel 718Abhilash and Chakradhar. (2020) [85]ANN Using Inconel 718 as the research subject not only highlights the unique advantages of WEDM in machining difficult-to-cut materials, but also addresses critical issues such as wire breakage and lack of spark during the process.
Yusoff et al. (2018) [93]CFNN By employing a CFNN-based approach, this study effectively tackles the multi-objective performance prediction challenges in WEDM of Inconel 718, significantly reducing experimental cost and time while enhancing prediction accuracy.
Paturi et al. (2021) [98]SVM combined with GA The selection of Inconel 718 as the workpiece material not only verifies the advantages of WEDM in processing difficult-to-machine alloys, but also achieves precise optimization of process parameters through machine learning, thereby resolving the core issue of surface-roughness control.
Wang et al. (2018) [106]Unsupervised machine learning This research overcomes the limitations of traditional broaching techniques for Inconel 718 by integrating intelligent monitoring, which substantially improves machining efficiency and surface quality, providing an innovative solution for the manufacturing of critical aerospace components.
Abhilash and Chakradhar. (2021) [109]NB Inconel 718 is chosen due to its high representativeness as a typical difficult-to-machine material. The study resolves the wire breakage issue during its WEDM processing, offering technical assurance for precision manufacturing of high-value components.
Nimonic 263Singh and Misra. (2019) [94]Box–Behnken design combined with ANN Through a systematic investigation into the WEDM characteristics of Nimonic 263, the study addresses the challenges of achieving high-efficiency and high-quality machining of difficult-to-cut materials, offering valuable process insights and technical support to relevant industrial sectors.
Rectangular plate of Udimet-L605Nain et al. (2018) [96]SVM combined with GRA In the WEDM machining of Udimet-L605, this study solves the problems of low efficiency and severe tool wear encountered in conventional machining of high-temperature alloys. Machine learning-based parameter optimization is introduced to provide both theoretical and practical guidance for high-performance material processing in the aerospace industry.
Udimet-L605Nain et al. (2018) [101]RF and M5PTree combined with sensitivity analysis and PSO This work aims to address the issues of efficiency, quality, and precision in the machining of Udimet-L605, thereby enhancing its application potential in the aviation industry.
Tool steelsTungsten carbide tool (YG15 grade)Ming et al. (2016) [84]ANN combined with LWPA method The proposed method resolves challenges in traditional WEDM of YG15, including unstable surface quality, high machining cost, and low efficiency in parameter tuning. It validates the effectiveness of intelligent optimization algorithms in complex process-parameter matching.
Zhang et al. (2015) [92]BPNN-GAand NSGA-II YG15 tungsten carbide is selected as the workpiece to address its machining difficulties in the mold industry due to its high hardness and wear resistance. Parameter optimization enhances improvements in both surface quality and mechanical properties, and a reduction in rejection rates in industrial applications.
SKD11Zhang et al. (2013) [91]BPNN-GA Focusing on SKD11, the study aims to solve the process parameter-optimization problem in MS-WEDM. After optimization using BPNN-GA, the relative residuals of Ra and MRR were reduced to 1.2% and 5.24%, respectively.
Zhang et al. (2015) [100]SVM-RF As a representative die steel, SKD11 demonstrates that pulse-off time plays a critical role in reducing harmful discharges, such as arcing and short circuits, achieving a balance between high material-removal rate and stable machining.
Tungsten carbideZhang et al. (2015) [100]SVM-RF Tungsten carbide, as a difficult-to-machine material, is adopted to reveal the discharge behavior under high energy and long pulse duration, offering guidance for selecting parameters in hard material machining.
Tool steelHuang et al. (2018) [95]SVM Using tool steel as the workpiece, this study addresses the key challenge of machining variable-height components in RT-WEDM. The proposed method achieved an estimation error of less than 2 mm and reduced machining time by more than 30%.
Non-ferrous metals and alloysTitanium alloyPaturi et al. (2022) [88]ANNWEDM experiments on titanium alloy, combined with ANN for performance prediction, yields an R-value of 0.9995 between predicted and measured values, indicating high predictive accuracy.
Aluminum alloy 7075-T6Ulas et al. (2020) [107]W-ELM Using Al7075 T6 aluminum alloy as the workpiece, WEDM coupled with W-ELM prediction of Ra significantly reduced experimental cost and time, overcoming the limitations of conventional machining in achieving high-precision processing.
Copper blockZhang et al. (2015) [100]SVM-RF As a highly conductive material, copper is used to validate the influence of pulse-on time on spark discharge ratio, thereby optimizing machining stability.
Special materialsTungsten and tungsten alloysChen et al. (2010) [90]BPNN and SAA Pure tungsten and tungsten alloys were employed to solve the processing difficulties caused by their high hardness, melting point, and brittleness. High-efficiency machining of high-purity tungsten is successfully achieved using BPNN-SAA.
Tungsten carbide–cobalt (WC-Co) compositeSaha et al. (2008) [89]Second-order multivariable regression model and BPNN WC-Co composites are machined using WEDM, and the Cs and Ra values are predicted using BPNN and multiple regression models. The overall mean prediction error for Cs and Ra is 3.29%. This approach addresses the instability, poor surface quality, and parameter optimization difficulties arising from the hardness and compositional variability of such materials.

4. Deep Learning

Deep learning technology has brought a revolutionary breakthrough to the optimization of the WEDM process (in the context of intelligent transformation). Its core advantage lies in solving the prediction bottleneck of traditional methods under complex conditions through multi-level feature extraction and nonlinear modeling. The processing quality of WEDM is influenced by the coupling of multiple factors, such as discharge state, material properties, and dynamic parameters. Deep learning models (e.g., CNN and Deep Neural Networks (DNNs)), with their powerful ability to capture spatiotemporal features (e.g., local discharge image analysis) and end-to-end learning mechanisms (e.g., multimodal signal fusion), can accurately predict processing performance indicators (e.g., SR and MRR) and enable real-time process control. In recent years, methods such as CNN combined with time-series modeling (e.g., Gated Recurrent Units (GRUs)) and DNN integrated with optimization algorithms (e.g., manta ray foraging optimization (MRFO)) have shown significant advantages in discharge state recognition, defect detection, and multi-objective optimization, driving WEDM toward high-precision, adaptive processing. This section will systematically elaborate on the application progress of deep learning models, such as CNN and DNN, in WEDM; analyze their technical principles and engineering value; and provide methodological support for process optimization in intelligent manufacturing scenarios.

4.1. Convolutional Neural Network

In recent years, deep learning technology has made significant progress in optimizing discharge-state prediction for WEDM. Researchers have employed methods such as CNN combined with time-series modeling, data augmentation, and multimodal fusion to enhance prediction accuracy and model generalization, providing new approaches for process optimization. For example, Liu et al. [111] proposed a deep learning model based on CNN and GRU for predicting discharge states. They also innovatively introduced a quantitative labeling method for the processing state, significantly improving the stability of the model. Experimental results demonstrated that the prediction method based on spark images could accurately reflect and track the processing state. The “image-to-sequence” model achieved a prediction accuracy of 95% across the entire dataset, while the “sequence-to-sequence” model had an accuracy of 90%, fully validating the effectiveness of the proposed method. Yang et al. [112] developed an efficient dual-input deep learning model—batch-related time-series convolutional neural network (BRTCN)—and applied it to construct an acoustic emission (AE) model to precisely predict discharge states. Additionally, they proposed a novel labeling method. Experimental results indicated (as shown in Figure 8a) that the dual-input model performed excellently in training with acoustic emission signals. The best model (256-BR(gru)TCN-NE) had a test loss that was only 58.08% of the single-input model’s test loss, and the MSE on the test set was as low as 0.009, significantly improving prediction accuracy and model performance. Ye et al. [113] introduced the mix-up data augmentation method into the electric discharge-state dataset and combined it with the CNN-GRU network, proposing a mix-up data augmentation-based CNN-GRU hybrid network algorithm. Experimental results depicted that this model could effectively capture the “temporal features” and “local features” in the data, improving the recognition accuracy to 96%, a 3.9% increase compared to the original CNN-GRU model. This result indicates that the new data generated by mix-up data augmentation significantly enhanced the generalization ability of the CNN-GRU network on small sample datasets, effectively strengthening the model’s robustness and predictive performance. Tang et al. [114] designed and optimized a Field-Programmable Gate Array -based CNN-GRU hybrid neural network model for real-time identification of WEDM discharge gaps. Experimental results revealed that this model achieved a recognition accuracy of 97.35%, outperforming both CNN (90.26%) and GRU (96.86%), meeting the high-precision requirements of electrical discharge machining.
At the same time, CNNs have demonstrated great potential in optimizing the surface quality of WEDM processes. Researchers have developed various CNN models by combining scanning electron microscope (SEM) image analysis, defect detection, and surface morphology prediction techniques to improve the accuracy and efficiency of surface quality assessment. These methods not only optimized the prediction accuracy of MRR and Ra, but also enhanced defect recognition capabilities, providing intelligent solutions for WEDM process optimization. Rahul et al. [115] analyzed the surface morphology of Ni-Ti-Hf-based alloys using SEM and, based on pixel intensity histograms, applied a CNN model for MRR classification of SEM images. To optimize MRR and Ra, the PSO algorithm was introduced. The results confirmed (as shown in Figure 8b) that the one-dimensional CNN model had low computational costs and high processing efficiency when analyzing the Ni50.3-Ti34.7-Hf15 high-temperature shape-memory alloy (SMA), achieving a classification accuracy of 92.67%. Experimental verification results indicated that, under two sets of input parameters, the MRR error was 2.01% and 2.046%, and the Ra error was 3.86% and 1.611%, values that are highly consistent with the model’s predictions, proving the reliability of the proposed method. Li et al. [116] proposed a CNN-based defect detection method based on ResNet-50 that can identify, classify, and locate defects such as cracks, notches, burrs, and deformation in the WEDM process. Experimental results depicted that, after 40 rounds of training, the model achieved 95.3% accuracy, 94.2% recall, 95.8% precision, and 94.7% F1 score on the test set. Among these, crack detection had the highest accuracy (96.5%), while deformation detection was lower (92.2%), indicating room for further optimization. Compared to traditional methods, this deep learning model outperformed in all metrics. Furthermore, in WEDM process optimization, Gonzalez-Sanchez et al. [117] also applied the theoretical knowledge of convolutional neural networks to study the surface morphology of WEDM, achieving significant results.

4.2. Deep Neural Network

In recent years, AI technologies have been increasingly applied in advanced manufacturing, and DNN methods have been adopted to optimize the WEDM process to improve machining performance and surface quality. Kumar and Das [118] experimentally analyzed the effects of P-OFF, P-ON, SV, and IP on MRR, recast layer thickness (RLT), and Ra, comparing the results with the RSM model. A hybrid DNN model based on ANOVA and MRFO was applied to improve prediction accuracy. The results demonstrated that the DNN+MRFO method achieved an average prediction accuracy of 90%. The maximum MRR obtained experimentally was 2.657 mm3/min, with the minimum RLT and Ra being 7.9 µm and 0.351 µm, respectively. Regression analysis indicated that the DNN+MRFO model outperformed the RSM, with expected values of 0.844 (RSM) and 0.921 (DNN+MRFO) in the validation analysis. Meanwhile, Kumar and Das [119] proposed a hybrid model combining DNN with the COOT optimization algorithm (DNN+COOT) to predict the MRR, microhardness (MH), Ra, and RLT performance indicators for UV-WEDM machining of AISI P20+Ni. The process parameters were optimized using the RSM, and the prediction performance was compared with traditional DNN and RSM models. The results showed that the DNN+COOT method achieved a prediction accuracy of 98.77%, significantly improving upon traditional DNN. The optimized UV-WEDM process could maximize MRR and MH (46% and 5–6% improvements, respectively) and minimize Ra and RLT (Ra improved by up to 69%) at a desirability of 0.65. The experimental error did not exceed 2%. The prediction errors of DNN+COOT were significantly reduced (RMSE: MRR, 0.0293; MH, 0.0292; RLT, 0.0276; and Ra, 0.0268), whereas the RMSE values of traditional DNN and RSM methods were higher (0.326 and 0.479, respectively). Therefore, DNN+COOT demonstrated outstanding performance in predicting UV-WEDM processing parameters.
At the same time, accurate prediction of discharge conditions in the WEDM process is crucial for improving machining stability and quality. With the development of deep learning technologies, researchers have attempted to use DNN methods to extract key features from machining signals to optimize the WEDM process and enhance anomaly detection capabilities. Sanchez et al. [120] performed DNN to identify hidden patterns from raw voltage signals during machining and assessed the feasibility of predicting sudden events in industrial WEDM processes. Their study compared the accuracy, recall, and F1 scores of different DNN models, including CNN, RNN, bidirectional RNNs, and combined CNN-RNN models, across multiple datasets. The results confirmed that the model with one convolutional layer and two GRUs performed optimally, achieving nearly 100% accuracy on multiple datasets. From the machining-process perspective, confusion matrix analysis indicated that the model could accurately predict changes in workpiece thickness of at least 2mm, with a prediction accuracy of up to 97.4%. This capability ensures rapid early warning responses, effectively preventing process degradation and providing ample time to adjust machining parameters.

4.3. Other

In addition to CNN and DNN, researchers have proposed various optimization methods, including intelligent prediction models, optimization algorithms, and deep learning techniques. These methods aim to improve machining accuracy, optimize machining parameters, and enhance MRR. The following studies demonstrate recent advances in WEDM process optimization: Conde et al. [121] proposed a prediction method based on an Elman-structured layer recurrent neural network (LRNN) to evaluate WEDM machining accuracy. Combined with a variable-radius wire path design algorithm, they utilized software compensation to correct machining deviations. By integrating LRNN prediction with simulated annealing (SA) optimization, the optimal wire path design was achieved, minimizing radial deviations caused by electrode-wire deformation. The results revealed (as shown in Figure 9a) that the optimal network structure consisted of seven hidden layer neurons and two feedback delays, resulting in the lowest MAPE. When compared to measurements from a coordinate-measuring machine (CMM), the maximum average deviation was <5.6 μm (CMM extended uncertainty ±5 μm), validating the model’s high prediction accuracy. Further experiments demonstrated that the solution was particularly effective for machining small-radius and large-height parts, with an 80% reduction in average deviation and a 43% reduction in the coefficient of variation. Kumar and Das [122] adjusted input parameters such as P-ON/OFF, IP, and SV to measure MRR, Ra, electrode wire wear, and KW, and optimized machining performance using the RSM and a deep belief network (DBN)-based search-and-rescue (SAR) optimization algorithm. The results showed that the RSM-BBD model successfully optimized WEDM process parameters, achieving a desirability of 0.79. The optimized input parameters were P-OFF of 60 µs, P-ON of 110 µs, SV of 5.1869 V, and IP of 11.00889 A. Additionally, the SAR-optimized DBN prediction model had an average accuracy of approximately 90%, confirming its efficiency and accuracy. Kumar and Jayswal [123] used deep learning to optimize MRR in WEDM. By generating datasets through experimental design, they analyzed the effects of pulse parameters, current, voltage, and WS on MRR. The results indicated that, among the evaluated activation functions (Sigmoid, Tanh, and ReLU), the Sigmoid activation function performed the best, with a MSE of 0.0004, a MAE of 0.0166, and a R2 of 0.9999. These results demonstrated a high correlation between predicted and actual values, highlighting the effectiveness of the Sigmoid activation function in this application.
Secondly, in the optimization of surface quality in WEDM (wire electrical discharge machining) processes, recent research has demonstrated the significant application of deep learning technologies in accurately detecting and predicting surface features. The following studies further advance this field and showcase the potential of combining deep learning with WEDM technology. Wang et al. [124] proposed a new method for detecting tolerance defects in turbine disc keyway machining, marking the first application of deep learning technology in the WEDM field. Their study involved machining industrial-grade turbine disc keyways and comparing the defect-prediction results from deep neural networks with traditional measurements from a CMM. The results depicted that, within the strict ±5 µm tolerance range, the network’s predictions were highly consistent with the traditional CMM results in 80% of the regions. Jiang and Yen [125] proposed a method called MTF-CLSTM, which combines Markov Transition Field (MTF), CNN, and Long Short-Term Memory Networks (LSTMs) to achieve accurate prediction of product quality after WEDM machining. Experimental results revealed that, under three-, four-, and five-state MTF, the MAPE of the method was 3.11%, 2.94%, and 3.24%, respectively, outperforming DNN and MC-DNN methods under the same experimental conditions and significantly improving prediction accuracy. Vakharia et al. [126] employed WEDM to machine nickel–titanium alloy bars and analyzed surface features using Field-Emission Scanning Electron Microscopy. They also applied Single-Image Generative Adversarial Networks (SinGANs) and Dense Net deep learning models to predict surface morphology and its correlation with machining parameters. The results indicated (as shown in Figure 9b) that the Dense Net model achieved a 100% average accuracy during both the training and testing phases, accurately predicting surface images. The model also achieved a 100% recognition rate in the other three metrics, performing best in surface morphology prediction. The next best-performing models were Alex Net (99.8%), KNN (99.4%), and MNB (98.9%).
Finally, the optimization of the WEDM process relies heavily on the accurate identification of discharge conditions. In recent years, deep learning technologies have gradually been applied to discharge state monitoring, offering more precise analysis of discharge characteristics compared to traditional methods, thus improving machining stability and efficiency. For example, Shen et al. [127] proposed a machine vision method based on deep network learning, utilizing the TensorFlow parallel computing framework to identify discharge conditions. The study compared three methods: K-nearest neighbors, logistic regression, and deep network learning. The results demonstrated that deep network learning achieved the best performance, with an identification accuracy of 95.63%. Considering both accuracy and training time, the best configuration for discharge state identification on wire EDM machines was found to be convolution layer feature parameters (C1 = 64; C2 = 128) and a training step size of 200. Wang et al. [128] proposed a time-domain–frequency-domain dual-memory network model based on deep learning, combining LSTM networks with wavelet transform to precisely identify spark discharge states during the wire electrical discharge machining process. The experimental results revealed that, after ten-fold cross-validation, the classification accuracy of the dual-output fusion model reached 99.69%, significantly outperforming traditional methods and effectively meeting the practical requirements for discharge state identification in WEDM processes.
Figure 9. (a) Algorithm framework combining LRNN prediction and SA optimization, the variation in the circular radius predicted by LRNN and experimental measurements with respect to the angle (radius, 4 mm; height, 50 mm; measurements at plane j = 5 (Z = −20 mm)) and the optimized average deviation (radius, 3 mm; height, 100 mm) (Reprinted with permission from Ref. [121]. Copyright 2018, Elsevier). (b) Flowchart of the proposed method and the confusion matrix of Dense Net training and testing prediction results (Reprinted from Ref. [126]).
Figure 9. (a) Algorithm framework combining LRNN prediction and SA optimization, the variation in the circular radius predicted by LRNN and experimental measurements with respect to the angle (radius, 4 mm; height, 50 mm; measurements at plane j = 5 (Z = −20 mm)) and the optimized average deviation (radius, 3 mm; height, 100 mm) (Reprinted with permission from Ref. [121]. Copyright 2018, Elsevier). (b) Flowchart of the proposed method and the confusion matrix of Dense Net training and testing prediction results (Reprinted from Ref. [126]).
Metals 15 00706 g009

4.4. Summary

In the optimization of WEDM using deep learning technologies, research mainly focuses on aspects such as surface quality, discharge conditions, and machining performance. Through a citation analysis of the important literature in the field (2018 to 2024), relevant studies were systematically categorized according to research topic and technical approach (see Figure 10). The research found that the majority of studies concentrate on optimizing discharge conditions and evaluating surface quality and mechanical properties. This is because WEDM technology is increasingly applied in precision manufacturing, and process parameters—especially discharge condition control—play a decisive role in machining accuracy and workpiece performance. The analysis revealed that compared to traditional deep learning methods (such as CNN and DNN) and emerging algorithms (like MTF-CLSTM and SinGAN–Dense Net), classical statistical models (such as RSM-DBN-SAR) dominate in citation count. This is due to their well-established theoretical foundation, clear interpretability, lower computational demands, and long-term validation in industrial scenarios. For example, the method proposed by Sanchez et al. [120] has become a representative high-impact study with 84 citations. In contrast, despite the stronger predictive capabilities of deep learning technologies, their application is limited by high data-quality requirements and the complexity of model training, which has resulted in relatively fewer citations for studies like those by Shen et al. [127]. With the optimization of algorithms and the upgrading of computational hardware, it is expected that deep learning and hybrid models (such as LRNN-SA) will gradually become mainstream in research on complex working conditions.
The application of deep learning methods in the optimization of metallic workpiece materials is increasingly becoming a research hotspot in the field of WEDM. In recent years, studies on WEDM process optimization have progressively focused on deep learning techniques, which have demonstrated significant advantages in enhancing machining accuracy, efficiency, and surface quality. Table 3 summarizes several key studies that illustrate the effectiveness of integrating deep neural network models—such as CNN, LSTM, and DNN—with heuristic algorithms, like the COOT algorithm, PSO, and GRU, in optimizing various metallic materials, including tool steels and shape memory alloys. For instance, Kumar and Das [119] successfully combined DNN with the COOT algorithm to optimize WEDM parameters for AISI P20+Ni tool steel, achieving a 45.88% increase in MRR and a 69.81% reduction in Ra, marking a significant breakthrough in the precision machining of complex alloys. Another representative study by Rahul et al. [115] utilized a hybrid CNN-PSO approach for the WEDM of Ni-Ti-Hf shape-memory alloys, attaining simultaneous improvements in both MRR and surface quality, while preserving the material’s phase-transformation characteristics. Moreover, investigations on steels such as SKD11 and AISI D2 have also shown that deep learning models like Dense Net and LRNN exhibit excellent adaptability and generalization capabilities in WEDM parameter prediction and control. These results collectively underscore that deep learning-based optimization methods not only outperform traditional algorithms but also provide robust support for intelligent decision-making in complex manufacturing environments. As the integration of deep learning technologies with WEDM processes continues to evolve, they are expected to play an increasingly pivotal role in the high-efficiency and precision manufacturing of high-performance metallic materials.
Table 3. Summary of key publications on deep learning methods for workpiece metal materials and optimization approaches during the process of WEDM.
Table 3. Summary of key publications on deep learning methods for workpiece metal materials and optimization approaches during the process of WEDM.
Metal (Material)Author(s)/YearOptimization MethodComments
High-temperature alloysInconel 718Wang et al. (2019)
[124]
Deep learning techniques The integration of WEDM and deep learning techniques has successfully addressed the challenge of high-precision geometric defect detection in the fabrication of turbine disk fir-tree slots using Inconel 718 superalloy.
Tool steelsAISI P20+NiKumar and Das (2023)
[118]
DNN integrated with ANOVA and MRFO AISI P20+Ni tool steel, known as one of the hardest steels for plastic-mold manufacturing, presents significant machining challenges. By employing zinc-coated brass wire electrodes and optimizing WEDM parameters, this study effectively achieves the synergistic optimization of MRR, recast layer thickness, and surface roughness.
Kumar and Das (2024)
[119]
DNN combined with COOT Using AISI P20+Ni tool steel as the workpiece material, UV-assisted WEDM can enhance a 45.88% improvement in MRR and a 69.81% reduction in SR, overcoming the issues of low efficiency and poor surface quality associated with conventional machining.
Kumar and Das (2023)
[122]
RSM and DBN integrated with SAR The machining behavior of AISI P20+Ni tool steel in WEDM has been thoroughly investigated, resolving key problems such as low efficiency, inferior surface quality, and excessive tool wear commonly encountered in high-hardness mold-steel processing. The advantages of low-temperature-treated zinc-coated wire electrodes in high-performance machining are confirmed, and advanced optimization algorithms are incorporated to achieve accurate process modeling and parameter tuning.
SKD61Jiang and Yen. (2021)
[125]
Integration of MTF, CNN and LSTM For SKD61 steel, the proposed MTF-CLSTM method effectively predicted surface roughness during WEDM. By transforming dynamic data into images and extracting spatiotemporal features, the model achieves a minimum MAPE of 2.94%, significantly improving prediction accuracy.
SKD11Shen et al. (2019)
[127]
Deep neural network learning SKD11, a high-hardness and wear-resistant tool steel, exhibits complex discharge behavior during EDM. These variations pose greater demands on the sensitivity and accuracy of detection methods.
Carbon steels and alloy steelsAISI 1045Liu et al. (2021)
[111]
CNN combined with GRU By analyzing spark image features and their relationship with discharge conditions during WEDM of AISI 1045 carbon steel, this study resolves the instability and latency issues inherent in conventional monitoring methods that rely solely on electrical parameters (e.g., voltage and current).
Carbon steelYang et al. (2022)
[112]
BRTCN The use of carbon steel as the workpiece material in WEDM enabled the investigation of discharge behavior, addressing limitations of traditional monitoring methods in terms of efficiency, stability, and real-time performance. Furthermore, AE signals are validated as effective in distinguishing different discharge states, such as short-circuiting, partial short-circuiting, and open-circuiting.
AISI D2Sanchez et al. (2018)
[120]
DNN By integrating the material characteristics of AISI D2 tool steel with deep learning techniques, the study tackles quality degradation in industrial WEDM caused by sudden parameter changes (e.g., workpiece-thickness variation), offering a novel method for real-time monitoring and proactive intervention in intelligent manufacturing of high-precision, high-value components.
Conde et al. (2018)
[121]
LRNN combined with SA When machining AISI D2 tool steel, dimensional accuracy loss caused by wire electrode deformation is effectively mitigated through the combination of an Elman-based LRNN and SA optimization. The proposed method reduces mean deviation by up to 80% and CV by 43%.
Non-ferrous metals and alloysAl6061Shen et al. (2019)
[127]
Deep neural network learning Al6061, a lightweight aluminum alloy with excellent conductivity and machinability, differs significantly from steel in its discharge characteristics. Evaluating this material provides insight into the adaptability of algorithms to different electrical conductivities.
BrassShen et al. (2019)
[127]
Deep neural network learning Brass, favored for its good electrical conductivity and stability, is commonly used in precision-component machining. Analyzing its discharge waveforms further validates the reliability of detection methods under fine-machining conditions.
Special materialsNi-Ti-Hf SMARahul et al. (2023)
[115]
CNN integrated with PSO The use of Ni-Ti-Hf SMAs in WEDM addressed the limitations of conventional NiTi alloys under high-temperature applications. By employing PSO and CNN-based image classification, machining parameters are optimized to enhance MRR and Ra while preserving high-temperature shape memory performance, overcoming processing challenges inherent to this alloy.
Ni55.8Ti SMAVakharia et al. (2022)
[126]
Dense Net-based deep learning approach The challenges of machining Ni55.8Ti SMAs, such as high hardness, rapid work hardening, pseudoelasticity, significant tool wear, and poor surface finish, were effectively addressed using DenseNet deep learning. This approach significantly improves machinability and surface quality where traditional methods fail.

5. Discussion

5.1. Model Analysis Comparison

In this study, a systematic evaluation of various intelligent models was conducted to assess their performance across different machining performance indicators, revealing their potential applications in process optimization. As shown in Figure 11, the performance of different models across multiple machining performance indicators (such as Ra, MRR, KW, wire breakage prediction, discharge state identification, and crack detection) is comprehensively compared and visualized using various evaluation metrics (R2, accuracy, prediction error, RMSE, MAPE, MSE, and MAE).
Figure 11 uses a color-coded approach to visualize the performance of the various models on different WEDM process metrics: green indicates excellent performance, yellow indicates moderate performance, and red represents poor performance. According to the figure, different models display notable variation in their predictive ability and applicability with respect to different process performance indicators. Among them, the DNN-COOT model performs well in the three key indicators of Ra, MRR, and KW, which are all labeled in green, demonstrating high prediction accuracy and stability. This result highlights the advantages of combining deep neural networks with co-optimization algorithms, as doing so is especially suitable for multi-objective co-optimization tasks. The W-ELM model shows outstanding performance in KW prediction, as well as robust performance in Ra and MRR, which shows a good balance between practicality and computational efficiency. In contrast, traditional models such as SVM and BPNN-SAA, while having some advantages in single metrics (e.g., Ra), are weak in handling more challenging tasks (e.g., wire break prediction vs. discharge state identification), making it difficult to meet the demands under complex working conditions. Meanwhile, some deep learning models (e.g., ResNet-50-CNN, CNN-PSO, and CNN-GRU) show red flags in specific tasks, especially poor performance in discharge state identification, which may originate from overfitting or insufficient ability to extract key features. This suggests that deep models must also consider the importance of structural fitness and task-specific feature engineering while improving prediction performance. Overall, deep learning models with integrated optimization algorithms (e.g., DNN-COOT) show stronger adaptability and generalization potential in WEDM process modeling and multidimensional performance optimization, which is a worthy direction for further research and application in the future.

5.2. Comparison and Analysis of Traditional Methods and Artificial Intelligence

The traditional optimization of WEDM processes has primarily relied on empirical methods and statistical techniques [133]. Empirical methods, being the earliest and most commonly utilized approaches, depend heavily on the operator’s accumulated experience and skills developed through long-term practice. Optimization is achieved through trial and error and the fine-tuning of process parameters [134,135,136]. For example, operators typically adjust key parameters such as pulse duration, current intensity, and WFD based on their previous experience to achieve better cutting quality or higher machining efficiency. While this approach is intuitive and flexible, it is also highly subjective and uncertain. It heavily depends on individual expertise, making it difficult to standardize or transfer knowledge and nearly impossible to establish universally applicable parameter optimization rules. In contrast, statistical methods introduce the concepts of mathematical modeling and experimental design. Commonly utilized techniques include RSM, the Taguchi method, and ANOVA [137,138,139]. These methods involve the systematic DOE, data collection from various parameter combinations, and statistical analysis to establish mathematical relationships between input parameters and machining quality [140,141]. It is worth noting that mainstream WEDM equipment is generally equipped with adaptive control systems, which can adjust processing parameters in real time to ensure processing stability and reduce the risk of wire breakage. However, this also means that operators cannot directly set traditional fixed processing parameters; they can only choose from several preset “processing settings.” Therefore, the input parameters used in many studies are actually the control settings of the equipment, rather than the real-time physical parameters during the processing. This reality poses significant challenges for the precise modeling and optimization of process parameters. To address this limitation, this paper focuses on the mapping relationship between processing settings and processing performance during the experimental design and data analysis stages, employing repeated experiments and statistical methods to mitigate the impact of parameter fluctuations caused by adaptive control. Additionally, robust strategies are integrated into the subsequent construction of intelligent optimization models to enhance the model’s adaptability to dynamic changes in internal equipment parameters. With the expected opening of equipment interfaces and the development of real-time monitoring technology, it will become possible in the future to obtain more accurate machining-parameter data, thereby further improving the precision and reliability of WEDM process optimization. Compared to empirical approaches, statistical methods are more scientific and reproducible, allowing the quantification of the effects of different factors on machining outcomes such as Ra, CS, and MRR. However, these methods have limitations. On the one hand, they require extensive experimental data to ensure model reliability, which can be costly and time-consuming. On the other hand, traditional statistical models struggle to accurately capture the behavior of highly nonlinear and strongly coupled multivariable systems, limiting their effectiveness in revealing complex process dynamics. In contrast, the introduction of AI technologies has brought groundbreaking advances in WEDM process optimization [142,143].
AI uses data-driven approaches that do not rely on specific physical models or manual assumptions. Instead, AI can automatically learn and extract complex nonlinear relationships between parameters from large volumes of historical machining data. Typical AI methods include ANNs, SVMs, and, more recently, deep learning techniques. AI not only fits existing data and predicts from it but also exhibits strong generalization capabilities, enabling it to adapt to variations in materials, machine conditions, and processing requirements [144,145]. This significantly enhances the stability, robustness, and intelligence of the WEDM process. Moreover, AI technologies support multi-objective optimization and dynamic adjustment capabilities. They can adapt strategies in real time during the machining process, balancing multiple goals, such as machining precision and cutting efficiency—something that traditional methods struggle to achieve simultaneously. For instance, by integrating evolutionary algorithms with machine-learning models, adaptive optimization systems can be developed that respond autonomously to uncertainties like electrode wear, dielectric fluid contamination, or material hardness fluctuations. These systems continuously refine parameter settings during operation.
At a deeper level, true synergy is reflected in the closed-loop optimization of material–process–performance relationships. For commonly used materials in hydraulic machinery—such as stainless steels and wear-resistant coatings—developing specialized electrode materials and working media tailored to their machining characteristics can significantly reduce the thickness of the recast layer, thereby ensuring adequate fatigue strength under dynamic loading conditions [3,146]. Additionally, the optimization of process parameters should align with the full life-cycle management of hydraulic equipment. For example, controlling discharge energy can help minimize residual stress, enabling processed components to maintain structural stability under long-term water hammer impacts. This interdisciplinary integration of technologies not only enhances the machining quality of individual components but also drives the overall design philosophy of hydraulic equipment toward functional integration and environmentally sustainable manufacturing. This intelligent nature positions WEDM for a transformation toward “self-learning, self-decision-making, and self-adaptation,” driving the process into a new era of efficient, intelligent, and sustainable manufacturing.

5.3. Analysis of the Advantages and Disadvantages of Traditional Machine Learning and Deep Learning

In the optimization of WEDM processes, both traditional machine-learning methods (such as ANN and SVM) and deep-learning approaches offer distinct advantages and limitations. Traditional machine-learning models are characterized by simpler architectures and faster training speeds, making them suitable for small to medium-sized datasets and parameter optimization tasks. For instance, ANNs can model nonlinear relationships through hidden layers and have shown strong performance in predicting Ra. SVMs, on the other hand, utilize kernel functions to map input data into high-dimensional spaces, enabling effective classification and optimization of process parameters. However, these methods typically rely on manual feature engineering and are sensitive to data noise, which can limit their generalization ability. In contrast, emerging deep learning methods—such as CNNs and LSTM networks—can automatically extract hierarchical features through multiple layers. They are capable of learning high-level representations directly from raw data (e.g., machining current waveforms and acoustic emission signals), leading to significantly improved model accuracy. For example, CNNs are well-suited for processing image data from WEDM (e.g., micro-crack detection), while LSTMs excel at capturing long-term dependencies in time-series data (e.g., discharge pulse intervals). However, the high performance of deep learning models comes with certain costs [147,148]. They require large volumes of labeled data, consume significant computational resources, and often function as “black boxes,” making them less interpretable. This lack of transparency may reduce user trust and acceptance in industrial settings where stability and safety are critical [149]. Therefore, in practical applications, it is essential to balance model complexity and usability based on task requirements, data availability, and the application context. For example, transfer learning can be employed to mitigate the issue of limited data, and ensemble learning strategies can be adopted to combine traditional machine-learning with deep-learning models. This hybrid approach leverages the strengths of both, ensuring high performance while improving interpretability and enhancing industrial applicability and adaptability.

5.4. Application of Workpiece Materials During the Process of WEDM

In the optimization of WEDM processes, the workpiece material significantly influences machining performance and final outcomes, and the machining strategy must be adapted to accommodate the required material properties. Different materials exhibit distinct characteristics in terms of electrical conductivity, thermophysical properties, and machining response, all of which directly influence the discharge behavior and overall machining quality. Figure 12 presents a statistical summary of the commonly used workpiece materials and their distribution in recent WEDM optimization studies, highlighting the wide applicability of this technique across various industrial materials. WEDM has been extensively applied to a diverse range of workpiece materials, including tool steels, high-temperature alloys, non-ferrous metals and their alloys, and carbon and alloy steels, as well as a variety of advanced materials. Among these, tool steels account for approximately 30% of the studies, underscoring their dominant role due to their high hardness and wear resistance, which are particularly advantageous in mold manufacturing and precision component fabrication. High-temperature alloys, such as Inconel 718, represent around 14% of the applications, owing to their superior thermal strength and oxidation resistance, which are critical for aerospace and high-temperature service components [150]. Non-ferrous metals like titanium and aluminum alloys are favored in industries requiring lightweight structures and high thermal conductivity, such as medical devices and electronics. Moreover, with the increasing demand for functional materials, WEDM has been progressively employed for processing emerging materials such as shape memory alloys and metallic glasses, demonstrating promising adaptability and future potential. These findings highlight the strong industrial relevance of WEDM technology. To facilitate implementation in real-world manufacturing environments, the insights gained from material-specific studies can be integrated into digital machining databases or intelligent control platforms. Such tools can assist engineers in selecting optimal process parameters for specific materials, thereby improving machining efficiency, reducing trial-and-error costs, and accelerating the development of high-performance components.

6. Outlook

With the continuous advancement of precision manufacturing and intelligent machining technologies, the optimization of WEDM processes is encountering new opportunities and challenges. Although current optimization approaches—ranging from traditional empirical methods to advanced machine-learning and deep-learning techniques—have significantly improved machining performance, there remains considerable room for improvement in terms of adaptability, real-time control capabilities, and model interpretability. Looking ahead, future research should place greater emphasis on bridging theoretical innovation with practical industrial applications. The following outlook highlights several promising directions worthy of further exploration:
  • Traditional optimization methods are expected to achieve breakthroughs in areas such as intelligent integration, real-time responsiveness, and adaptive modeling. By incorporating intelligent algorithms such as GA and PSO, multi-objective optimization capabilities can be significantly enhanced. Real-time performance can be improved through dynamic adjustment based on sensor feedback, while the integration of fuzzy logic and grey system theory can increase adaptability and robustness under complex machining conditions. Traditional methods are poised to gain renewed vitality through their integration with intelligent technologies.
  • The development of machine learning will focus on enhancing generalization, enabling lightweight deployment, and establishing closed-loop data systems. Transfer learning and ensemble methods will improve model adaptability across different materials and equipment. Model compression and edge deployment will support on-site industrial applications. By integrating active learning and feedback mechanisms, a “prediction–optimization” closed-loop control system can be achieved, enhancing the practicality of models under small-sample conditions.
  • Deep learning will continue to advance in the areas of cross-modal integration, edge intelligence, and enhanced interpretability. Multimodal models (such as Transformers) can fuse images, signals, and process data to achieve more refined state perception. Lightweight models (such as Mobile Net) will facilitate the implementation of real-time optimization. Explainable AI (XAI) techniques, including attention mechanisms and causal inference, will help improve model transparency, promoting a shift from “passive optimization” to “autonomous decision-making.”
  • Collaborative innovation is a matter of concern, thus balancing the relationship between technological progress and the costs of industrialization. Looking ahead, there is a need to establish a seamless integration between process databases and design standards so as to develop a knowledge base of machining parameters suitable for specific operating conditions. In addition, the application of digital twin technology could facilitate the virtual validation of process results. This systematic optimization path will accelerate the transformation of WEDM machining from mere “machinability” to “optimized machining.” For example, in the field of hydraulic engineering, continuous support is provided for the reliable manufacturing of large hydraulic engineering equipment.

Author Contributions

Conceptualization, S.D. and W.M.; methodology, S.D., W.M. and X.Z.; formal analysis, B.D. and X.Z.; investigation, B.D. and S.D.; data curation, B.D. and X.Z.; writing and original draft preparation, X.Z. and B.D.; writing review and editing, X.Z. and W.M.; supervision, S.D. and W.M.; funding acquisition, W.M. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the Guangdong Basic and Applied Basic Research Foundation (No. 2022A1515140083 and 2022A1515140066), and the Science and Technology Development Program of Henan Province (No. 252102210253).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

No new data were created in this study.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Spedding, T.A.; Wang, Z.Q. Study on Modeling of Wire EDM Process. J. Mater. Process. Technol. 1997, 69, 18–28. [Google Scholar] [CrossRef]
  2. Ho, K.H.; Newman, S.T.; Rahimifard, S.; Allen, R.D. State of the Art in Wire Electrical Discharge Machining (WEDM). Int. J. Mach. Tools Manuf. 2004, 44, 1247–1259. [Google Scholar] [CrossRef]
  3. Slătineanu, L.; Dodun, O.; Coteaţă, M.; Nagîţ, G.; Băncescu, I.B.; Hriţuc, A. Wire Electrical Discharge Machining—A Review. Machines 2020, 8, 69. [Google Scholar] [CrossRef]
  4. Natarajan, K.; Ramakrishnan, H.; Gacem, A.; Vijayan, V.; Karthiga, K.; Ali, H.E.; Prakash, B.; Mekonnen, A. Study on Optimization of WEDM Process Parameters on Stainless Steel. J. Nanomater. 2022, 2022, 6765721. [Google Scholar] [CrossRef]
  5. Camposeco-Negrete, C. Analysis and Optimization of Sustainable Machining of AISI O1 Tool Steel by the Wire-EDM Process. Adv. Manuf. 2021, 9, 304–317. [Google Scholar] [CrossRef]
  6. Chakraborty, S.; Mitra, S.; Bose, D. Evaluation of Response Characteristics Using Sensitivity Analysis and TLBO Technique of Powder Mixed Wire EDM of Ti6Al4V Alloy. CIRP J. Manuf. Sci. Technol. 2023, 47, 260–272. [Google Scholar] [CrossRef]
  7. Zheng, J.; Ren, Y.; Qi, T.; Lin, F.; Shi, J.; Hu, X.; Pan, Q.; Yao, J.; Ling, W.; Guan, A.; et al. Modeling and Optimization of Energy Consumption in Wire Cut Electrical Discharge Machining. Comput. Ind. Eng. 2024, 191, 110167. [Google Scholar] [CrossRef]
  8. Liao, Y.S.; Woo, J.C. Design of a Fuzzy Controller for the Adaptive Control of WEDM Process. Int. J. Mach. Tools Manuf. 2000, 40, 2293–2307. [Google Scholar] [CrossRef]
  9. Takale, A.; Chougule, N. Optimization of Process Parameters of Wire Electro Discharge Machining for Ti49.4Ni50.6 Shape Memory Alloys Using the Taguchi Technique. Int. J. Struct. Integr. 2019, 10, 548–568. [Google Scholar] [CrossRef]
  10. Wang, S.-M.; Wu, J.-X.; Gunawan, H.; Tu, R.-Q. Optimization of Machining Parameters for Corner Accuracy Improvement for WEDM Processing. Appl. Sci. 2022, 12, 10324. [Google Scholar] [CrossRef]
  11. Abhilash, P.M.; Chakradhar, D. Wire EDM Failure Prediction and Process Control Based on Sensor Fusion and Pulse Train Analysis. Int. J. Adv. Manuf. Technol. 2022, 118, 1453–1467. [Google Scholar] [CrossRef]
  12. Abhilash, P.M.; Chakradhar, D. Performance Monitoring and Failure Prediction System for Wire Electric Discharge Machining Process through Multiple Sensor Signals. Mach. Sci. Technol. 2022, 26, 245–275. [Google Scholar] [CrossRef]
  13. Ozcanli, A.K.; Yaprakdal, F.; Baysal, M. Deep Learning Methods and Applications for Electrical Power Systems: A Comprehensive Review. Int. J. Energy Res. 2020, 44, 7136–7157. [Google Scholar] [CrossRef]
  14. Cen, J.; Yang, Z.; Liu, X.; Xiong, J.; Chen, H. A Review of Data-Driven Machinery Fault Diagnosis Using Machine Learning Algorithms. J. Vib. Eng. Technol. 2022, 10, 2481–2507. [Google Scholar] [CrossRef]
  15. Berus, L.; Hernavs, J.; Potocnik, D.; Sket, K.; Ficko, M. Enhancing Manufacturing Precision: Leveraging Motor Currents Data of Computer Numerical Control Machines for Geometrical Accuracy Prediction Through Machine Learning. Sensors 2025, 25, 169. [Google Scholar] [CrossRef]
  16. Zhang, L.; Du, J.; Zhuang, X.; Wang, Z.; Pei, J. Geometric Prediction of Conic Tool in Micro-EDM Milling with Fix-Length Compensation Using Simulation. Int. J. Mach. Tools Manuf. 2015, 89, 86–94. [Google Scholar] [CrossRef]
  17. Pramanik, A.; Basak, A.K.; Littlefair, G.; Debnath, S.; Prakash, C.; Singh, M.A.; Marla, D.; Singh, R.K. Methods and Variables in Electrical Discharge Machining of Titanium Alloy—A Review. Heliyon 2020, 6, e05554. [Google Scholar] [CrossRef]
  18. Zeng, K.; Wu, X.; Jiang, F.; Shen, J.; Zhu, L.; Li, L. A Comprehensive Review on the Cutting and Abrasive Machining of Cemented Carbide Materials. J. Manuf. Process. 2023, 108, 335–358. [Google Scholar] [CrossRef]
  19. Koshy, P.; Dumitrescu, P.; Ziada, Y. Novel Methods for Rapid Assessment of Tool Performance in Milling. Int. J. Mach. Tools Manuf. 2004, 44, 1599–1605. [Google Scholar] [CrossRef]
  20. Savas, V.; Ozay, C. The Optimization of the Surface Roughness in the Process of Tangential Turn-Milling Using Genetic Algorithm. Int. J. Adv. Manuf. Technol. 2008, 37, 335–340. [Google Scholar] [CrossRef]
  21. Sarıkaya, M.; Yılmaz, V.; Güllü, A. Analysis of Cutting Parameters and Cooling/Lubrication Methods for Sustainable Machining in Turning of Haynes 25 Superalloy. J. Clean. Prod. 2016, 133, 172–181. [Google Scholar] [CrossRef]
  22. Fuse, K.; Dalsaniya, A.; Modi, D.; Vora, J.; Pimenov, D.Y.; Giasin, K.; Prajapati, P.; Chaudhari, R.; Wojciechowski, S. Integration of Fuzzy AHP and Fuzzy TOPSIS Methods for Wire Electric Discharge Machining of Titanium (Ti6Al4V) Alloy Using RSM. Materials 2021, 14, 7408. [Google Scholar] [CrossRef] [PubMed]
  23. Mahapatra, S.S.; Patnaik, A. Parametric Optimization of Wire Electrical Discharge Machining (WEDM) Process Using Taguchi Method. J. Braz. Soc. Mech. Sci. Eng. 2006, 28, 422–429. [Google Scholar] [CrossRef]
  24. Mukherjee, R.; Chakraborty, S.; Samanta, S. Selection of Wire Electrical Discharge Machining Process Parameters Using Non-Traditional Optimization Algorithms. Appl. Soft Comput. 2012, 12, 2506–2516. [Google Scholar] [CrossRef]
  25. Çaydaş, U.; Hasçalık, A.; Ekici, S. An Adaptive Neuro-Fuzzy Inference System (ANFIS) Model for Wire-EDM. Expert Syst. Appl. 2009, 36, 6135–6139. [Google Scholar] [CrossRef]
  26. Ishfaq, K.; Sana, M.; Ashraf, W.M. Artificial Intelligence—Built Analysis Framework for the Manufacturing Sector: Performance Optimization of Wire Electric Discharge Machining System. Int. J. Adv. Manuf. Technol. 2023, 128, 5025–5039. [Google Scholar] [CrossRef]
  27. Grigoriev, S.N.; Pivkin, P.M.; Kozochkin, M.P.; Volosova, M.A.; Okunkova, A.A.; Porvatov, A.N.; Zelensky, A.A.; Nadykto, A.B. Physicomechanical Nature of Acoustic Emission Preceding Wire Breakage during Wire Electrical Discharge Machining (WEDM) of Advanced Cutting Tool Materials. Metals 2021, 11, 1865. [Google Scholar] [CrossRef]
  28. Zheng, J.; Zheng, W.; Chen, A.; Yao, J.; Ren, Y.; Zhou, C.; Wu, J.; Ling, W.; Bai, B.; Wang, W.; et al. Sustainability of Unconventional Machining Industry Considering Impact Factors and Reduction Methods of Energy Consumption: A Review and Analysis. Sci. Total Environ. 2020, 722, 137897. [Google Scholar] [CrossRef]
  29. Choudhary, K.; DeCost, B.; Chen, C.; Jain, A.; Tavazza, F.; Cohn, R.; Park, C.W.; Choudhary, A.; Agrawal, A.; Billinge, S.J.L.; et al. Recent Advances and Applications of Deep Learning Methods in Materials Science. npj Comput. Mater. 2022, 8, 59. [Google Scholar] [CrossRef]
  30. Song, J.; Wang, B.; Hao, X. Optimization Algorithms and Their Applications and Prospects in Manufacturing Engineering. Materials 2024, 17, 4093. [Google Scholar] [CrossRef]
  31. Sun, Y.; Zeng, W.; Ma, X.; Xu, B.; Liang, X.; Zhang, J. A Hybrid Approach for Processing Parameters Optimization of Ti-22Al-25Nb Alloy during Hot Deformation Using Artificial Neural Network and Genetic Algorithm. Intermetallics 2011, 19, 1014–1019. [Google Scholar] [CrossRef]
  32. Liu, S.; Yang, C. Machine Learning Design for High-Entropy Alloys: Models and Algorithms. Metals 2024, 14, 235. [Google Scholar] [CrossRef]
  33. Durodola, J.F. Machine Learning for Design, Phase Transformation and Mechanical Properties of Alloys. Prog. Mater. Sci. 2022, 123, 100797. [Google Scholar] [CrossRef]
  34. Gu, Z.; Sharma, S.; Riley, D.A.; Pantawane, M.V.; Joshi, S.S.; Fu, S.; Dahotre, N.B. A Universal Predictor-Based Machine Learning Model for Optimal Process Maps in Laser Powder Bed Fusion Process. J. Intell. Manuf. 2023, 34, 3341–3363. [Google Scholar] [CrossRef]
  35. Alzubaidi, L.; Zhang, J.; Humaidi, A.J.; Al-Dujaili, A.; Duan, Y.; Al-Shamma, O.; Santamaría, J.; Fadhel, M.A.; Al-Amidie, M.; Farhan, L. Review of Deep Learning: Concepts, CNN Architectures, Challenges, Applications, Future Directions. J. Big Data 2021, 8, 53. [Google Scholar] [CrossRef] [PubMed]
  36. Holm, E.A. In Defense of the Black Box. Science 2019, 364, 26–27. [Google Scholar] [CrossRef]
  37. Ma, W.; Cheng, F.; Xu, Y.; Wen, Q.; Liu, Y. Probabilistic Representation and Inverse Design of Metamaterials Based on a Deep Generative Model with Semi-Supervised Learning Strategy. Adv. Mater. 2019, 31, 1901111. [Google Scholar] [CrossRef]
  38. Pendokhare, D.; Chakraborty, S. A Review on Multi-Objective Optimization Techniques of Wire Electrical Discharge Machining. Arch. Comput. Methods Eng. 2024, 32, 1797–1839. [Google Scholar] [CrossRef]
  39. Liao, Y.S.; Yan, M.T.; Chang, C.C. A Neural Network Approach for the On-Line Estimation of Workpiece Height in WEDM. J. Mater. Process. Technol. 2002, 121, 252–258. [Google Scholar] [CrossRef]
  40. Yan, M.-T.; Chien, H.-T. Monitoring and Control of the Micro Wire-EDM Process. Int. J. Mach. Tools Manuf. 2007, 47, 148–157. [Google Scholar] [CrossRef]
  41. Modrak, V.; Pandian, R.S.; Kumar, S.S. Parametric Study of Wire-EDM Process in Al-Mg-MoS2 Composite Using NSGA-II and MOPSO Algorithms. Processes 2021, 9, 469. [Google Scholar] [CrossRef]
  42. Jadam, T.; Datta, S.; Masanta, M. Study of Surface Integrity and Machining Performance during Main/Rough Cut and Trim/Finish Cut Mode of WEDM on Ti–6Al–4V: Effects of Wire Material. J. Braz. Soc. Mech. Sci. Eng. 2019, 41, 151. [Google Scholar] [CrossRef]
  43. Hewidy, M.S.; El-Taweel, T.A.; El-Safty, M.F. Modelling the Machining Parameters of Wire Electrical Discharge Machining of Inconel 601 Using RSM. J. Mater. Process. Technol. 2005, 169, 328–336. [Google Scholar] [CrossRef]
  44. Kung, K.-Y.; Chiang, K.-T. Modeling and Analysis of Machinability Evaluation in the Wire Electrical Discharge Machining (WEDM) Process of Aluminum Oxide-Based Ceramic. Mater. Manuf. Process. 2008, 23, 241–250. [Google Scholar] [CrossRef]
  45. Li, Z.; Grandhi, R.V.; Srinivasan, R. Distortion Minimization during Gas Quenching Process. J. Mater. Process. Technol. 2006, 172, 249–257. [Google Scholar] [CrossRef]
  46. Kumar, K.; Agarwal, S. Multi-Objective Parametric Optimization on Machining with Wire Electric Discharge Machining. Int. J. Adv. Manuf. Technol. 2012, 62, 617–633. [Google Scholar] [CrossRef]
  47. Srivastava, V.; Basu, B.; Prabhu, N. Application of Machine Learning (ML)-Based Multi-Classifications to Identify Corrosion Fatigue Cracking Phenomena in Naval Steel Weldments. Mater. Today Commun. 2024, 39, 108591. [Google Scholar] [CrossRef]
  48. Bai, Y.; Sun, Z.; Zeng, B.; Long, J.; Li, L.; de Oliveira, J.V.; Li, C. A Comparison of Dimension Reduction Techniques for Support Vector Machine Modeling of Multi-Parameter Manufacturing Quality Prediction. J. Intell. Manuf. 2019, 30, 2245–2256. [Google Scholar] [CrossRef]
  49. Yang, N.; Yang, C.; Xing, C.; Ye, D.; Jia, J.; Chen, D.; Shen, X.; Huang, Y.; Zhang, L.; Zhu, B. Deep Learning-Based SCUC Decision-Making: An Intelligent Data-Driven Approach with Self-Learning Capabilities. IET Gener. Transm. Distrib. 2022, 16, 629–640. [Google Scholar] [CrossRef]
  50. Li, L.; Rong, S.; Wang, R.; Yu, S. Recent Advances in Artificial Intelligence and Machine Learning for Nonlinear Relationship Analysis and Process Control in Drinking Water Treatment: A Review. Chem. Eng. J. 2021, 405, 126673. [Google Scholar] [CrossRef]
  51. Cai, M.; Pipattanasomporn, M.; Rahman, S. Day-Ahead Building-Level Load Forecasts Using Deep Learning vs. Tradit. Time-Ser. Techniques. Appl. Energy 2019, 236, 1078–1088. [Google Scholar] [CrossRef]
  52. Mayer, R.; Jacobsen, H.-A. Scalable Deep Learning on Distributed Infrastructures: Challenges, Techniques, and Tools. ACM Comput. Surv. 2021, 53, 1–37. [Google Scholar] [CrossRef]
  53. Wang, J.; Ma, Y.; Zhang, L.; Gao, R.X.; Wu, D. Deep Learning for Smart Manufacturing: Methods and Applications. J. Manuf. Syst. 2018, 48, 144–156. [Google Scholar] [CrossRef]
  54. Sharma, P.; Chakradhar, D.; Narendranath, S. Analysis and Optimization of WEDM Performance Characteristics of Inconel 706 for Aerospace Application. Silicon 2018, 10, 921–930. [Google Scholar] [CrossRef]
  55. Ikram, A.; Mufti, N.A.; Saleem, M.Q.; Khan, A.R. Parametric Optimization for Surface Roughness, Kerf and MRR in Wire Electrical Discharge Machining (WEDM) Using Taguchi Design of Experiment. J. Mech. Sci. Technol. 2013, 27, 2133–2141. [Google Scholar] [CrossRef]
  56. Bobbili, R.; Madhu, V.; Gogia, A.K. Effect of Wire-EDM Machining Parameters on Surface Roughness and Material Removal Rate of High Strength Armor Steel. Mater. Manuf. Process. 2013, 28, 364–368. [Google Scholar] [CrossRef]
  57. Thankachan, T.; Soorya Prakash, K.; Loganathan, M. WEDM Process Parameter Optimization of FSPed Copper-BN Composites. Mater. Manuf. Process. 2018, 33, 350–358. [Google Scholar] [CrossRef]
  58. Kavimani, V.; Soorya Prakash, K.; Thankachan, T. Multi-Objective Optimization in WEDM Process of Graphene—SiC-Magnesium Composite through Hybrid Techniques. Measurement 2019, 145, 335–349. [Google Scholar] [CrossRef]
  59. Xiao, Y.; Ming, W.; Shen, D.; He, W.; Ma, J.; Jiao, J. Wolf Pack Algorithm for Optimisation of Cutting Parameters in WEDM Using Taguchi Method. Int. J. Internet Manuf. Serv. 2019, 6, 139–154. [Google Scholar] [CrossRef]
  60. Zaman, U.K.U.; Khan, U.A.; Aziz, S.; Baqai, A.A.; Butt, S.U.; Hussain, D.; Siadat, A.; Jung, D.W. Optimization of Wire Electric Discharge Machining (WEDM) Process Parameters for AISI 1045 Medium Carbon Steel Using Taguchi Design of Experiments. Materials 2022, 15, 7846. [Google Scholar] [CrossRef]
  61. Deshmukh, S.S.; Shaikh Zubair, A.; Jadhav, V.S.; Shrivastava, R. Optimization of Process Parameters of Wire Electric Discharge Machining on AISI 4140 Using Taguchi Method and Grey Relational Analysis. Mater. Today Proc. 2019, 18, 4261–4270. [Google Scholar] [CrossRef]
  62. Khan, Z.A.; Siddiquee, A.N.; Khan, N.Z.; Khan, U.; Quadir, G.A. Multi Response Optimization of Wire Electrical Discharge Machining Process Parameters Using Taguchi Based Grey Relational Analysis. Procedia Mater. Sci. 2014, 6, 1683–1695. [Google Scholar] [CrossRef]
  63. Lingadurai, K.; Nagasivamuni, B.; Muthu Kamatchi, M.; Palavesam, J. Selection of Wire Electrical Discharge Machining Process Parameters on Stainless Steel AISI Grade-304 Using Design of Experiments Approach. J. Inst. Eng. India Ser. C 2012, 93, 163–170. [Google Scholar] [CrossRef]
  64. Singh, V.; Bhandari, R.; Yadav, V.K. An Experimental Investigation on Machining Parameters of AISI D2 Steel Using WEDM. Int. J. Adv. Manuf. Technol. 2017, 93, 203–214. [Google Scholar] [CrossRef]
  65. Sarkar, S.; Sekh, M.; Mitra, S.; Bhattacharyya, B. Modeling and Optimization of Wire Electrical Discharge Machining of γ-TiAl in Trim Cutting Operation. J. Mater. Process. Technol. 2008, 205, 376–387. [Google Scholar] [CrossRef]
  66. Sivaprakasam, P.; Hariharan, P.; Gowri, S. Optimization of Micro-WEDM Process of Aluminum Matrix Composite (A413-B4C): A Response Surface Approach. Mater. Manuf. Process. 2013, 28, 1340–1347. [Google Scholar] [CrossRef]
  67. Soundararajan, R.; Ramesh, A.; Mohanraj, N.; Parthasarathi, N. An Investigation of Material Removal Rate and Surface Roughness of Squeeze Casted A413 Alloy on WEDM by Multi Response Optimization Using RSM. J. Alloys Compd. 2016, 685, 533–545. [Google Scholar] [CrossRef]
  68. Mohanraj, T.; Sakthivel, G.; Pramanik, A. Use of RSM Desirability Approach to Optimize WEDM of Mild Steel. Phys. Scr. 2024, 99, 105976. [Google Scholar] [CrossRef]
  69. Yan, M.T.; Liao, Y.S. Adaptive Control of the WEDM Process Using the Fuzzy Control Strategy. J. Manuf. Syst. 1998, 17, 263–274. [Google Scholar] [CrossRef]
  70. Lin, C.-T.; Chung, I.-F.; Huang, S.-Y. Improvement of Machining Accuracy by Fuzzy Logic at Corner Parts for Wire-EDM. Fuzzy Sets Syst. 2001, 122, 499–511. [Google Scholar] [CrossRef]
  71. Fard, R.K.; Afza, R.A.; Teimouri, R. Experimental Investigation, Intelligent Modeling and Multi-Characteristics Optimization of Dry WEDM Process of Al–SiC Metal Matrix Composite. J. Manuf. Process. 2013, 15, 483–494. [Google Scholar] [CrossRef]
  72. Maher, I.; Sarhan, A.A.D.; Barzani, M.M.; Hamdi, M. Increasing the Productivity of the Wire-Cut Electrical Discharge Machine Associated with Sustainable Production. J. Clean. Prod. 2015, 108, 247–255. [Google Scholar] [CrossRef]
  73. Goyal, A.; Gautam, N.; Pathak, V.K. An Adaptive Neuro-Fuzzy and NSGA-II-Based Hybrid Approach for Modelling and Multi-Objective Optimization of WEDM Quality Characteristics during Machining Titanium Alloy. Neural Comput. Appl. 2021, 33, 16659–16674. [Google Scholar] [CrossRef]
  74. Jithendra, T.; Basha, S.S.; Das, R.; Gajjela, R. Modeling and Optimization of WEDM of Monel 400 Alloy Using ANFIS and Snake Optimizer: A Comparative Study. Proc. Inst. Mech. Eng. Part C J. Mech. Eng. Sci. 2024, 238, 1573–1589. [Google Scholar] [CrossRef]
  75. Naresh, C.; Bose, P.S.C.; Rao, C.S.P. Artificial Neural Networks and Adaptive Neuro-Fuzzy Models for Predicting WEDM Machining Responses of Nitinol Alloy: Comparative Study. SN Appl. Sci. 2020, 2, 314. [Google Scholar] [CrossRef]
  76. Tosun, N. The Effect of the Cutting Parameters on Performance of WEDM. KSME Int. J. 2003, 17, 816–824. [Google Scholar] [CrossRef]
  77. Yuan, J.; Wang, K.; Yu, T.; Fang, M. Reliable Multi-Objective Optimization of High-Speed WEDM Process Based on Gaussian Process Regression. Int. J. Mach. Tools Manuf. 2008, 48, 47–60. [Google Scholar] [CrossRef]
  78. Pasam, V.K.; Battula, S.B.; Madar Valli, P.; Swapna, M. Optimizing Surface Finish in WEDM Using the Taguchi Parameter Design Method. J. Braz. Soc. Mech. Sci. Eng. 2010, 32, 107–113. [Google Scholar] [CrossRef]
  79. Ming, W.; Zhang, Z.; Zhang, G.; Huang, Y.; Guo, J.; Chen, Y. Multi-Objective Optimization of 3D-Surface Topography of Machining YG15 in WEDM. Mater. Manuf. Process. 2014, 29, 514–525. [Google Scholar] [CrossRef]
  80. Nguyen, T.-T.; Duong, Q.-D. Optimization of WEDM Process of Mould Material Using Kriging Model to Improve Technological Performances. Sādhanā 2019, 44, 154. [Google Scholar] [CrossRef]
  81. Xu, J.; Li, M.; Zhong, J.; Hou, Y.; Xia, S.; Yu, P. Process Parameter Modeling and Multi-Response Optimization of Wire Electrical Discharge Machining NiTi Shape Memory Alloy. Mater. Today Commun. 2022, 33, 104252. [Google Scholar] [CrossRef]
  82. AbouHawa, M.; Eissa, A. Corner Cutting Accuracy for Thin-Walled CFRPC Parts Using HS-WEDM. Discov. Appl. Sci. 2024, 6, 130. [Google Scholar] [CrossRef]
  83. Singh, R.; Hussain, S.A.I.; Dash, A.; Rai, R.N. Modelling and Optimizing Performance Parameters in the Wire-Electro Discharge Machining of Al5083/B4C Composite by Multi-Objective Response Surface Methodology. J. Braz. Soc. Mech. Sci. Eng. 2020, 42, 344. [Google Scholar] [CrossRef]
  84. Ming, W.; Hou, J.; Zhang, Z.; Huang, H.; Xu, Z.; Zhang, G.; Huang, Y. Integrated ANN-LWPA for Cutting Parameter Optimization in WEDM. Int. J. Adv. Manuf. Technol. 2016, 84, 1277–1294. [Google Scholar] [CrossRef]
  85. Abhilash, P.M.; Chakradhar, D. Prediction and Analysis of Process Failures by ANN Classification during Wire-EDM of Inconel 718. Adv. Manuf. 2020, 8, 519–536. [Google Scholar] [CrossRef]
  86. Shandilya, P.; Jain, P.K.; Jain, N.K. Neural Network Based Modeling in Wire Electric Discharge Machining of SiCp/6061 Aluminum Metal Matrix Composite. Adv. Mater. Res. 2012, 383–390, 6679–6683. [Google Scholar] [CrossRef]
  87. Chou, P.-H.; Hwang, Y.-R.; Yan, B.-H. The Study of Machine Learning for Wire Rupture Prediction in WEDM. Int. J. Adv. Manuf. Technol. 2022, 119, 1301–1311. [Google Scholar] [CrossRef]
  88. Paturi, U.M.R.; Cheruku, S.; Salike, S.; Pasunuri, V.P.K.; Reddy, N.S. Estimation of Machinability Performance in Wire-EDM on Titanium Alloy Using Neural Networks. Mater. Manuf. Process. 2022, 37, 1073–1084. [Google Scholar] [CrossRef]
  89. Saha, P.; Singha, A.; Pal, S.K.; Saha, P. Soft Computing Models Based Prediction of Cutting Speed and Surface Roughness in Wire Electro-Discharge Machining of Tungsten Carbide Cobalt Composite. Int. J. Adv. Manuf. Technol. 2008, 39, 74–84. [Google Scholar] [CrossRef]
  90. Chen, H.-C.; Lin, J.-C.; Yang, Y.-K.; Tsai, C.-H. Optimization of Wire Electrical Discharge Machining for Pure Tungsten Using a Neural Network Integrated Simulated Annealing Approach. Expert Syst. Appl. 2010, 37, 7147–7153. [Google Scholar] [CrossRef]
  91. Zhang, G.; Zhang, Z.; Guo, J.; Ming, W.; Li, M.; Huang, Y. Modeling and Optimization of Medium-Speed WEDM Process Parameters for Machining SKD11. Mater. Manuf. Process. 2013, 28, 1124–1132. [Google Scholar] [CrossRef]
  92. Zhang, Z.; Ming, W.; Huang, H.; Chen, Z.; Xu, Z.; Huang, Y.; Zhang, G. Optimization of Process Parameters on Surface Integrity in Wire Electrical Discharge Machining of Tungsten Tool YG15. Int. J. Adv. Manuf. Technol. 2015, 81, 1303–1317. [Google Scholar] [CrossRef]
  93. Yusoff, Y.; Mohd Zain, A.; Sharif, S.; Sallehuddin, R.; Ngadiman, M.S. Potential ANN Prediction Model for Multiperformances WEDM on Inconel 718. Neural Comput. Appl. 2018, 30, 2113–2127. [Google Scholar] [CrossRef]
  94. Singh, B.; Misra, J.P. Surface Finish Analysis of Wire Electric Discharge Machined Specimens by RSM and ANN Modeling. Measurement 2019, 137, 225–237. [Google Scholar] [CrossRef]
  95. Huang, G.; Xia, W.; Qin, L.; Zhao, W. Online Workpiece Height Estimation for Reciprocated Traveling Wire EDM Based on Support Vector Machine. Procedia CIRP 2018, 68, 126–131. [Google Scholar] [CrossRef]
  96. Nain, S.S.; Garg, D.; Kumar, S. Evaluation and Analysis of Cutting Speed, Wire Wear Ratio, and Dimensional Deviation of Wire Electric Discharge Machining of Super Alloy Udimet-L605 Using Support Vector Machine and Grey Relational Analysis. Adv. Manuf. 2018, 6, 225–246. [Google Scholar] [CrossRef]
  97. Bijeta Nayak, B.; Sankar Mahapatra, S. An Intelligent Approach for Prediction of Angular Error in Taper Cutting Using Wire-EDM. Mater. Today Proc. 2018, 5, 6121–6127. [Google Scholar] [CrossRef]
  98. Paturi, U.M.R.; Cheruku, S.; Pasunuri, V.P.K.; Salike, S.; Reddy, N.S.; Cheruku, S. Machine Learning and Statistical Approach in Modeling and Optimization of Surface Roughness in Wire Electrical Discharge Machining. Mach. Learn. Appl. 2021, 6, 100099. [Google Scholar] [CrossRef]
  99. Sharma, V.; Misra, J.P.; Singhal, S. Surface Roughness Modeling Using Machine Learning Approaches for Wire Electro-Spark Machining of Titanium Alloy. Int. J. Struct. Integr. 2022, 13, 999–1012. [Google Scholar] [CrossRef]
  100. Zhang, Z.; Ming, W.; Zhang, G.; Huang, Y.; Wen, X.; Huang, H. A New Method for On-Line Monitoring Discharge Pulse in WEDM-MS Process. Int. J. Adv. Manuf. Technol. 2015, 81, 1403–1418. [Google Scholar] [CrossRef]
  101. Nain, S.S.; Garg, D.; Kumar, S. Performance Evaluation of the WEDM Process of Aeronautics Super Alloy. Mater. Manuf. Process. 2018, 33, 1793–1808. [Google Scholar] [CrossRef]
  102. Upadhyay, V.; Misra, J.P.; Singh, B. Wire-Breakage Prediction during WEDM of Ni-Based Superalloy Using Machine Learning-Based Classifier Approaches. Int. J. Interact. Des. Manuf. 2024, 18, 3739–3749. [Google Scholar] [CrossRef]
  103. Jai Rajesh, P.; Balambica, V.; Achudhan, M. Optimizing Ti6Al4V Alloy Wire Edm Parameters Using Regression Analysis and Metaheuristic Algorithm. J. Phys. Conf. Ser. 2024, 2837, 012045. [Google Scholar] [CrossRef]
  104. Saha, S.; Arunkumar, T.; Debnath, K.; Chaurasia, S. Optimization of Kerf Width in WEDM of Sandwich Woven CFRP-An Ensemble Machine Learning Based Approach. Arab. J. Sci. Eng. 2024, 1–12. [Google Scholar] [CrossRef]
  105. Konda, R. Predicting Machining Rate in Non-Traditional Machining Using Decision Tree Inductive Learning. Ph.D. Thesis, Nova Southeastern University, Fort Lauderdale, FL, USA, 2010. [Google Scholar]
  106. Wang, J.; Sanchez, J.A.; Ayesta, I.; Iturrioz, J.A. Unsupervised Machine Learning for Advanced Tolerance Monitoring of Wire Electrical Discharge Machining of Disc Turbine Fir-Tree Slots. Sensors 2018, 18, 3359. [Google Scholar] [CrossRef] [PubMed]
  107. Ulas, M.; Aydur, O.; Gurgenc, T.; Ozel, C. Surface Roughness Prediction of Machined Aluminum Alloy with Wire Electrical Discharge Machining by Different Machine Learning Algorithms. J. Mater. Res. Technol. 2020, 9, 12512–12524. [Google Scholar] [CrossRef]
  108. Trung-Thanh, N.; Khoa, D.N. Multi-Attribute Optimization of the WEDM Process for Surface Characteristics and Productivity. Teh. Vjesn. 2021, 28, 473–480. [Google Scholar] [CrossRef]
  109. Abhilash, P.M.; Chakradhar, D. Sustainability Improvement of WEDM Process by Analysing and Classifying Wire Rupture Using Kernel-Based Naive Bayes Classifier. J. Braz. Soc. Mech. Sci. Eng. 2021, 43, 64. [Google Scholar] [CrossRef]
  110. Vijayakumar, N.; Chandradass, J. Optimizing Wire Electrical Discharge Machining Parameters for Enhanced Gear Machining Performance: A Machine Learning Approach on 20MnCr5 Steel. J. Mater. Eng. Perform. 2024. [Google Scholar] [CrossRef]
  111. Liu, C.; Yang, X.; Peng, S.; Zhang, Y.; Peng, L.; Zhong, R.Y. Spark Analysis Based on the CNN-GRU Model for WEDM Process. Micromachines 2021, 12, 702. [Google Scholar] [CrossRef]
  112. Yang, X.; Liu, C.; Peng, L.; Peng, S.; Zhang, Y.; Xie, N.; Zhong, R.Y. A New BRTCN Model for Predicting Discharge Status of WEDM Based on Acoustic Emission. J. Manuf. Syst. 2022, 64, 409–423. [Google Scholar] [CrossRef]
  113. Ye, Z.Q.; Zhong, Z.P.; Deng, Y.C.; Zhang, Y.J.; Su, G.K. Discharge state recognition of WEDM based on CNN-GRU deep learning with Mixup data augmentation. Machinery 2023, 50, 8–15. [Google Scholar]
  114. Tang, Y.G.; Su, G.K.; Ye, Z.Q.; Chen, H.; Zhang, Y.J. Online pulse recognition of WEDM based on FPGA high-speed data acquisition. J. Xi’an Technol. Univ. 2024, 44, 273–282. [Google Scholar] [CrossRef]
  115. Rahul, V.M.; Balaji, V.; Narendranath, S. Optimization of Wire-EDM Process Parameters for Ni–Ti-Hf Shape Memory Alloy through Particle Swarm Optimization and CNN-Based SEM-Image Classification. Results Eng. 2023, 18, 101141. [Google Scholar] [CrossRef]
  116. Li, X.; Wang, Y.; Wang, Z. Convolutional Neural Network-Based Wire-Cut Image Recognition and Defect Detection Research. Eng. Res. Express 2025, 7, 015559. [Google Scholar] [CrossRef]
  117. Gonzalez-Sanchez, E.; Saccardo, D.; Esteves, P.B.; Kuffa, M.; Wegener, K. Automatic Characterization of WEDM Single Craters Through AI Based Object Detection. Int. J. Autom. Technol. 2024, 18, 265–275. [Google Scholar] [CrossRef]
  118. Kumar, B.K.; Das, V.C. Study and Optimisation of WEDM Parameters of AISI P20+Ni Using RSM and Hybrid Deep Neural Network. Adv. Mater. Process. Technol. 2023, 9, 1299–1327. [Google Scholar] [CrossRef]
  119. Kumar, B.K.; Das, V.C. Prediction and Optimization of Ultrasonic Vibration Assisted Wire EDM Process for AISI P20 + Ni Using COOT Optimization Algorithm Based Deep Neural Network. J. Vib. Eng. Technol. 2024, 12, 613–632. [Google Scholar] [CrossRef]
  120. Sanchez, J.A.; Conde, A.; Arriandiaga, A.; Wang, J.; Plaza, S. Unexpected Event Prediction in Wire Electrical Discharge Machining Using Deep Learning Techniques. Materials 2018, 11, 1100. [Google Scholar] [CrossRef]
  121. Conde, A.; Arriandiaga, A.; Sanchez, J.A.; Portillo, E.; Plaza, S.; Cabanes, I. High-Accuracy Wire Electrical Discharge Machining Using Artificial Neural Networks and Optimization Techniques. Robot. Comput.-Integr. Manuf. 2018, 49, 24–38. [Google Scholar] [CrossRef]
  122. Kumar, B.K.; Das, V.C. Study and Parameter Optimization with AISI P20 + Ni in Wire EDM Performance Using RSM and Hybrid DBN Based SAR. Int. J. Interact. Des. Manuf. 2023, 17, 679–701. [Google Scholar] [CrossRef]
  123. Kumar, S.; Jayswal, S.C. Deep Learning Approach for Optimizing Material Removal Rate in Wire Electrical Discharge Machining Using Neural Networks. Eng. Res. Express 2025, 7, 015519. [Google Scholar] [CrossRef]
  124. Wang, J.; Sanchez, J.A.; Iturrioz, J.A.; Ayesta, I. Geometrical Defect Detection in the Wire Electrical Discharge Machining of Fir-Tree Slots Using Deep Learning Techniques. Appl. Sci. 2019, 9, 90. [Google Scholar] [CrossRef]
  125. Jiang, J.-R.; Yen, C.-T. Product Quality Prediction for Wire Electrical Discharge Machining with Markov Transition Fields and Convolutional Long Short-Term Memory Neural Networks. Appl. Sci. 2021, 11, 5922. [Google Scholar] [CrossRef]
  126. Vakharia, V.; Vora, J.; Khanna, S.; Chaudhari, R.; Shah, M.; Pimenov, D.Y.; Giasin, K.; Prajapati, P.; Wojciechowski, S. Experimental Investigations and Prediction of WEDMed Surface of Nitinol SMA Using SinGAN and DenseNet Deep Learning Model. J. Mater. Res. Technol. 2022, 18, 325–337. [Google Scholar] [CrossRef]
  127. Shen, D.L.; Liu, D.M.; Du, J.G.; Ming, W.Y. Discharge state detection method for medium-speed WEDM based on deep neural network learning. Mech. Des. Manuf. 2019, 8, 146–149. [Google Scholar]
  128. Wang, F.; Li, J.W.; Zhang, Y.D.; Zhang, Y.J.; Huang, Z.G. Discharge state recognition of WEDM based on a dual-path deep learning network in time and frequency domains. Electro-Mach. Mould. 2022, 23–29, 40. [Google Scholar]
  129. Straka, Ľ.; Čorný, I. Simulation and Optimization of Surface Roughness and Process Performance during Machining of HSS by Micro-WEDM Technology. Micromachines 2024, 15, 372. [Google Scholar] [CrossRef]
  130. Pradhan, B.B.; Masanta, M.; Sarkar, B.R.; Bhattacharyya, B. Investigation of Electro-Discharge Micro-Machining of Titanium Super Alloy. Int. J. Adv. Manuf. Technol. 2009, 41, 1094–1106. [Google Scholar] [CrossRef]
  131. Singh Nain, S.; Garg, D.; Kumar, S. Prediction of the Performance Characteristics of WEDM on Udimet-L605 Using Different Modelling Techniques. Mater. Today Proc. 2017, 4, 546–556. [Google Scholar] [CrossRef]
  132. Phate, M.R.; Toney, S.B. Modeling and Prediction of WEDM Performance Parameters for Al/SiCp MMC Using Dimensional Analysis and Artificial Neural Network. Eng. Sci. Technol. Int. J. 2019, 22, 468–476. [Google Scholar] [CrossRef]
  133. Huang, H.; Zhang, Z.; Ming, W.; Xu, Z.; Zhang, Y. A Novel Numerical Predicting Method of Electric Discharge Machining Process Based on Specific Discharge Energy. Int. J. Adv. Manuf. Technol. 2017, 88, 409–424. [Google Scholar] [CrossRef]
  134. Lodhi, B.K.; Agarwal, S. Optimization of Machining Parameters in WEDM of AISI D3 Steel Using Taguchi Technique. Procedia CIRP 2014, 14, 194–199. [Google Scholar] [CrossRef]
  135. Nayak, B.B.; Mahapatra, S.S. An Intelligent Approach for Multi-Response Optimisation of WEDM Parameters. Int. J. Ind. Syst. Eng. 2017, 25, 197–227. [Google Scholar] [CrossRef]
  136. Sarkar, S.; Sekh, M.; Mitra, S.; Bhattacharyya, B. A Novel Method of Determination of Wire Lag for Enhanced Profile Accuracy in WEDM. Precis. Eng. 2011, 35, 339–347. [Google Scholar] [CrossRef]
  137. Meena, V.K.; Azad, M.S. Grey Relational Analysis of Micro-EDM Machining of Ti-6Al-4V Alloy. Mater. Manuf. Process. 2012, 27, 973–977. [Google Scholar] [CrossRef]
  138. Mahapatra, S.S.; Patnaik, A. Optimization of Wire Electrical Discharge Machining (WEDM) Process Parameters Using Taguchi Method. Int. J. Adv. Manuf. Technol. 2007, 34, 911–925. [Google Scholar] [CrossRef]
  139. Sharma, N.; Khanna, R.; Gupta, R.D.; Sharma, R. Modeling and Multiresponse Optimization on WEDM for HSLA by RSM. Int. J. Adv. Manuf. Technol. 2013, 67, 2269–2281. [Google Scholar] [CrossRef]
  140. Zhang, G.; Chen, Z.; Zhang, Z.; Huang, Y.; Ming, W.; Li, H. A Macroscopic Mechanical Model of Wire Electrode Deflection Considering Temperature Increment in MS-WEDM Process. Int. J. Mach. Tools Manuf. 2014, 78, 41–53. [Google Scholar] [CrossRef]
  141. Ming, W.; Ma, J.; Zhang, Z.; Huang, H.; Shen, D.; Zhang, G.; Huang, Y. Soft Computing Models and Intelligent Optimization System in Electro-Discharge Machining of SiC/Al Composites. Int. J. Adv. Manuf. Technol. 2016, 87, 201–217. [Google Scholar] [CrossRef]
  142. Ming, W.; Shen, F.; Zhang, G.; Liu, G.; Du, J.; Chen, Z. Green Machining: A Framework for Optimization of Cutting Parameters to Minimize Energy Consumption and Exhaust Emissions during Electrical Discharge Machining of Al 6061 and SKD 11. J. Clean. Prod. 2021, 285, 124889. [Google Scholar] [CrossRef]
  143. Chen, Y.; Hu, S.; Li, A.; Cao, Y.; Zhao, Y.; Ming, W. Parameters Optimization of Electrical Discharge Machining Process Using Swarm Intelligence: A Review. Metals 2023, 13, 839. [Google Scholar] [CrossRef]
  144. Elenchezhian, M.R.P.; Vadlamudi, V.; Raihan, R.; Reifsnider, K.; Reifsnider, E. Artificial Intelligence in Real-Time Diagnostics and Prognostics of Composite Materials and Its Uncertainties—A Review. Smart Mater. Struct. 2021, 30, 083001. [Google Scholar] [CrossRef]
  145. Liu, Y.; Yang, Z.; Yu, Z.; Liu, Z.; Liu, D.; Lin, H.; Li, M.; Ma, S.; Avdeev, M.; Shi, S. Generative Artificial Intelligence and Its Applications in Materials Science: Current Situation and Future Perspectives. J. Mater. 2023, 9, 798–816. [Google Scholar] [CrossRef]
  146. Ayesta, I.; Izquierdo, B.; Flaño, O.; Sánchez, J.A.; Albizuri, J.; Avilés, R. Influence of the WEDM Process on the Fatigue Behavior of Inconel® 718. Int. J. Fatigue 2016, 92, 220–233. [Google Scholar] [CrossRef]
  147. Taylor, B.; Marco, V.S.; Wolff, W.; Elkhatib, Y.; Wang, Z. Adaptive Deep Learning Model Selection on Embedded Systems. ACM Sigplan Not. 2018, 53, 31–43. [Google Scholar] [CrossRef]
  148. Sze, V.; Chen, Y.-H.; Yang, T.-J.; Emer, J.S. Efficient Processing of Deep Neural Networks: A Tutorial and Survey. Proc. IEEE 2017, 105, 2295–2329. [Google Scholar] [CrossRef]
  149. Wang, P.; Zhang, Z.; Wei, S.; Hao, B.; Chang, H.; Huang, Y.; Zhang, G. Study on Plasma Behaviour, Ablation Mechanism, and Surface Morphology of CFRP by Underwater Laser-Induced Plasma Micro-Machining. J. Mater. Process. Technol. 2025, 338, 118757. [Google Scholar] [CrossRef]
  150. Chang, H.; Zhang, Z.; Wang, P.; Zhang, G. A Novel Composite Film with Superhydrophobic Graphene for Anti-Icing/Deicing via Chemical-Assisted Magnetically Controllable Picosecond Laser Writing. Mater. Today Phys. 2025, 54, 101726. [Google Scholar] [CrossRef]
Figure 1. Working principle for the process of WEDM for cutting metal materials.
Figure 1. Working principle for the process of WEDM for cutting metal materials.
Metals 15 00706 g001
Figure 3. Fuzzy algorithm: (a) Schematic diagram of the multivariable multi-zone fuzzy controller, the time evolution of the abnormal ratio and shutdown time in the spark-frequency monitoring and control system during the finishing process (Reprinted with permission from Ref. [69]. Copyright 1998, Elsevier). (b) ANFIS architecture of the three-input Sugeno fuzzy model, the relationship between SR and changes in P-ON, IP, WT, and discharge energy (Reprinted with permission from Ref. [72]. Copyright 2015, Elsevier).
Figure 3. Fuzzy algorithm: (a) Schematic diagram of the multivariable multi-zone fuzzy controller, the time evolution of the abnormal ratio and shutdown time in the spark-frequency monitoring and control system during the finishing process (Reprinted with permission from Ref. [69]. Copyright 1998, Elsevier). (b) ANFIS architecture of the three-input Sugeno fuzzy model, the relationship between SR and changes in P-ON, IP, WT, and discharge energy (Reprinted with permission from Ref. [72]. Copyright 2015, Elsevier).
Metals 15 00706 g003
Figure 4. Citation distribution of process optimization techniques (including Taguchi method, response surface methodology, fuzzy algorithms, and regression analysis) in the field of WEDM from 1998 to 2024. (Note: The number of citations shown in the figure is weighted based on the logarithmic transformation of the original values using base 10) (Adapted from Refs. [44,54,55,56,57,58,65,67,69,70,71,72,73,76,77,78,79]).
Figure 4. Citation distribution of process optimization techniques (including Taguchi method, response surface methodology, fuzzy algorithms, and regression analysis) in the field of WEDM from 1998 to 2024. (Note: The number of citations shown in the figure is weighted based on the logarithmic transformation of the original values using base 10) (Adapted from Refs. [44,54,55,56,57,58,65,67,69,70,71,72,73,76,77,78,79]).
Metals 15 00706 g004
Figure 7. Citation distribution of different machine-learning techniques (neural networks, SVM, RF, and other algorithms) in the field of WEDM process optimization, based on the literature statistics from 2008 to 2024. (Note: The citation counts in the figure are weighted by calculating the logarithm of the original values with base 10) (Adapted from Refs. [84,85,88,89,90,91,92,93,94,95,96,98,100,101,106,107,109]).
Figure 7. Citation distribution of different machine-learning techniques (neural networks, SVM, RF, and other algorithms) in the field of WEDM process optimization, based on the literature statistics from 2008 to 2024. (Note: The citation counts in the figure are weighted by calculating the logarithm of the original values with base 10) (Adapted from Refs. [84,85,88,89,90,91,92,93,94,95,96,98,100,101,106,107,109]).
Metals 15 00706 g007
Figure 8. (a) BRTCN model framework diagram, (256-BR (gru)TCN-NE) model prediction results, and the matrix heatmap of real values vs. predicted values (G) (Reprinted with permission from Ref. [112]. Copyright 2022, Elsevier). (b) Workflow diagram and comparison analysis of CNN model’s predicted MRR and Ra values with experimental data (Reprinted from Ref. [115]).
Figure 8. (a) BRTCN model framework diagram, (256-BR (gru)TCN-NE) model prediction results, and the matrix heatmap of real values vs. predicted values (G) (Reprinted with permission from Ref. [112]. Copyright 2022, Elsevier). (b) Workflow diagram and comparison analysis of CNN model’s predicted MRR and Ra values with experimental data (Reprinted from Ref. [115]).
Metals 15 00706 g008
Figure 10. Citation distribution of different deep learning techniques (CNN, DNN, and other deep learning algorithms) in the field of WEDM process optimization, based on the literature review from 2018 to 2024. (Note: The citation count in the figure is weighted using the logarithmic calculation with base 10 of the original values.) (Adapted from Refs. [111,112,113,115,118,120,121,122,124,125,126,127]).
Figure 10. Citation distribution of different deep learning techniques (CNN, DNN, and other deep learning algorithms) in the field of WEDM process optimization, based on the literature review from 2018 to 2024. (Note: The citation count in the figure is weighted using the logarithmic calculation with base 10 of the original values.) (Adapted from Refs. [111,112,113,115,118,120,121,122,124,125,126,127]).
Metals 15 00706 g010
Figure 11. Comparison of different models in WEDM process-optimization performance (Adapted from Refs. [129,130,131,132].
Figure 11. Comparison of different models in WEDM process-optimization performance (Adapted from Refs. [129,130,131,132].
Metals 15 00706 g011
Figure 12. Distribution of workpiece materials used in the optimization of WEDM processes.
Figure 12. Distribution of workpiece materials used in the optimization of WEDM processes.
Metals 15 00706 g012
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Zhao, X.; Dong, B.; Dong, S.; Ming, W. Research Progress on Process Optimization of Metal Materials in Wire Electrical Discharge Machining. Metals 2025, 15, 706. https://doi.org/10.3390/met15070706

AMA Style

Zhao X, Dong B, Dong S, Ming W. Research Progress on Process Optimization of Metal Materials in Wire Electrical Discharge Machining. Metals. 2025; 15(7):706. https://doi.org/10.3390/met15070706

Chicago/Turabian Style

Zhao, Xinfeng, Binghui Dong, Shengwen Dong, and Wuyi Ming. 2025. "Research Progress on Process Optimization of Metal Materials in Wire Electrical Discharge Machining" Metals 15, no. 7: 706. https://doi.org/10.3390/met15070706

APA Style

Zhao, X., Dong, B., Dong, S., & Ming, W. (2025). Research Progress on Process Optimization of Metal Materials in Wire Electrical Discharge Machining. Metals, 15(7), 706. https://doi.org/10.3390/met15070706

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop