A Review of the Role of Modeling and Optimization Methods in Machining Ni-Cr Super-Alloys

Abstract

Ni-Cr alloys are some of the most important materials being utilized in the manufacturing industry. Their unique properties make them attractive for various applications, especially in the aerospace and automobile industries. Since machining these materials is challenging due to their properties, it is necessary to understand their machining processes and how to improve them. As a result, time and again, effort has been made to understand and model the machining of Ni-Cr alloys. In this action, different approaches, i.e., neural networks, fuzzy systems, simulations, etc., have been of great help. At the same time, efforts have been made to optimize the machining processes to find how to obtain the best outputs from the processes. Different methods, such as multi-criteria decision-making, meta-heuristic algorithms, desirability functions, etc., have been utilized in this respect. This work aims to prepare an exhaustive review of the methods used for modeling and optimization of the machining of Ni-Cr alloys. It considers five major machining operations and collects data on how these methods or algorithms have been used to improve the machining and to what extent. The use of newer advanced algorithms in manufacturing processes is on the rise, and this manuscript aims to record the methods used, their effectiveness, and their shortcomings. It also provides an insight into the methods and their compatibility. Suggestions for future work are also discussed at the end of this study.

1. Introduction

Modern industries like aerospace, power generation, and chemical processing require materials that endure harsh conditions. Aerospace demands lightweight, durable materials for fuel efficiency and resistance to extreme temperatures. Power generation requires high-temperature and corrosion-resistant materials, while chemical and petrochemical industries need materials that can withstand corrosive substances and maintain structural integrity. Recent advancements in nickel alloy technology, including improved metallurgy and processing methods, have led to the development of new alloys that overcome previous limitations. These new alloys are more reliable, cost-effective, and versatile, expanding their applications across various industries [1]. The primary advantages of nickel-based alloys include their ability to withstand high temperatures, maintain strong mechanical and chemical properties under heat, and exhibit high melting points. They are also highly resistant to corrosion, thermal stress, and wear [2,3]. Nickel and nickel-based alloys provide a higher weight-to-strength ratio than steel. Their strength comes from a nickel matrix with elements like chromium, tungsten, and rhenium. Tantalum boosts high-temperature performance but can be substituted by titanium, lowering resistance to heat and oxidation. Nickel alloy turbine blades can function at temperatures as high as 520 °C [4].

Machining nickel-based superalloys presents significant challenges because they are poor heat conductors, have a tendency to harden during work, high hot hardness, chemical reactivity with tool materials, and the presence of abrasive carbide particles in their structure [2,3,5,6]. These issues force the aerospace industry to adjust traditional tool failure standards (ISO 3685 for turning and ISO 8688 for milling), discarding tools well before they reach the typical wear limits to avoid cutting edge damage and maintain the surface integrity of critical safety components [7]. Nickel-based superalloys are difficult to machine using conventional methods like drilling or turning due to their mechanical properties. To address these challenges, robust non-traditional machining techniques such as EDM, WEDM, USM, and RUM are preferred, though they significantly increase the machining cost [8]. In addition, nickel-based alloys face problems during heat treatment with maintaining dimensional accuracy, requiring tighter process control or extra allowances during the machining process [9]. Machinability is a complex function consisting of various functions. Among them, the cutting tool material is considered the most fundamental factor influencing machinability. However, selecting the appropriate tool and evaluating its machining performance remain significant challenges for manufacturers [10,11,12,13]. Compared to 1080 steel, which is rated at 100% machinability, Inconel 718 has a much lower machinability rating of only 10–12%, highlighting the difficulty of machining nickel-based alloys [14].

Modeling and optimization play a crucial role in machining various alloys. For decades, developing predictive models for machining performance metrics like tool wear, surface roughness, cutting forces, part accuracy, and chip behavior has been a key research focus. The interrelated nature of these metrics, combined with the complex interactions among the tool, chip, and workpiece, has hindered progress in creating effective models. One major issue is carbide cracking, which leads to residual cavities and surface cracks on the machined part. Moreover, studies [15,16] have shown that debris in the form of microchips is highly sensitive to cutting speed, further complicating the machining process of these alloys. Thus, optimization of sensitive process parameters is necessary to enhance the machinability of these alloys. Additionally, the challenge lies in identifying precise predictive parameters to ensure optimal productivity and quality in real-world applications [17]. Numerical models for dry and near-dry machining have become a precise method for analyzing complex phenomena during machining, including dynamic recrystallization and phase transformation [18,19]. Modeling approaches, such as the Johnson–Cook (JC) model [20], are commonly used to predict workpiece responses like flow stress under thermomechanical loads during machining. These are typically implemented using tools like MATLAB, FEA or analytical models [21,22]. In the absence of sophisticated experimental methods like high-speed in situ analysis, FEM models are commonly used to investigate only certain metallurgical material states [17,23]. Traditional machining optimization primarily focuses on cost and productivity objectives. While these are important, a more crucial aspect is optimizing the interaction of various machining performance measures, including tool life, surface roughness, and chip form or breakability [24,25]. Challenges also remain due to limited understanding of process-specific constitutive and friction behavior [21,26], especially in extreme conditions where extrapolation is needed. This affects the accuracy of machining parameter predictions and thus component quality [27]. Additionally, standard calibration methods using mechanical tests are not fully representative of actual machining conditions [28].

Optimization of machining parameters is gaining popularity due to its impact on efficiency, quality, cost, and sustainability [29]. Bi et al. [30] reported up to 67% energy savings in a drilling process through optimization, while Liow [31] found that conventional machines can consume 800 times more energy than micro-milling setups, mainly due to inefficient spindle usage. The literature [32,33,34] has shown that nature-inspired algorithms like GA, PSO, ABC, HS, and ACO are effective in solving various engineering optimization problems, providing optimal or near-optimal solutions.

Few reviews have been reported earlier. Benardos and Vosniakos [35] reviewed methods for predicting surface roughness in machining, focusing on how various optimization techniques enhance production efficiency. Rao and Kalyankar [36] explored advanced optimization methods for machining parameters in electric discharge, abrasive jet, and ultrasonic machining to boost productivity. Razlan Yusoff et al. [37] presented a literature review on optimizing machining parameters to reduce chatter during operations, addressing aspects like spindle design, tool path, and cutting processes. Additionally, Aggarwal and Singh [38] examined the optimization of machining parameters specifically in turning operations to improve productivity.

However, a complete inquiry is needed to reduce the gap between the state-of-the-art modeling optimization and machining of Ni-Cr alloys. By synthesizing advanced modeling and optimization methodologies with practical machining insights, such a review can facilitate the development of innovative strategies that enhance machining performance, improve material utilization, and reduce production costs. Ultimately, this integrated approach can drive advancements in manufacturing Ni-Cr alloys, enabling their broader application across various industries. The scope of this article is presented in Figure 1.

The review is sectioned as follows: firstly, Ni and Ni-Cr alloys—the most-used alloys in the aerospace, chemical, and power-generation industries—are described, along with their composition, properties, and applications; secondly, a broad methodology is outlined that encompasses the primary optimization methods and modeling techniques used in machining: Response Surface Methodology (RSM), Artificial Neural Network (ANN), Adaptive Neuro-Fuzzy Inference Systems (ANFIS), Taguchi signal-to-noise (S/N) ratio, desirability function (DF), Grey Rational Analysis (GRA), Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS), genetic algorithm (GA), Non-dominated Sorting Genetic Algorithm II (NSGA-II). The methods of modeling and optimization are explored in depth, focusing on their application to turning, milling, drilling, grinding, and electrical discharge machining (EDM)/wire electrical discharge machining (WEDM) as discussed in the existing literature. The final section focuses on the experimentation with Ni-Cr alloys and the future of comprehending machining processes, specifically regarding the modeling and optimization of Ni-Cr alloys. This review aims to examine the trends in utilizing modeling and optimization techniques while working with Ni-Cr alloys, addressing aspects such as machinability, roughness, power, temperature, surface integrity, force, tool life, and productivity to align with the demands of contemporary manufacturing. Figure 2 illustrates a structure to make this article more elucidated.

2. Nickel–Chromium Alloys: Classification and Properties

Nickel is a lustrous, white, face-centered cubic crystal structured metal. The density of nickel (Ni) is 8.90 g/cm³ [39]. Nickel is elastically rigid, having a modulus of elasticity of 207 GPa [40]. Because of high strength and corrosion resistance, nickel is a good choice as an alloying element. Nickel and nickel–chromium alloys have ultimate tensile strengths ranges from 317 to 1375 MPa. Nickel–chromium alloys have melting temperatures at the range of 1260–1907 °C, which is more than most of the aluminum and steel alloys. Nickel is not easily corroded by air, water, or alkaline substances [39]. These special properties make Ni-Cr alloys suitable for aerospace and automotive components, high-performance machinery, and medical implants. Nickel-based alloys account for about 80% of superalloys used in aerospace sectors [41].

Nickel–chromium alloys have a broad category based on their composition and properties. Among them, Inconel, Nimonic, and Hastelloy have wider area of applications.

Table 1. Chemical composition of nickel and its alloys in the reviewed articles in Section 4.

Materials	Chemical Composition (Almost) in %															Properties
	Ni	Cr	Fe	Mo	Co	C	Mn	W	Al	Si	S	Cu	Nb	Ti	Ta	M.P. (°C)	T.S. (MPa)
Pure Ni	99.5	-	-	-	-	-	-	-	-	0.2	-	-	-	-	-	1455	317
Inconel 600	73.63	15.68	8.79	-	-	0.054	0.33	-	-	0.24	0.006	0.028	-	-	-	1350	671
Inconel 625	60.99	21.4	4.27	8.96	-	0.039	0.32	-	0.53	0.4	-	-	3.29	0.21	<0.01	1350	860–960
Inconel 718	53.12	17.65	18.63	3.07	-	0.04	-	-	0.6	-	0.03	0.02	4.79	0.86	-	1290–1350	1375
Inconel 800	35	21	Bal.	-	-	0.08	1.2	-	0.6	0.5	0.015	0.7	-	0.5	-	1385	690–790
Hastelloy C-276	Bal.	14.5–15.5	4–7	15–17	2.5	0.01	1	3.5–4.5	-	-	-	-	-	-	-	1323–1371	690
Hastelloy X	Bal.	21.7	19.73	8.44	0.79	-	0.64	0.68	0.07	0.35	-	0.35	-	0.04	-	1260–1355	615
Nimonic 80A	Bal.	19.5	1.5	-	-	-	-	-	1.7	-	-	-	-	2.5	-	1320–1365	1250
Nimonic 90	Bal.	18–21	2	-	15–21	0.13	1	-	1–2	1	-	-	-	2–3	-	1310–1370	1271
															Range	1260–1907	317–1375

shows the chemical compositions of different Ni-Cr alloys.

Inconel 600 is a solid solution Ni-Cr based superalloy. It has good corrosion and oxidation resistance up to 1204 °C, making it suitable for extreme environmental applications such as nuclear plant, marine, and aerospace components [42,43,44,45,46]. It shows brittle behavior at room temperature but ductility at high temperature [47]. The grain size is smaller in this alloy. The Cr-rich carbides along the grain boundaries make this alloy susceptible to creep fracture [42].

Inconel 625 is a popular choice for applications involving extreme heat, including aerospace, petrochemical, marine, and nuclear industries [48]. The potential applications of this alloy include steam turbine casting and heat-exchanger tubes [49,50]. It has an austenitic microstructure [48], and its mechanical properties are anisotropic [51] (vary depending on the direction of measurement). Cr and Mo in this alloy resist corrosion, Fe lowers the cost, and Nb works as solid-strengthener [52].

The most widely used nickel alloy, Inconel 718, accounts for 35% of the yearly volume production of nickel alloys and over 50% of the total weight in aerospace engines [41,53]. Inconel 718 is a popular nickel–iron–chromium alloy that can handle extreme heat and tough conditions [54,55]. Its main strengthening phase is a body-centered tetragonal structure, which is represented as γ* Ni₃Nb. Inconel 718 does not become brittle up to 650 °C. The quantity of delta-phase precipitation that occurs at grain boundaries directly affects Inconel 718′s resistance to creep. Because of the precipitation, Nb is removed from the grain boundaries, increasing the likelihood of failure [56].

The aerospace industry uses Inconel 800 because of its exceptional qualities, which include great chemical and physical strength against high temperatures and good corrosion resistance [57]. It is lighter and more elongated than Inconel 718 [58]. Inconel 800 is a common type of Ni-Fe-Cr alloy used in supercritical water (SCW) applications. Alloy 800 has a fine-grained, austenitic structure that helps to form protective Cr oxides, resisting corrosion. However, high temperature and high contents of salt and oxygen decrease its corrosion resistance notably [59,60].

Ni-Cr-Mo alloys are the most adaptable among Ni alloys, functioning well in environments with both reducing and oxidizing conditions. Hastelloy C-276 is such an alloy that provides resistance to stress corrosion cracking, localized pitting and various acidic environments due to its higher percentage of W, Fe [61], Cr, and Mo [62]. It is a promising material for next-generation nuclear reactors [63]. Recently, it has been used for making superconducting tapes like YBCO and MgB₂ [64].

Even though solid solution strengthened Ni-based superalloys exhibit high ductility, the extra strengthening provided by the as-solidified fine dendritic microstructures may be beneficial for their intermediate strengths and maintaining ductility. Due to the lack of Nb in its composition, Hastelloy X is an excellent applicant for this use [65]. It is a strong alloy typically used in gas turbines to make parts like the combustion chamber, connecting parts, and parts near the exhaust. According to Haynes International, Hastelloy X is easy to fabricate, is highly resistant to stress-corrosion cracking in petrochemical environments, and exhibits good ductility after prolonged exposure at high temperature.

The interaction of Ti with Ni has been proven to be stronger than that with Al [66,67]. Nimonic 80A is a Ni-Cr alloy, in which the addition of Ti and Al increase the strength [68]. The quality of this superalloy depends on various parameters, mostly on the formation of γ’ phase, retaining a high degree of order up to the melting temperature [69,70], which makes this alloy important for aerospace and high-temperature applications. Another Ni-Cr-Co alloy is Nimonic 90, which is popular for its good chemical and corrosion resistance, high creep, and rupture strength. It thus is widely used at high temperature applications in various industries [71,72,73,74].

3. Overview of Optimization and Modeling Techniques

In this section, several modeling and optimization techniques that can be found in the latter section of this paper have been discussed briefly. The section is comprised of two subparts. The first subsection discusses three modeling techniques, and the latter discusses several optimization techniques. While many methods are available other than those mentioned here, only the ones that can be found dominantly in this work have been considered.

3.1. Modeling Methods

In manufacturing works, modeling is usually conducted to correlate various machining parameters with outputs of the machining process. This modeling can be performed with the help of simple mathematical tools, numerical approaches, or advanced algorithms. They are chosen based on their complexity and performance. Though modeling can be conducted for classification and regression analysis of responses, the works on the latter outnumber the former by a very large scale in the context of machining materials. The following are three such methods that can perform regression:

3.1.1. RSM

Response Surface Methodology is a set of mathematical and statistical tools that approaches a complex problem with the help of multiple parameters and their extent of interactions. The extent to which RSM can understand underlying relations between the independent and dependent variables depends on the Design of Experiment or DOE. DOE can be performed in several ways, such as full factorial, Central Composite, D-optimal, etc. DOE also decides the number of experiments to be conducted. After conducting experiments, a regression model is built that can represent the process. Sometimes, the response surface developed by RSM is used for optimization, but the process becomes complicated when the number of independent variables is more than two [75]. More details about the DOEs and other aspects of RSM can be found in the study by Kumar et al. [76].

3.1.2. ANN

Artificial Neural Networks (ANNs) are information-processing models stimulated by biological neurons. They are also called processing elements. ANNs perform through three key components [77]: neuron character, network formula, and learning. The neurons are the building elements of ANNs that have a weight assigned to their input connection. When the weighted sum of a neuron crosses a specific threshold, the neuron is activated, and it produces an output. The neurons are arranged into arrays to build a neural network. The usual layers in ANNs are input, output, and hidden. The number of layers and nodes in layers is important, as it affects the performance. Finally, through learning, ANNs keeps updating the weights and improve their architecture. Elaborate details about these learning processes can be found in the work of Jain et al. [78].

3.1.3. ANFIS

Adaptive Neuro-Fuzzy Inference Systems are based on fuzzy interference networks. They were introduced in 1993 [79], and they work through five layers [80]: fuzzification, rule, normalization, defuzzification, and summation. First, each model node transforms the crisp input values into fuzzy values by applying a membership function to the values. From there, each node determines the firing strength of the rule that it represents. Then, the firing strength of each node is normalized, and the weighted values of rules for each node are calculated. Finally, the output is provided by summing the weighted values. Detailed explanations of ANFIS can be found in the work by Kabini et al. [81].

3.2. Optimization Algorithms

Optimization, in its simplest form, means finding the values of a variable that either maximizes or minimizes an objective function. In the context of manufacturing, optimization is performed to find values of machining parameters that result in expected output responses. Like modeling methods, optimization can also be carried out with simple mathematical tools to complex evolutionary algorithms. A few such methods are discussed as follows:

3.2.1. Taguchi S/N Ratio

The signal-to-noise ratio, or S/N ratio, developed by Taguchi can help determine optimal parameters for a given criteria of output response. In this context, signal refers to desirable output, and noise refers to uncontrollable variations. Although there are over 60 S/N ratios [82], the three most used S/N ratios are the following [83,84]:

• Smaller-the-better: ${\displaystyle \frac{S}{N}}=-10log{\displaystyle \frac{1}{n}}\left(\Sigma y^2\right)$

• Larger-the-better: ${\displaystyle \frac{S}{N}}=-10log{\displaystyle \frac{1}{n}}\left(\Sigma {\displaystyle \frac{1}{y^2}}\right)$

• Nominal-the-best: ${\displaystyle \frac{S}{N}}=10log{\displaystyle \frac{\overline{y}}{{s_y}^2}}$

$y=$ observed value of the response, $\overline{y}=$ average of y $s_y=$ variance of y, $n=$ no. of observations

For all three cases, the higher the S/N ratio, the better the output.

3.2.2. DF

The desirability function is known for its ability to convert multiple variables into one score. There are several types of DFs [85], but the most common is Derringer and Suich. In this method, an overall desirability is obtained for each experiment, and the best levels of each parameter are determined based on the experiment with the highest overall desirability. Overall desirability is obtained through the following formula [86]:

$$d_0=\sqrt[w]{\left({d_1}^{w1} \ast {d_2}^{w2} \ast \dots \dots \dots . \ast {d_n}^{wn}\right)}; w=w1+w2+\dots +wn$$

$d_i=$ desirability index of ith response, $wi=$ weight of ith response.

3.2.3. GRA

Grey Relational Analysis (GRA) is a component of Grey Systems Theory, which was introduced by Julong Deng [87]. GRA can be used to evaluate systems with uncertainty and poor information [88]. GRA works by first determining the Grey Relational Coefficient (GRC) and then obtaining the Grey Relational Grade (GRG) from the GRC for each experiment [89,90]. Upon finding the GRC, it is multiplied by the relative weight, and the GRG is obtained by summing the products of GRC and the weight. After finding the GRG for each experimental run, the run with the highest GRG is declared as the best run.

3.2.4. TOPSIS

Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) is Hwang and Yoon’s multi-criteria decision-making (MCDM) method [91]. TOPSIS aims to select the solution that is closest to the ideal and farthest from the worst. The distance is measured in Euclidean terms [92]. First, the positive ideal response and the negative ideal response are found, and then the distance from the ideal solutions is found for each experiment. Finally, the closeness coefficient (CCO) is calculated using the following formula:

$$C_i={\displaystyle \frac{{D_i}^{-}}{{{D_i}^{-}+D_i}^{+}}}$$

${D_i}^{+}=$ separation from the positive ideal solution; ${D_i}^{-}=$ separation from the negative ideal solution.

The higher the CCO value, the better the experiment. Usually, the experiments are ranked based on the CCO value.

3.2.5. GA

Genetic algorithm is a type of evolutionary algorithm, which itself is a division of metaheuristics algorithms, that is based on concepts of biological selection and evolution [93]. John Holland, the developer of GA, drew inspiration from Darwin’s evolutionary theory to unfold GA. While the algorithm has gone through many changes throughout the years [94], its key elements are selection, crossover, and mutation [95]. In the selection step, two chromosomes are selected from a random population of chromosomes for crossover. These chromosomes are representative of the expected solution to the problem. From the chosen chromosomes, an offspring is produced through crossover that represents a better solution than its parents. As a crossover merges already-existing genetic elements, a mutation is induced in the chromosome to introduce variation. Then, from the mutated offspring, a new population of chromosomes is generated, and the process is repeated until a termination criterion is reached. More about the crossover, mutation methods, and the overall functionality of GA can be found in the work of Katoch et al. [93].

3.2.6. NSGA-II

Non-dominated Sorting Genetic Algorithm II (NSGA-II) is the updated version of the first NSGA. After NSGA was criticized for its incapability to include elitism and complex computation [93], NSGA-II was developed by Deb et al. [96]. NSGA-II can optimize multiple objectives simultaneously on its own and is based on Pareto dominance. Using Pareto dominance, NSGA-II works with the following elements [97].

Non-dominated Sorting: Before starting the sorting, a new population is obtained from crossover and mutation of the initial population, and both of the populations are merged to form a new population on which sorting is conducted. Solutions are sorted into different fronts based on their non-domination levels.

Elite-preserving Operator: Elitism makes sure that non-dominated members of a generation get passed on to the next generation.

Crowding Distance: For a given solution, crowding distance is found by averaging the distances of the two solutions that are on either side of the said solution. The solution that is in a less-crowded region will have a larger crowding distance. This is considered while selecting elements for new generation from the fronts.

Selection Operator: The population for the next generation is selected based on the elitism and the crowd surrounding a certain solution. For population members of different rank, the member with the higher rank is selected for next generation. And for population members of same rank, the member with the higher crowding distance is selected.

These steps are recursively applied until a termination condition is satisfied. The process can be summed up using the flow chart in Figure 3:

4. Modeling and Optimization Approaches in Machining Ni-Cr Alloys

4.1. Turning

Turning is a major conventional machining process used to produce rotational parts by removing material from a workpiece using a single-point cutting tool. The workpiece, generally held in a chuck, rotates at high speed while the single-point cutting tool feeds into it, removing material as chips to form the desired shape. The cutting tools used for turning can be categorized into solid tools and inserts. Tools can be coated or uncoated, which plays a great role in machining difficult-to-machine materials such as Inconel, Hastelloy, etc. In recent years, the use of conventional lathes has declined due to the advancement of CNC machines, which offer greater productivity and precision. Turning remains widely used across various industries, with a significant portion of research on machining nickel-based alloys focusing on this process. Moreover, turning is the primary machining method used in the production of gas turbine disks.

The key independent parameters of turning operation are cutting speed, depth of cut, and feed rate. Almost all of the studies included here use these three parameters and sometimes some other, less-used ones. For machining responses, a large number of responses can be found, with surface roughness taking the lead.

Tebassi et al. [98,99] studied surface roughness in Inconel 718 turning using mathematical models, RSM, and ANN and the study found that ANN outperform RSM in reliably and accurately predicting the non-linear behavior of surface roughness and cutting force models, showing better correlation and lower errors. Jafarian et al. [100] analyzed surface improvement in finish turning using a Genetically Optimized Neural Network System (GONNS) for analysis.

In another study by Jafarian et al. [101], studied how tool wear and machining time affect surface roughness. Intelligent methods like ANN and multi-objective optimization (MOO) were applied for simultaneous process optimization. The results showed that tool wear had minimal impact on surface quality when machining time was under 120 s. Optimal conditions were achieved at a cutting speed of 150 m/min, feed rate of 0.08 mm/rev, and depth of cut of 0.2 mm. The conclusions highlight that neural networks are often strong performers in surface roughness prediction.

Such performance for surface roughness can also be observed by RSM. Frifita et al. [102] studied the effects of feed rate, cutting speed, and tool type on cutting force and surface finish using a carbide tool, identifying nose radius and cutting speed as key factors through RSM and DF. The developed models showed high accuracy, with R² values of 95–98% and a maximum error of 9% in confirmation tests. RSM identified optimal cutting parameters with a desirability score of 0.83. Similarly, Ali et al. [103] optimized cutting parameters under MQL using ANOVA, finding feed rate the most influential for surface roughness, with 84% contribution. A study on the surface roughness of Hastelloy X was undertaken by Oschelski [104]. The study considered three input parameters, i.e., cutting speed, depth of cut, and cooling conditions. The optimized parameters using Box–Behnken Design (BBD) yielded low surface roughness and high MRR. While different optimization methods were used (RSM, ANOVA, BBD, S/N ratios), all achieved high model accuracy and low prediction errors, confirming the effectiveness of parameter tuning.

Satynarayana et al. [105] studied the surface roughness of Inconel 600 using a precision lathe with uncoated carbide tools under dry and MQL conditions. They analyzed the effects of rake angle, cutting speed, and feed rate, using S/N ratios to find optimal settings. Their developed model predicted surface roughness with less than 10% error. Sometimes, work on surface roughness is conducted as a part of a multi-objective problem that also includes evaluation of other responses. Such a work is the two-part manuscript by Pusavec et al. [106,107] on turning using a 890-grade carbide tool. In the first part, a prediction model was built using RSM with inputs like feed rate, cutting speed, depth of cut, and cooling/lubrication (C/L) conditions, predicting outputs such as cutting forces, tool wear, surface roughness, and MRR. The second part used GA to optimize this model, adding chip breakability as an additional input. Optimization was performed with and without considering chip breakability, resulting in slightly higher feed and cutting speeds when it was included.

Buddaraju et al. [108,109] presented works aimed at surface roughness and MRR of Inconel 600. In both manuscripts, the Taguchi method was used for optimization of the process parameters. The results also indicated that feed rate has the highest influence on both of the output responses. Gupta et al. [110] used PSO, TLBO, and DF to optimize turning of Inconel 800. Cutting force, tool wear, surface roughness, and length of tool-chip contact were chosen as the responses. Considering cutting speed, feed rate, and the side cutting edge angle of the tool as the inputs, a combined objective equation was built. Following that, all three approaches were applied to minimize the equation. While PSO achieved better success (90%), TLBO was faster (Figure 4), and they both performed better than the DF approach.

Taking the same inputs, same insert, and somewhat the same parameters into account, Gupta & Sood [111] conducted a study where PSO, Bacterial Foraging Optimization (BFO), and DF were used for optimization. Like the aforementioned work, PSO achieved better success, and in this case, it was also faster. The manuscripts also concluded that MQL conditions are better than dry conditions for machining aerospace materials. These studies reflect a growing trend of using metaheuristic evolutionary algorithms like PSO and GA for multi-objective optimization. This inclination is driven by their capability of flexible objective handling and ease of implementation.

Another work [112] compared Al₂O₃, MoS₂, and graphite-based nanofluids during turning of Inconel 800 using similar tools and parameters as the previous work. Using the Box–Cox RSM approach for modeling and composite desirability approach (CDA) for optimization, graphite nanofluids were found to perform the best.

Durairaj and Gowri [113] studied micro-turning of Inconel 600 with a titanium carbide-coated tool. Two non-linear regression models were built for responses, tool wear, and surface roughness, considering cutting speed, feed rate, and depth of cut as the input parameters. Finally, optimization was performed with GA, whose fitness functions were the functions obtained from said regression models. Optimal results suggested that low cutting speed, low feed rate, and low depth of cut are ideal for micro-machining Inconel 600.

In two studies, Altin [114,115] investigated how feed rate, cutting speed, and tool type affect cutting force and surface roughness during machining of Hastelloy X and Inconel 625 using three tool types, provided in

Table 2. Properties of the cutting tools. Reproduced from [114] with copyright permission from Elsevier, 2013.

Coating Material (Top Layer)	Coating Method and Layers	ISO Grade of Material (Grade)	Geometric Form	Manufacturer and Code
TiN	CVD (TiN, Al2O3, TiCN, TiN, WC)	P25-40, M20-30	CNMG120404RP	Kennametal KC9240
TiN	PVD (TiN, TiCN, TiN, WC)	P25-40, M20-30	CNMG120404FN	Kennametal KT315
WC-CO	Uncoated	P25-40, M20-30	CNMG120404MS	Kennametal K313

. The first study focused on Hastelloy X, while the second compared both materials. Using Taguchi S/N ratio for optimization, they found that minimum cutting force occurs at the same feed rate and tool level but different cutting speeds, while minimum surface roughness occurs at the same cutting speed but varies with feed rate and tool. The studies highlight that optimal conditions differ among different Ni-Cr alloys, especially in terms of cutting speed.

Gowd et al. [116] attempted to build a model that can represent hard turning of Inconel 600. The input parameters were cutting speed, feed rate, and depth of cut; output responses were all three cutting forces and surface finish. The CCD approach of RSM was used to build the model, and its adequacy was measured in various ways, such as ANOVA and multiple regression coefficients. A manuscript by Korkmaz and Günay [117] studied the effect of the same parameters on the same responses during turning of Nimonic 80A. A second-degree regression model was constructed for each response. Percentage deviation between actual and predicted values from models proved the models to be significant. Researchers have also focused on cutting force, tool wear, and surface roughness of Hastelloy C-276 [118]. Modi et al. [119] used DF to optimize turning parameters for Hastelloy C-276, considering feed rate, cutting speed, and depth of cut. Cutting force, tool wear, and surface roughness were the responses. Individual desirability was combined into a composite one for multi-objective optimization, as shown in Figure 5.

These studies indicate that statistical approaches, especially ones such as RSM, are generally capable of modelling cutting forces during turning adequately.

The same efficacy, however, cannot be easily reproduced using fine-element modeling. In their study of high-speed cooling, Liu et al. [120] employed heat transfer and Fluid–Structure Interaction (FSI) simulation to better understand the effects of cooling. While the output of the FSI simulation was accurate enough to validate some assumptions, it lacked precision in the numerical results for cutting force and tool–chip contact length. This discrepancy occurred as the model assumed a specific uncut chip thickness, as opposed to varying thickness in an actual experiment.

DF was also used by Singh et al. [121], who optimized the turning of Hastelloy C-276 using green lubricants like synthetic oil, waste motor oil, and vegetable oil. Using the CCD-RSM approach with four input variables and three output responses, they applied DF for multi-response optimization. Validation showed good agreement between the predicted and actual results, as shown in Figure 6. Synthetic oil gave the lowest temperature, while vegetable oil offered the best surface finish and chip-reduction coefficient.

In another study by the same author [122], sustainability assessment was performed during turning of the same material under dry, flood, and vegetable oil MQL conditions using an uncoated CNMG120408 tool. Using TOPSIS, performance indices were calculated for 27 experiments. With equal response weights, MQL was most sustainable; with varied weights, dry machining ranked highest as indicated in Figure 7.

Prasad et al. [123] optimized cutting parameters, i.e., cutting speed, feed rate, and depth of cut, to minimize tool wear and surface roughness during dry turning of Inconel 800. After examinations with signal-to-noise (S/N) ratio and ANOVA, GRA was used for multi-objective optimization. GRA results showed that the cutting parameters significantly affect the machining on Inconel 800, with cutting speed having the highest percentage contribution (33.3%). Joshi et al. [124] conducted a similar study with different levels of parameters. It also compared two different MQL conditions (150 mL/h and 230 mL/h) with flooded (600 mL/h) and dry conditions and concluded that MQL2 (230 mL/h) was the most favorable condition.

Naidu et al. [125] optimized cutting speed, depth of cut, and feed rate for SR and MRR of Hastelloy C-276. Both dry and wet conditions were considered for the study, and optimal parameters were found for both cases using S/N ratio. The optimal level of parameters was different for both cases, but in both scenarios, feed rate was the most significant parameter.

Dabhade et al. [126] conducted a multi-objective optimization of Inconel 625 during dry turning with a sialon ceramic insert. The goal was to observe how surface roughness and MRR act concerning cutting speed, feed, and depth of cut. Through ANOVA testing, it was detected that feed and depth of cut are the primary factors affecting surface roughness and MRR respectively. RSM’s D-optimal approach was used to minimize surface roughness and maximize MRR. Kosaraju et al. [127] reached somewhat similar results with the Taguchi-GRA approach. Overall, it was consistently found that feed rate and cutting speed are the most significant factors affecting surface roughness and tool wear in machining alloys like Inconel and Hastelloy.

With the objective of finding the optimal levels of cutting parameters, i.e., feed rate, cutting speed, and depth of cut, for low cutting force and good surface roughness, Venkatesan et al. [128] conducted a study using statistical approaches. The experiment focused on the dry turning of Inconel 625 using a PVD-coated carbide tungsten insert. The results showed that the regression models for output responses are highly significant, with predicted values closely matching the experimental results, indicating strong model accuracy for machining performance. An identical study was conducted by Zeqiri et al. [129], with the only exception being that in this case, both heat-treated and untreated materials were considered. Singh et al. [130] used regression analysis and the Taguchi method to study surface roughness and temperature of Hastelloy C-276 during turning, finding minimum standard error between the fitted and observed values for output responses in between 0.0149 and 1.80 units. Thus, using statistical and regression models with optimization methods can successfully predict and improve cutting parameters to obtain better outputs in different nickel-based alloys.

Sonawane et al. [131] investigated the relationships among vibrations, surface roughness, and machining parameters (depth of cut, feed rate, and speed) for Inconel 600. The study used Grey Relational Rating (GRR) to determine optimal settings, achieving a GRR of 0.7949 for the best results.

Penteado et al. [132] used Simulated Annealing (SA) and GA to optimize turning of Nimonic 80A. The corresponding inputs were cutting speed, feed rate, depth of cut, cooling conditions (MQL and flood), and two types of ceramic inserts (TP2500 and CP250). The study used surface roughness and cutting length as output responses, employing agglutination methods and regression models before optimization with both SA and GA. The results revealed that SA performed similarly to GA at a 5% significance level. The desirability function gave satisfactory results compared to the agglutination method for multi-response optimization.

Thakur et al. [133] used S/N ratio for optimizing high-speed turning of Inconel 718. The confirmation tests validated the additive model’s adequacy with 95% confidence. Regression coefficients ranged from 0.93 to 0.99, indicating strong correlation. The optimized cutting parameters improved the output responses, ranging from 145% to 180%.

As the turning process generally produces a lot of heat, the effectiveness of different types of cooling conditions has also been studied. In a detailed study, Eskandari et al. [134] compared turning of Inconel 718 under cryogenic conditions with flooded conditions. The cryogenic condition was created using liquid nitrogen (LN₂), and it was found that while performance under cryogenic conditions is generally lacking, using medium levels of parameters can lead to a performance that is close to flooded conditions. Following that revelation, DF-RSM was applied to find such values of parameters, i.e., feed rate, cutting speed, and depth of cut, that would maximize MRR while minimizing flank wear, cutting forces, and surface roughness.

Another extensive study to compare dry, wet, and MQL turning of Inconel 718 was carried out by Khan et al. [135]. Considering feed rate, cutting speed, depth of cut, and cooling conditions as the input parameters and TWR, SR, specific cutting energy, and carbon emission as the output responses, the study conducted a total of nine experiments based on Taguchi DOE. Finally, to find the best experimental run, TOPSIS was used, which concluded that MQL provides a balance between dry and wet machining. The optimum condition defined by TOPSIS led to a reduction of 23%, 8%, 15%, and 12% in the mentioned outputs, respectively.

Other than the usual machining responses, some peculiar responses, such as plastic deformation, power consumption, etc., have also been pursued. Subhas et al. [136] studied plastic deformation and dimensional instability of Inconel 718 during machining in a lathe machine. Using multiple DF, the work optimized five process parameters, i.e., cutting speed, feed, depth of cut, nose radius, and rake angle, for three ranges of dimensional tolerance band. Individual desirability functions of the responses, i.e., residual stress, tool life, surface finish, and MRR, were found and then combined into one function. Modeling was conducted using ANN during rough turning of Inconel 718 with an uncoated carbide insert [137]. The input parameters were cutting speed and cutting time; the response was flank wear. Mital and Mehta [138] built prediction models for Inconel 718 and many other materials during fine turning and compared their behavior, finding 0.187 standard error of mean for surface finish prediction model of Inconel 718.

Khan and Gupta [139] conducted a study on power consumption during the turning of Inconel 600, analyzing factors like cutting speed, feed rate, depth of cut, and tool texture. They used RSM and TOPSIS for optimization. Before optimization, cutting speed was found to have the most significant impact on power consumption, as shown in Figure 8. The two optimization methods gave different results, with RSM leading to lower power consumption.

Khanna et al. [140] focused on understanding power consumption and surface roughness during ultrasonic-assisted turning (UAT) of Nimonic 90, with inputs like cutting speed, feed rate, depth of cut, and tool frequency. Two methods were used for optimization. One was PSO, and the other was a hybrid of PSO and simplex method. Using both methods, both mono and bi-objective optimization were performed. While various modeling techniques, such as ANN, RSM, and hybrid optimization methods (e.g., PSO-Simplex), were used in different studies, hybrid approaches generally showed faster convergence and better accuracy, with differences observed in the effectiveness of optimization methods like RSM and TOPSIS, shown in Figure 9 and Figure 10.

4.2. Electric Discharge Machining

Electric discharge machining (EDM) is a non-traditional machining process that removes material using electrical sparks without direct contact between the tool and workpiece. It is ideal for machining hard, electrically conductive materials like nickel-based superalloys. The process relies on converting electrical energy into heat, reaching temperatures around 10,000 °C. Dielectric fluids—often water-, oil-, or gas-based—are used to control sparks and remove debris, and adding powders to the fluid can enhance surface finish and material removal rate. Electrode motion, such as rotation or vibration, further improves efficiency. EDM is widely used in aerospace, automotive, mold-making, military, and medical industries for producing complex and precise components.

Electric discharge machining’s independent parameters are vastly different from those that are found for traditional machining, like turning, milling, grinding, etc. The input parameters are usually pulse-on-time, pulse-off-time, servo voltage, discharge current, etc. The machining responses that are most important for industrial purposes, however, stay the same. As a result, most of the works found that fall within the scope of this review are on surface roughness and MRR.

Bhardwaj et al. [141] optimized gap voltage, peak current, and pulse-on time to maximize MRR and minimize surface roughness during EDM of Hastelloy C-276 (HC-276). Machining was carried out with a copper electrode, and de-ionized water was the dielectric fluid. GRA was performed using the values of S/N ratio of the responses. ANOVA testing of GRG showed that pulse-on time is the most crucial parameter, followed by peak current and gap voltage. Ramesh et al. [142] performed drilling with EDM on Hastelloy C-276 and investigated the effect of current, pulse-on time, and pulse-off time on surface roughness and MRR of the material. A main effect plot was utilized to find optimal values of inputs for maximizing MRR and minimizing roughness. Finally, regression equations were formulated for the responses, and the values predicted by the equations were compared with experimental values. The close similarity between the experimental and predicted values indicates that the formulated equations can predict the responses fairly accurately.

Khan et al. [143] optimized the process parameters for similar outputs in Nimonic-90. The study developed second-order quadratic models using RSM to link the inputs and responses. ANOVA testing of the model validated its fitness, with an R² value of 0.9516. Finally, DF was used to find the level of parameters that minimize surface roughness and maximize MRR.

Vishnu et al. [144] used Box–Behnken design (BBD) of RSM for DOE in optimizing MRR, surface roughness (SR), and tool wear rate (TWR). The materials used were Nimonic 80A as the workpiece and copper as the electrode. Upon validating the RSM model with ANOVA, DF was used to optimize the responses. Pulse-on-time was found to be most significant for MRR and current for SR and TWR. Sahu et al. [145] focused on the same area, and utility theory was utilized for optimization. A two-factor, three-level, full factorial design was conducted for DOE, and the experimental run with the highest overall utility was chosen as the optimal run. The study concluded that an increase in energy input (peak current or pulse-on time) caused higher MRR and TWR. This conclusion aligns with the previous work that found pulse-on-time to be a significant factor.

Senkathir et al. [146] used a conductive molybdenum electrode and de-ionized water to machine Nimonic 80A, focusing on parameters such as duty factor, gap voltage, and wire feed rate. They optimized these parameters using GRA and RSM. A response optimizer (Figure 11a) was used to obtain the optimized responses (Figure 11b). The results showed that the duty factor had the greatest impact on MRR, surface roughness, and tool wear rate, while feed rate also played a significant role. In contrast, gap voltage negatively affected MRR.

Liu et al. [147] based their work on MRR, kerf width, and surface roughness of Inconel 718 and considered servo voltage, pulse-on time, pulse-off time, and wire tension as the independent factors. Taguchi Data Envelopment Analysis-based Ranking (DEAR) was used with the help of the L27 OA. Using the DEAR approach, a Multi-Response-Performance-Index (MRPI) value was obtained for each experimental run. From this index, finally, an optimal set of parameters was found. The optimal values were validated experimentally, and it was seen that the results derived by only 1.1%.

Baig and Venkaiah [148] focused on minimizing kerf width (width of cut) during machining of HC-276. In this study, the inputs were pulse-on time, pulse-off time, discharge current, and servo voltage. After using S/N ratio for mono-objective optimization, GRA was applied for multi-objective optimization. ANOVA test outed discharge current as the predominant variable for MRR and kerf width individually and also together. Darji and Pillai [149] used Taguchi’s S/N ratio to optimize polarity, peak current, rotational speed, and pulse-on time for minimizing MRR. The predicted and experimental data are compared in

Table 3. Predicated optimal value and confirmation experimental result. Reproduced from [149] with copyright permission from IJMERR, 2012.

	Predicted	Experimented
Optimal parameter combination	Negative, 150 rpm, 2 A, 9 µs	Negative, 150 rpm, 2 A, 9 µs
Significant parameter	Negative, 150 rpm, 2 A	Negative, 150 rpm, 2 A
MRR (mm3/min)	0.24517	0.29887

, showing a good closeness between the data.

Paul et al. [150] optimized responses, i.e., MRR and surface roughness against inputs, i.e., pulsed current, pulse-on time, and pulse-off time during machining of Inconel 800. MOORA was used with the integration of Principal Component Analysis (PCA) for optimization. Error in prediction for the conventional MOORA model and MOORA-PCA model was compared, and MOORA-PCA had a better average error of 4.552% against 6.707% of the conventional MOORA model. Chambhare et al. [151] optimized the process parameters in powder mixed wire cut EDM (PMWEDM) of Nimonic-90 using both single and multi-objective (GRA) methods. Kerf width, MRR, and surface roughness were calculated according to the DOE. Nano powder of titanium carbide was used along with the dielectric. Both the optimization methods identified powder concentration at 4% as beneficial, indicating consistency across methods.

These studies show that in modeling (and sometimes optimizing) surface roughness, MRR, kerf width, and some other parameters during EDM, statistical approaches are often preferred. These approaches range from simple regression equations and ANOVA models to complex MCDM methods. And these methods perform well enough to justify their application in understanding complex machining responses. However, this does not imply that advanced machine learning algorithms are ignored in modeling of these parameters.

WEDM of Hastelloy X was studied by Parmar et al. [152]. The study considered pulse-on time, pulse-off time, wire feed, and servo voltage as the inputs; MRR and surface roughness were the target outputs. A model was formulated using ANN and the comparison plot for the model is displayed in Figure 12 and Figure 13. The MAPE for the MRR and SR model was 6.371% and 5.92%, respectively.

Considering surface roughness as the output variable, a study by Singh et al. [153] compared the predictability of various evolution algorithms such as support vector machine (SVM), Gaussian Process (GP), and ANN. For both GP and SVM, the same three kernels, Puk, Poly, and RBF, were used. Among the total seven distinct algorithms, the GP-Puk model showed the lowest RMSE value and dominated other models during training. During testing, however, the GP-Poly model superseded the others. ANN performed moderately well in both cases.

Hewidy & Salem [154] conducted a multi-objective optimization study on machining Inconel 718, focusing on volumetric material removal rate (VMRR) and surface roughness (SR). They used ANFIS, ANN, and RSM for prediction modeling and the Pareto search algorithm for optimization. The total percentage errors of the models are as follows: ANFIS: 1.539% in VMRR, 1.069% in Ra; ANN: 0.8738% in VMRR, 2.111% in Ra; RSM: 6.387% in VMRR, 4.301% in Ra. These results show that both ANN and ANFIS models are more satisfactory than those of RSM. The work of Lalwani et al. [155] is similar to the preceding manuscript. The study considered pulse-in-time, pulse-off-time, servo voltage, peak current, and wire tension as the input parameters and optimized kerf width, surface roughness, and MRR. Prediction models were developed using both RSM and ANN, and the confirmation experiments concluded that ANN was a better option, due its lower MSE (1.49%), than RSM (5.71%). Finally, the NSGA-II method was used for optimizing the responses.

Natarajan et al. [156] investigated the WEDM of HC-276, focusing on parameters like current, pulse-on time, and pulse-off time, with responses such as MRR, surface roughness, dimensional deviation, and tolerance errors. The study used GRA for multi-response optimization and developed an ANN model to predict GRG from the GRC of input variables (Figure 14). The predicted GRG values closely matched the calculated ones, with an RMSE of 0.0086.

These studies prove that neural networks models are as adept in handling surface roughness data for non-traditional machining as they are adept in handling traditional machining such as turning. The effectiveness of evolutionary algorithms, especially genetic algorithms, in optimizing surface roughness for non-traditional operations can also be confirmed in the same way from the following studies.

Kumar et al. [72] machined Nimonic-90 and optimized the inputs for improving SR using GA. A zinc-coated brass wire and distilled water were utilized as the electrode and dielectric fluid, respectively. The study proposed a Central Composite Design of RSM for modeling the input parameters. GA found the minimum surface roughness to be 0.97717 µm, which was 0.1% greater than the experimental value. Selvam and Kumar [157] investigated surface roughness and kerf width of HC-276. The input variables were pulse-on time, pulse-off time, wire feed rate, current, and voltage. Two regression models were built using RSM, and the equations were the focus of optimization with GA. GA optimized the responses separately, and, in both cases, it reached its solution within 100 iterations, proving its efficiency and speed. Another such example is the use of GA to optimize MRR and kerf width while micro-wire electric discharge machining (Micro-WEDM) Inconel 718 with zinc-coated brass wire [158]. The optimal values provided by GA was validated using experimental data, and the error for MRR and kerf width were 0.35% and 1.4%.

Biswas et al. [159] built a Multi-Objective, Multilayer Neural Network (MOMLNN) model with the assistance of GA and PSO and used that to study machining of Inconel 718 along with Inconel 625. The study focused on MRR, kerf width, and surface roughness, along with recast layer thickness. The prediction results showed that the GA integrated model had a better performance than the PSO integrated model.

Other than evolutionary algorithms, which are stochastic, deterministic algorithms are also sometimes utilized to optimize machining responses. Shukla and Priyadarshini [160] used existing data of wire cut-EDM of HC-276 from the literature to showcase the gradient descent method’s optimization ability. Two equations for the responses, i.e., surface roughness and kerf width, were developed using the normal equation algorithm, and gradient descent was successfully used for multi-objective optimization.

Other than the usual responses, cutting rate and forces are sometimes also included in multi-objective optimization. In a study by Majumder et al. [161], the PCA-GRA approach was administered to optimize cutting time and surface roughness of Inconel 800 during WEDM with a brass electrode. After normalization of the responses, relational grade was calculated for each test experiment, and the optimal combination of parameters was found. Error comparison between PCA-GRA and GRA models revealed that PCA-GRA has a better average error.

Nayak & Mahapatra [162] conducted an experiment on taper cutting of Inconel 718 using WEDM, focusing on parameters like part thickness, pulse duration, taper angle, discharge current, wire tension, and wire speed. They combined the responses (cutting speed, surface roughness, and angular error) into a single performance metric using the Maximum Deviation Theory and optimized it with Taguchi’s method. They then developed an ANN prediction model and optimized it using the Bat algorithm, another swarm-based evolutionary algorithm. The results showed that the combining ANN with Bat provide higher accuracy in terms of the composite score.

A response that has received somewhat less exposure is the TWR. Reviewing works focused on TWR reveals that statistical method is typically used to model and optimize this response, like it is for surface roughness, MRR, and kerf width. Kumar J and Reddy [163] optimized discharge current, pulse-on time, pulse-off time, and electrode material in EDM of HC-276 to enhance MRR and reduce TWR. They tested copper, cryogenic-treated copper, brass, and phosphor bronze electrodes. ANOVA revealed that electrode material was the most influential factor. Taguchi and MOORA methods identified cryogenic-treated copper as the best electrode for both single- and multi-objective optimizations. Dutta and Sarma [164] applied the BBD method of RSM to model µ-EDM of Hastelloy C-276, using capacitance, gap voltage, and pulse-on time as inputs and diametral overcut (DOC), MRR, and tool wear rate (TWR) as outputs. They developed quadratic models for each response and analyzed the impact of inputs on them. Multi-objective optimization was carried out using two methods: the DF of RSM, to maximize MRR and minimize the other responses, and a multi-objective genetic algorithm (MOGA). The results from both methods were compared in Figure 15. It is evident that while Taguchi and RSM are deterministic and suitable for simpler problems, metaheuristic algorithms, such as GA, PSO, and Bat, are better suited for complex, non-linear, and multi-response problems.

As WEDM is an unconventional machining process, there are some peculiar machining responses that are only found in its manuscripts. Several such responses are DOC, recast layer thickness, radial overcut, etc. Faithful observations of these responses have been carried out with both MCDMs and machine learning algorithms. The effect of capacitance, gap voltage, and powder concentration on MRR and DOC was studied by Dutta and Sharma. [165]. µ-EDM of HC 276 was conducted with graphene nano powder. The S/N ratio was used for single-response optimization, and GRA was used for multi-response optimization, which concluded that along with the higher gap voltage, a high value of capacitance and powder concentration results in maximum MRR and minimal diametral overcut.

Alhodaib et al. [166] worked on Nimonic 90, with kerosene as the dielectric and silicon as the powder. GRA was used for optimizing the process parameters for better surface roughness and recast layer thickness. Experimental validation showed that the optimum parameters decreased surface roughness and recast layer thickness by 50.04% and 25.81%, respectively.

Durairaj et al. [167] used GRA in the optimization of Wire-Cut EDM of Inconel 800, further confirming its effectiveness as a multi-objective optimization method for machining processes. A novel study by Roopak et al. [168] united EDM with a magnetic field environment, called magnetic field-assisted EDM (MFEDM). The study also compared between EDM and MFEDM in terms of MRR, SR, and radial overcut and optimized the process parameters using NSGA-II and TOPSIS. The validation results showed acceptable error margins between 5 and 10%, supporting the precision of the methods.

Dry-EDM, an eco-friendly approach to conventional EDM, was investigated by Cortés-Mendoza et al. [169]. The study worked with both Inconel 625 and Titanium Grade 2 and considered pulse-on-time, current, voltage, and gas pressure as the input parameters. The study also defined a Palatnik index, a constant referring to the machinability of a material by thermal erosion processes, for both materials and found that including the index as an input improves the prediction of MRR, electrode wear, electrode velocity, and SR. Prediction models used by the manuscript were LR, RF, SVR, and ANN, with ANN outperforming the other models.

Another unconventional machining, die-sinking EDM, has been the focus of Paul et al. [170]. Machining was performed on Inconel 800, and the CCD approach of RSM was used, which showed that CCD is a powerful tool for experimental diagrams and statistical modeling. ANOVA testing proved the model to be statistically significant, with an R-Sq value of 97.9%. Die-sinking EDM of HC–276 was investigated by Basha et al. [171]. Input parameters considered were pulse-on time, pulse-off time, discharge current, and type of electrode, and S/N ratio was utilized to find the best levels of parameters for maximizing MRR and minimizing surface roughness. Confirmation testing revealed that the difference between the optimum and expected values is not large, validating the approach’s accuracy.

4.3. Milling

Milling is a highly versatile machining process that uses a rotating, multi-edge cutting tool to remove material from a workpiece, enabling the creation of prismatic, polyhedral, or free-form shapes through programmed feed movements in almost any direction. Unlike other machining operations like drilling or turning, milling allows for greater precision and complexity, especially with the advent of CNC-controlled five-axis milling centers that can handle intricate geometries at high feed rates. Initially limited to flat surface generation until the 1960s, milling now plays a crucial role in manufacturing complex components such as plastic injection molds and aerospace parts. In modern aircraft, large monolithic structures like ribs, stringers, and bulkheads are milled from solid blocks, with up to 90–95% of the material removed to achieve lightweight yet strong components, which are then joined with milled aircraft skins to optimize the strength-to-weight ratio.

With milling, the focus returns to conventional independent parameters, such as cutting speed, feed rate, depth of cut, width of cut, number of flutes on cutting tool, etc. Among the machining responses, surface roughness during milling is just as important as it was for the previous two machining processes.

Zahoor et al. [172] explored the roughness of Inconel 718 during CNC milling. The study considered three parameters: feed rate, cutting speed, and axial depth of cut; it ran the data through three optimization models, i.e., PSO, GA, and DF. The results showed that PSO provided a better performance than the other two, with only 0.05% error and 3.19 min of processing time. Ozcelik et al. [173] conducted a similar study to minimize surface roughness during end milling of Inconel 718. The study implemented an ANN-GA approach to first model the dataset and then optimize it. Two kinds of optimization were carried out—one without any constraint on the values of maximum surface roughness and minimum MRR and one with constraints. The performance of GA was satisfactory in both cases, with the study concluding that a neural network coupled with GA is an effective approach to studying surface roughness.

Surface roughness, cutting force, cutting temperature, and tool wear of Inconel 718 were the focus of the study conducted by Rubaiee et al. [174]. A combined objective (CO) function was developed, and the optimization was performed by Teaching–Learning-Based Optimization (TLBO) and Non-dominated Sorting Genetic Algorithm II (NSGA-II). Both methods concluded similar levels of parameters as optimal (

Table 4. Optimal parameter settings. Reproduced from [174] with copyright permission from Elsevier, 2022.

Parameters	Vc (m/min)	f (mm/rev)	LM	CO
NSGA-II	90	0.05	4	0.960
TLBO	90	0.05	4	0.962
Experimental	90	0.05	4	0.97

). This proves that both methods are equally robust in optimizing. Zhou et al. [175] carried out optimization of surface roughness and compressive residual stress of Inconel 718. The manuscript used an integration of GRA, RBF, and PSO, which proved to be better than traditional GRA. This was ensured by conducting experiments using optimal parameters found from both approaches and realizing that the parameters from the integrated model resulted in a lower surface roughness. These studies reiterate the efficacy of evolutionary algorithms in optimizing surface roughness during traditional machining.

Other than evolutionary machine learning models, statistical approaches have also been made to model and optimize the surface roughness of material during milling operations. Motorcu et al. [176] optimized surface roughness during dry milling of Inconel 718, considering cutting speed, milling direction, coating layer, and number of inserts of the tool holder as the factors. The Taguchi method was used to optimize the parameter levels that minimize the surface roughness. The study concluded that down milling results in a lower surface roughness. Kar et al. [177] attempted to optimize axial cutting force, surface roughness, and MRR during CNC milling of Inconel 718. First, the multi-objective problem was turned into a single-objective problem with the help of DF, and then a fuzzy system was used for optimization. The results of the study were conclusive enough to comment that DF–fuzzy logic is a feasible approach in understanding roughness.

A study by Maiyar et al. [178] optimized surface roughness and MRR of Inconel 718 during end milling in a CNC machine. Optimization was performed using GRA, and it was found that the optimal inputs resulted in a 64.8% increase in MRR and a 9.52% decrease in surface roughness. A study from R. Bangar et al. [179] focused on the SR and MRR of ball nose end milling on Inconel 600, preheated at 650 °C. The Taguchi optimization was used in this study to find the optimal process parameters: spindle speed, depth of cut, feed rate, and number of flutes. The results showed that even though both higher spindle speed and higher depth of cut lead to lower roughness and higher MRR, lower feed rate ensures lower surface roughness, and higher feed rate contributes to higher MRR.

While both machine learning algorithms and statistical approaches are common for understanding surface roughness, in the case of studying properties of the tool, such as wear, ML takes the lead. Singh et al. [57] investigated tool wear optimization in milling Inconel 718 using an AlTiN + TiN coated tool under dry, flood, and MQL conditions. The study used BFO and PSO for optimization and found that PSO outperformed the former, a conclusion that was seen during tool wear prediction during turning as well. The study also concluded that MQL had the best cooling performance.

Tool wear prediction during milling of Hastelloy C276 was investigated by Sen et al. [180]. The machining was carried out using PVD TiAlN-coated carbide inserts, and an alumina nanofluid-enhanced MQL system was used to ascertain its effectiveness in reducing tool wear. To predict tool wear, Deep Neural Network (DNN), XGBoost, and SVR were used. The comparison of the models showed that XGBoost outperformed the other models by a long margin. The study also noted that despite the obvious advantages of these AI-driven methods, variability in machining conditions still poses a danger to the accuracy of the models.

Kaya et al. [181] proposed a tool condition monitoring method for Inconel 718 machining. The study found that cutting parameters did not significantly affect tool life, while torque and z-axis force were more sensitive to flank wear. An ANN model was built to reliably predict tool wear. A study by Sen et al. [182] predicted flank wear for MQL-milling of Inconel 690. Gene Expression Programming (GEP) and ANN were used for prediction. Comparing the methods based on R², RMSE, MAPE, Theil uncertainty, and Kullback–Leibler divergence revealed GEP as the superior.

These studies indicate that whether it is evolutionary algorithms, decision trees, or neural networks, they all perform satisfactorily concerning tool properties. Tool properties inherently involve a more complex relationship with the machining parameters, something that advanced algorithms can learn easily. Statistical approaches, however, are often pursued for modeling cutting force and energy. Shahir Kasim et al. [183] constructed a model using the BBD approach of RSM to correlate cutting parameters, i.e., cutting speed, feed rate, depth, and width of cut, with cutting force while end milling Inconel 718. ANOVA testing found the model to be satisfactory with 0.987 R² and 0.984 adjusted R² value. The prediction error of the model was below 3%. Sen et al. [184] aimed to find out the ideal MQL condition during milling of Inconel 690. The parameters were cutting speed, feed rate, depth of cut, and flow rate. RSM was applied to find the correlation between the parameters and the responses, i.e., surface roughness, cutting temperature, and cutting force. The prediction error in this case was less than 1%. Finally, NSGA-II and TOPSIS were used for optimizing the machining responses.

Hsiao et al. [185] also used an RSM model to corelate machining parameters with specific cutting energy, cutting power, surface roughness, and MRR during nanoparticle-suspended lubrication milling of Inconel 800. Authenticity of this RSM model was assured as the R² value for all the outputs was greater than 0.95. Then, NSGA-II was applied to minimize energy and roughness for one case and to minimize power and maximize MRR for another. The optimal results from the Pareto front were validated, confirming the reliability of the RSM-NSGA-II approach. From these studies, a conclusion can be drawn that RSM can easily model cutting forces and energies with high accuracy, which leads to an effective optimization process. It is also possible to attain similar accuracies as RSM with alternative approaches. An RBF-NSGA-II approach was used by Vu et al. [186] during flat-end milling of Inconel 800 in a three-axis milling machine. The R² values obtained from the RBF model were 0.93 and 0.94 for cutting force and specific cutting force, respectively. The study aimed to maximize MRR while minimizing cutting force and specific cutting forces. It also explored the relations between the parameters and responses and found that cutting velocity and depth of cut had the greatest impact on cutting force (Figure 16).

Another machining response that has been focused on time and again during milling is the cutting temperature. V. Anandan et al. [187] optimized machining of Nimonic 80A using a modified Gower TOPSIS method, considering surface roughness and temperature under dry, wet, and cryogenic MQL conditions. Cryogenic MQL was found to be the most effective, as all nine experiments with CMQL were ranked as the first nine runs by TOPSIS. The study also concluded that environment and feed rate are the key factors.

Kumar et al. [188] attempted to optimize the workpiece temperature of Inconel 625 during 2.5-D milling. A model was set up using Box–Behnken design (BBD). ANOVA was used for a significance test, and GA was used for optimizing temperature. The results revealed feed rate as the most crucial factor in controlling work-piece temperature, and that their relationship is positive. However, the result regarding to optimization was not satisfactory, as the experimental cutting temperatures were found to be greater than the optimized temperature during confirmation. The author suggests that this could be overcome with a better prediction model.

Another study focused on cutting temperature during milling of Inconel 625 was carried out by Karthik and Rao [189]. It was an extensive study that also included cutting force. First, an FE model was developed and validated against experimental data. Upon finding the FE model satisfactory, it was used to generate data to compare the performance of linear regression, GPR, Multilayer Perception, decision tree, SVR, ELM, RF, XGBoost, Stochastic Gradient Boosting, and AdaBoost. The study discovered that GPR, RF, XGBoost, Stochastic Gradient Boosting, and AdaBoost perform close to each other in predicting cutting temperature, with AdaBoost taking the lead. AdaBoost can do so by handling noise and variability more effectively than the others. In predicting cutting force, GPR, SVR, and AdaBoost perform similarly, with GPR taking the lead.

Feng et al. [190] also established a finite element prediction model for temperature variation in Inconel 718. However, this work is vastly different from the aforementioned work, as the authors are more focused on determining whether considering microstructure evolution helps to improve the prediction model. The proposed model was checked with both numerical and experimental methods, and it was found that the new model is indeed better.

4.4. Drilling

Drilling is a fundamental hole-making machining process that involves the use of a rotating drilling tool, typically secured in a machine spindle or lathe tailstock, while the workpiece is clamped securely using a vice or chuck. The operation relies on two primary motions, rotation and feed, which can be applied to either the tool or the workpiece, depending on the machine setup. As one of the most common and economically significant machining methods, drilling is often a final step in the fabrication of mechanical components. It has broad industrial applications, including automotive, aerospace, and injection molding industries, especially in producing deep holes for parts like crankshafts, fuel injectors, and cooling passages. In aerospace manufacturing, where precision and material integrity are critical, improper drilling of hard-to-machine materials can lead to surface anomalies and microstructural damage, potentially causing fatigue cracks and component failure. Hence, the work dedicated to drilling is mostly focused on the responses related to the hole.

Venkatesan et al. [191] studied the best size and shape of a hole drilled in Inconel 625 metal. The hole’s diameter, circularity error, overcut, taper ratio, cylindricity, and damage factor were analyzed and optimized using the ANOVA, Taguchi S/N ratio, and composite desirability approaches, achieving maximum error of less than 10% during the confirmation test. Another study by Venkatesan et al. [192] investigated and optimized the impact of micro-drilling parameters (tool diameter, spindle speed, and feed rate) on the quality of holes in Inconel 800 using similar optimization methods. The results reflected that the geometrical accuracy of the machined hole was primarily influenced by feed rate, followed by tool diameter and spindle speed.

Bronis et al. [193] aimed at a similar objective while drilling Inconel 718. The study used three kinematic systems of machining named KIN I, KIN II, and KIN III. Multifunctional statistical analysis was performed using Taguchi L27 orthogonal arrays. According to the ANOVA and simulation results, the third kinematic system was found to be the best option.

Gundrilling is a method used to drill very deep, precise holes. Ahmed et al. [194] used a novel Euler–Bernoulli beam theory-based dynamic behavior model for a gun drill to analyze the potency of different factors during drilling holes in Inconel 718. They also studied the tool wear and thrust forces during both conventional gundrilling (CG) and EDM-Gundrilling (EDMG). It showed that EDMG produces less wear on the drill bit than CG and also had lower straightness deviation, given in Figure 17 and Figure 18. The study also concluded that the dynamic behavior model is a good candidate in determining optimum parameters for drilling operations.

These studies indicate that there is a tendency to use basic statistical models for studying drilling. This is understandable, as these studies focus more on finding out how an independent parameter affects the overall drill properties and less on optimizing the parameters. There is an enormous scope for using different types of advanced algorithms in modeling and optimizing this process. Such studies would both ascertain the algorithm’s capability in understanding the drilling process and improve the machining process of drilling.

A few studies based on finite modeling, however, have been observed for drilling. A study by Wang et al. [195] compared three step drill designs with traditional twist drills to study how the second point angle affects thrust and torque, modeled using mechanical and finite element methods, at various spindle speeds. The study compared among the experimental, simulation, and predicted results of average thrust force and average torque for stable drilling, finding acceptable errors.

Residual stresses significantly impact machining and performance of Ni-Cr alloys as they influence fatigue life, dimensional stability, tool life, and machine efficiency. Lu et al. [196] used finite element simulation to predict residual stresses during micro-drilling of Inconel 718 and optimized cutting parameters. Their four-factor and three-level orthogonal simulation showed average relative errors of 12.20% (X-direction) and 16.96% (Z-direction). These findings suggest that while finite element simulations can provide valuable predictive insights, their reliability sometimes depends on factors such as mesh size and mesh ratio.

4.5. Grinding

Grinding is an abrasive machining process that uses a grinding wheel or abrasive belt as the cutting tool to remove material from a workpiece. Turbine blades in aircraft jet engines are typically made from nickel-based alloys due to their exceptional high-temperature strength and strong resistance to heat and corrosion. However, these same properties make them particularly challenging to grind, often resulting in high grinding forces, rapid wheel wear, and poor surface quality. Consequently, the modeling and optimization of grinding parameters have become increasingly important and have been the focus of extensive research in recent years.

Dawood et al. [197] predicted and optimized the input parameters while grinding Inconel 718. The article utilized Taguchi and ANOVA for optimization to achieve minimum surface roughness. Another study on the same material by Gupta et al. [198] optimized dry grinding parameters using RSM coupled with GA and PSO. ANOVA testing showed that wheel speed affects surface roughness the most, followed by depth of cut and table speed. It was also found that the combination of RSM and PSO yielded superior outcomes compared to RSM and GA. Though results by both methods indicated the same value of depth of cut and wheel speed for minimum roughness, they varied in the values of table speed. The roughness obtained by PSO was 0.2586 µm, as opposed to 0.2735 µm by GA.

Along with statistical approaches, fuzzy networks have also been utilized to predict surface roughness during grinding. Unune et al. [199] described a fuzzy logic artificial intelligence method to predict the MRR and SR of Nimonic 80A during abrasive-mixed electro-discharge diamond surface grinding (AMEDDSG). The model developed was based on Mamdani logic, and the centroid of area was used as the defuzzification method. Both of these approaches were used based on their wide acceptance. The model was validated using new experimental runs, and the accuracy was found to be 93.89%. Regression analysis, neural network, and ANFIS methods were used by Varma et al. [200] to predict the MRR and surface roughness of Inconel 800 during the cylindrical grinding process. Among the methods, ANFIS has the lowest error rate among others and predicts MRR and surface roughness with 91.15% and 90.96% accuracy, respectively. From these studies, it seems that surface roughness during the grinding operation is better estimated by fuzzy algorithms. However, more evidence is required before a generalization can be made in this regard.

Other than surface roughness and MRR, residual stress, grinding force, grinding temperature, etc. have also been the focus of researchers. Wang et al. [201] looked into the results of robotic belt grinding on Inconel 718. The article looked at how the grinding process affected the residual stress, surface roughness, and shape of the material. In this study, a new method for optimizing grinding parameters using a linear weighting function has also been suggested. It was found that, if the grinding force was very high (above 240 kPa), increasing the belt speed also increased the residual stress.

Das et al. [202] optimized grinding parameters using GRA to minimize the tangential force and average surface roughness while grinding Inconel 718. The researchers compared three different grinding methods: ultrasonic-assisted grinding (UAG), UAG with minimum quantity lubrication (MQL), and conventional grinding. The findings indicated that the depth of cut has the biggest impact on cutting force, while amplitude significantly affects both cutting force and surface finish. Also, the combination of UAG + MQL resulted in better output parameters than others. Singh et al. [203] used UAG with Ultrasonically Atomized Green Cutting Fluid (UaG-UaF) method while machining Nimonic 80A to optimize both grinding and vibration factors to improve surface roughness and minimize grinding forces. The study used a statistical approach of RSM-ANOVA for both modeling and optimization of the responses. The optimum parameters were validated experimentally, and the error was 7.67%, 9.98%, and 16.17% for normal force, tangential force, and surface roughness, respectively.

Esmaeili et al. [204] conducted a study aimed at reducing cutting force and temperature while simultaneously increasing material removal rate. An FE model was made to simulate the interaction between a single CBN grit and Inconel 718. This model was compared to real-world experiments and theoretical predictions using statistical analysis. The experimental results suggested that the addition of graphene nanoplatelets to a vegetable-based MQL system helped reduce the force and energy required for grinding compared to traditional dry, fluid, and MQL grinding methods. Sinha et al. [205] focused on the same area along with surface roughness, specific energy, and coefficient of friction. The study used random forest and Gaussian Process to create regression models and used MCDM methods, including VIKTOR, TOPSIS, and entropy weighting method (EWM), to identify the optimal conditions for grinding Inconel 615 using MQL. It was found that GPR consistently produced better results than RFR, with an R² value of over 0.95, because of its trial-and-error approach.

5. Experimentation with Ni-Cr Alloys

Nickel-based high-temperature alloys make up about 30% of the materials used to build an aircraft engine [206]. They are also used in space shuttle engines, cryogenic tanks, and pressure vessels for nickel–hydrogen batteries on space stations [207,208,209,210,211,212,213]. As mentioned, these alloys are very difficult to machine because of their tough properties. Research shows that they tend to react strongly with cutting tools, often causing issues like galling or the welding of chips to the tool [207,211,214]. Still, with the right setup, operations like turning, milling, drilling, and grinding can be performed on them. However, some modifications, including the usage of CBN, PVD, and PCD tools, coolants have been made to facilitate the machining [215,216]. Nickel-based alloys are tough to machine, but the right cutting tool geometry can help. Tools with larger noses and/or included angles tend to last longer because they are stronger and have more contact with the chip. In addition, ceramic tools, while brittle, can still be effective if used with techniques such as taper turning [217].

Most machining of these alloys in aerospace engine manufacturing is conducted at low cutting speeds and feed rates using carbide tools. This is mainly to extend tool life and to better manage the heat and stress produced during machining [54,211]. But carbide tools have poor hot-hardness, which makes them less effective for high-speed machining (HSM). Because of this, researchers have recently focused more on applying HSM techniques.

Along with works dedicated to finding newer and improved machining conditions, effort has been spent on making traditional methods more sophisticated. This requires a thorough understanding of the existing methods. Hence, studies focused on modeling and optimizing traditional methods have risen. While advanced cutting tools can ensure high speed and better heat resistance [218], understanding the traditional methods and their caveats will make the foundation of these processes stronger.

At high cutting speeds, the main role of the coolant is to cool the tool and workpiece, since there is not enough time for the fluid to reach the chip–tool interface or the wear land to act as a lubricant. Most of the studies found that the use of coolant results in a better surface finish, especially using cryogenic of MQL cooling condition in conventional machining of nickel-based superalloys [219,220]. High-pressure coolant achieves better chip removal than conventional coolants, producing small-segmented chips. However, flank wear and notching are the most dominant failure modes in high-pressure cooling conditions, which in turn, give shorter tool time than conventional cooling in most of the machining operations [221]. A survey revealed that the cost of coolant is more than three times higher than that of cutting tools. In addition to the high expense, the use of coolant can also pose health risks to operators. High-pressure coolant also does not show any advantage over conventional cooling in the case of machining nickel-based alloys, as ceramic tools can perform better under dry conditions. Given these concerns, dry machining presents itself as a more appealing option—both economically and environmentally [53,222].

Table 5. An overview of recent and pertinent studies concerning the modeling and optimization of Ni-Cr machining operations.

Author	Machining Process	Machining Material	Process Parameters	Machining Responses	Modeling Methods	Optimization Methods	Comments
Pusavec et al. [106,107]	Turning	Inconel 718	V, S, ap, C/L condition	F, TWR, SR, MRR	RSM	GA	RSM is effective for modeling and optimization but relies on predefined equations, limiting its ability to capture complex nonlinear behaviors. There is a lack of comparison with advanced methods, and most studies overlook factors like energy use, tool wear cost, and sustainability.
Gupta et al. [112]	Turning	Inconel 800	V, S, ap, C/L condition	SR, F, TW	RSM	CDA
Dabhade et al. [126]	Turning	Inconel 625	V, S, ap	SR, MRR	RSM	RSM
Singh et al. [121]	Turning	Hastelloy C-276	V, S, ap, C/L condition	SR, Tc, ξ	CCD-RSM	DF
Gowd et al. [116]	Hard turning	Inconel 600	V, S, ap	Fx, Fy, Fz, SR	CCD-RSM	-
Tebassi et al. [99]	Turning	Inconel 718	V, S, ap	SR, F	ANN, RSM	-	Few studies on turning processes use machine learning or multiple modeling approaches to assess predictive accuracy. Future research can address this gap by exploring and comparing such models.
Jafarian et al. [101]	Turning	Inconel 718	V, S, ap, machining time	SR	ANN	NSGA-II
Durairaj and Gowri [113]	Micro-turning	Inconel 600	V, S, ap	TW, SR	Statistical regression	GA
Penteado et al. [132]	Turning	Nimonic 80A	V, S, ap, C/L condition, inserts	SR, cutting length	-	SA, GA
Venkatesan et al. [128]	Turning	Inconel 625	V, S, ap	F, SR	-	S/N ratio	Few studies explore process parameters beyond cutting speed, feed rate, and depth of cut, indicating a common trend of focusing only on these. Additionally, more research should emphasize the use of advanced optimization algorithms.
Thakur et al. [133]	High-speed turning	Inconel 718	V, S, lubricant quality, pulse frequency, delivery pressure, direction of fluid	F, Tc, flank wear	-	Taguchi
Kumar et al. [72]	WEDM	Nimonic 90	Ip, Ton, Toff, SV	SR	CCD-RSM	GA	RSM-GA approach provides satisfactory results for SR and KW. Other designs of RSM can be checked in this regard. Use of NSGA-II instead of GA might produce better results.
Selvam and Kumar [158]	WEDM	Hastelloy C-276	Ton, Toff, Ip, Vf, WT, P	Kw, SR	RSM	GA
Khan et al. [143]	WEDM	Nimonic 90	SV, WT, pulse duration	MRR, SR	RSM	DF	Again, RSM results in good performance. However, neural networks or decision trees may prove to be better. Also, MCDMs like GRA, TOPSIS, etc. can be implemented.
Dutta and Sarma [165]	µ-EDM	Hastelloy C-276	Ton, capacitance, gap voltage	Diametral overcut, MRR, TWR	RSM	DF
					-	MOGA
Lalwani et al. [155]	WEDM	Inconel 718	Ip, Ton, Toff, SV, WT	Kw, SR, MRR	RSM, ANN	NSGA-II	ANN outperformed RSM and complemented NSGA-II better.
Nayak & Mahapatra [163]	Taper cutting WEDM	Cryo-treated Inconel 718	Part thickness, pulse duration, taper angle, discharge current, wire tension, wire speed	Cutting speed, SR, angular error	-	Taguchi	Bat and Pareto search are relatively unconventional approaches that can be explored further. Comparison with other methods may produce valuable results.
					ANN	Bat algorithm
Hewidy & Salem [154]	WEDM	Inconel 718	Ton, Toff, SV, Vf, P	VMRR, SR	ANFIS, ANN, RSM	Pareto search algorithm
Singh et al. [153]	WEDM	Nimonic 90	Ton, Toff, Ip, SV	Surface roughness	SVM, GP, ANN	-	This work demonstrated how kernel choice can affect performance.
Senkathir et al. [146]	WEDM	Nimonic 80A	Duty factor, gap voltage, Vf	MRR, SR, TWR	-	GRA	As dimensionality reduction improves the performance of MOORA, it can be investigated if PCA improves the other MCDMs as well.
Paul et al. [150]	Die-sinking EDM	Inconel 800	Ip, Ton, Toff	MRR, SR	-	MOORA, MOORA-PCA
Sen et al. [184]	Milling	Inconel 690	V, S, ap, flow rate of MQL	Maximum flank wear	GEP, ANN	FIS	ANN and GEP have proven effective in modeling complex, nonlinear, and multi-variable relationships without the need for predefined equations, making them well-suited for addressing the complexities of advanced machining processes.
Ozcelic et al. [174]	CNC milling	Inconel 718	V, S, ap, ae (radial)	SR, MRR	GEP, ANN	FIS
Kar et al. [178]	CNC milling	Inconel 718	V, S, ap	SR, F, MRR	ANN	GA
Zahoor et al. [173]	CNC Milling	Inconel 718	V, S, ap	SR	RSM	PSO, GA, DF	The trend shows consistent use of RSM with optimizers like PSO, GA, and NSGA-II, reflecting its reliability.
Kumar et al. [190]	CNC milling	Inconel 625	V, S, ap, step-over, cutter diameter	Temperature	RSM	GA
Sen et al. [186]	CNC milling	Inconel 690	V, S, ap, MQL flow rate	SR	RSM	NSGA-II, TOPSIS
Motorcu et al. [177]	Milling	Inconel 718	V, milling direction, coating layer, number of inserts	SR, TW	-	Taguchi	MCDM optimization techniques have proven effective in milling. Future research could explore even better results by first using advanced modeling methods and then applying MCDM for optimization.
Kar et al. [178]	CNC milling	Inconel 718	V, S, lubricating medium	F, TW, SR, Tc	Regression analysis	NSGA-II, TLBO
Rubaiee et al. [175]	milling	Nimonic 80A	V, S, cutting environment	SR, Tc	-	Gower TOPSIS
Ahmed et al. [196]	Gundrilling, EDM-Gundrilling	Inconel 718	V, coolant pressure	Hole-straightness, TW, Fy	Euler–Bernoulli beam model	-	Drilling studies mostly focus on hole quality, with limited use of MCDM and advanced modeling like data-driven, physics-based, or hybrid approaches. Future work could explore responses like tool wear, thrust force, and torque.
Lu et al. [198]	Micro-drilling	Inconel 718	V, S, ap, ae	Residual compressive stress	FE	Weight coefficient method
Bronis et el. [195]	Drilling	Inconel 718	V, S, type of kinematic system	Hole quality	RSM	-
Venkatesan et al. [193]	Micro-drilling	Inconel 625	V, S, ap	Hole’s diameter, circularity error, overcut, taper ratio, cylindricity, damage factor	-	Taguchi, DFA
Dawood et al. [199]	Cylindrical grinding	Inconel 718	Grit size, ap, coolants	SR	-	Taguchi S/N ratio	Statistical methods generally perform well in understanding grinding process. Other MCDMs can be tested and compared. Also, optimization using machine learning algorithms can be carried out, and results can be compared to find the best approach.
Singh et al. [205]	Ultrasonic-assisted grinding	Nimonic 80A	V, ap, intensity of vibration, air pressure, SOD	SR, grinding force	RSM, FEA	Desirability test
Sinha et al. [207]	CNC grinding	Inconel 625	V, table speed, ap	Fz, SR, specific energy, apparent coefficient of friction	RF, GPR	TOPSIS, VIKOR
Unune et al. [201]	Abrasive-mixed electro-discharge diamond surface grinding	Nimonic 80A	V, abrasive concentration, pulse current, pulse-on-time	MRR, SR	Fuzzy logic	-	Fuzzy logic is a good predictor of roughness. More comparison is needed to determine whether it outperforms ANN in all cases. Also, its prediction capability needs studying.
Verma et al. [202]	Cylindrical grinding	Inconel 800	V, S, ap	MRR, SR	Regression, ANN, ANFIS	-

provides a summary of the manuscripts focused on this area. The following are the symbols used in the table and what they represent:

Process parameters: cutting speed = V, feed rate = S, depth of cut (axial) = a_p, depth of cut (radial) = a_e, cooling/lubricant condition = C/L, pulse-on-time = T_on, pulse-off-time = T_off, discharge current = I_p, servo voltage = SV.

Machining responses: cutting force = F (feed = F_x, thrust = F_y, tangential = F_z), surface roughness = SR, material removal rate = MRR, volumetric material removal rate = VMRR, tool wear = TW, tool wear rate = TWR, cutting temperature = T_c, chip-reduction coefficient = ξ, wire feed rate = Vf, wire tension = WT, kerf width = K_w, flushing pressure = P.

6. Challenges and Future Scope

Over the years, Ni-Cr alloys have been used extensively in works focusing on modeling and optimization, maybe even more so than other alloys and materials. The challenges and future scope observed by our study are as follows:

• In turning nickel-based alloys, notching and flank wear are two of the most dominant failure modes. Work hardening of the alloys is another problem to consider. In this regard, coolant is useful, and challenges are faced while trading off between cost and performance to obtain sustainable machining processes.
• Characterization of surface texture is very complicated, largely depending on the machining process. Thus, analytical methods are less effective than AI-based, simulation-based, and hybrid methods in predicting surface texture.
• Several studies have explored machining aspects like vibration, cutting length, chip reduction coefficient, and temperature. However, power consumption is a promising area for deeper analysis. Input factors such as machining time, cooling conditions, and rake angle can offer fresh insights into turning Ni-Cr alloys.
• Studies consisting of EDM and WEDM highlighted challenges like limited accuracy in statistical methods. While hybrid approaches can give better results, challenges appear in parameter tuning, overfitting, cost and time investments, and the requirement for domain expertise to validate models.
• In milling, studies of tool properties, such as tool wear, tool wear rate, tool monitoring, etc., have mostly utilized advanced machine learning algorithms. While these methods have proved adequate in such studies, efforts can be made to judge how simple statistical and MCDM methods fare in these situations. If statistical methods can offer similar performance, complex algorithms would not have to be pursued for simple cases with few data.
• In the case of cutting forces and cutting energy, however, the ML algorithms are rare. The parameters are mostly modeled using the RSM approach and optimized using NSGA-II. As NSGA-II provides generally fair optimization based on RSM models, it is quite possible that the performance can be enhanced by using more sophisticated prediction models. Most notably, ANN and XGBoost should be used, as NSGA-II usually provides great performance with them.
• As for cutting temperature, this specific machining response has not been delved into properly in milling. The few available studies range from MCDM methods to ML algorithms to FE modeling. Though these studies provide favorable outcomes, more research is required on each type of model to ascertain its efficacy. RSM could be a great starting point to focus on, as this statistical approach has repeatedly proven its performance in understanding cutting temperature during turning and WEDM.
• While more attention is given to improving cutting temperature scholarships, studies of other responses, such as machining time, cutting length, cutting vibrations, residual stress, tool life, etc., should also be given a push, as these parameters also affect the milling of a material significantly.
• While robust models have shown significant growth in modeling and optimization methods, there is a negligible number of studies that focus on implementing these models in drilling. Future studies could apply advanced algorithms and compare their performance with existing algorithms.
• Multifunctional drilling analysis and micro-milling research are vital for the aerospace industry. However, issues like material buildup and high strain hardening make micro-milling challenging. Studying forces, torque, burrs, chips, and temperature is crucial, as optimizing these factors can drive manufacturing improvements and guide future research.
• Grinding, just like drilling, also received the shorter end of attention in these manuscripts. Of the little attention the grinding operation received, most of it was focused on surface roughness during grinding. This is not surprising, as one of the main goals of grinding operation is obtaining better surface quality. The surface roughness of the material during grinding has been modeled using ANOVA, RSM, fuzzy logic, etc. Given that no advanced robust algorithms have been utilized to model this roughness, there is an extensive scope of new research to be conducted. All sorts of algorithms, such as neural networks, decision trees, probabilistic models, etc., can be used to understand the roughness during grinding better.
• In optimization of this roughness, however, use of evolutionary algorithms has been noticed, a phenomenon common with that of turning, WEDM, and milling, though the algorithms have been limited in standard GA and PSO. Other evolutionary algorithms and other types of algorithms can be put to use in this regard. Several MCDMs can also be used if the available data is too small for machine learning algorithms.
• When it comes to studies on the various other machining responses during grinding, the exposure is very low. Whether it is grinding force, grinding temperature, or residual stress, they have been studied very few times. Other responses like grinding temperature, dimensional accuracy, chip morphology, wheel wear, etc. have almost been completely ignored. These responses have to be brought to attention as optimizing their values would help mitigate some industrial challenges, especially the reduction of wheel wear.
• During this study, the modeling methods that have been most encountered are RSM and ANN. While RSM does provide an efficient statistical way to model processes, an effort could be made to popularize the use of other statistical and numerical methods, such as SVM, KNN, FEM, etc. Again, there are neural network models other than ANN, such as OrthoANN, GONNS, CNN, LSTMs, etc. Another approach that can be popularized is tree-based algorithms such as XGBoost, CatBoost, LightGBM, and random forest. While these methods have seen some exposure, they should be focused on more, as they have outstripped the traditional performance methods.
• For optimization, the Taguchi method has been encountered most often, but other methods such as GA and GRA are also common. As GRA has already seen some traction in these works, there is room for other MCDMs such as MOORA, AHP, TOPSIS, COPRAS, etc. There is also scope for more extensive use of other evolutionary algorithms, such as PSO, TLBO, PSA, NSGA-II, etc. There is also a relatively fresh branch of algorithms called nature-inspired algorithms. Three such algorithms are Bacterial Foraging Optimization, Cuckoo Search Algorithm, and Flower Pollination Algorithm. As these algorithms have not been tested enough in manufacturing, there is potential for assessing these algorithms against commonly used ones.
• Finally, a key problem in using machine learning algorithms has always been the small quantity of available data. While, in this study, we have seen methods capable of producing great accuracy with a small dataset, a question remains whether the performance would have been greater with a larger dataset. The question can be answered through data augmentation. Augmentation methods such as random transformations, generative models, pattern mixing, decomposition, etc. can be utilized in any of these machining methods, and whether that augmentation provides anything useful can be determined.

7. Conclusions

• For modeling and optimization of Ni-Cr alloys, surface roughness and MRR are the most-used output responses. Such practice is reasonable considering that Ni-Cr alloys are a relatively new topic to be studied using advanced algorithms. So, understanding its roughness and MRR during machining takes precedence over other responses such as cutting force, cutting temperature, tool wear, power consumption, etc. However, these different responses have been coming into focus gradually, especially cutting force and tool wear. Some machining-specific outputs, such as kerf width and recast layer thickness for WEDM, are also being focused on to understand the material’s properties during said machining. And it is only a natural progression to focus on these new responses as the studies of the common responses are close to being exhausted. There is still a place for surface roughness and MRR with the new materials, whose research potential is yet to be scratched.
• As mentioned earlier, the most common modeling methods for Ni-Cr alloys are RSM and ANN. In our analysis, it has been noticed that the prediction accuracy or error of the said models does not stray very far from that of the less-used methods. It signifies that the reason behind the popularity of said models is not exactly their modeling and prediction skills but rather that they are simply popular methods used for understanding machining processes. Of course, their reputation as being easy to implement also helps their cause.
• In optimizing machining processes, several multi-criteria decision-making (MCDM) methods have become very useful. Most MCDMs assign a value to each experimental run, and the best levels of parameters are decided from the experimental run with the best score. The advantage of such methods is that they can themselves be used as optimization methods, or the grades provided by them can be used as output in some other optimizing algorithm, resulting in a hybrid optimization.
• Within our work’s scope, the material used most is Inconel 718. Inconel 718, the Ni-Cr alloy with the most applications, leads the aerospace and nuclear industries. While specific properties of other such alloys have made them useful in a few places, Inconel 718 will not be replaced by those materials anytime soon. And so, 718 will always be prioritized over other materials in any study.
• The amount of research based on turning, electrical discharge machining, and milling is somewhat similar. However, works focused on drilling and grinding are drastically low. Even though drilling and grinding are gaining more attention than before, the huge gap between the numbers proves that the former three processes are more important in real-life applications.
• Works exploring unconventional machining processes, such as ultrasonic-assisted turning (UAT), selective laser melting (SLM), micro-wire electric discharge machining (µ-EDM), and die-sinking EDM, are also very rare.
• The cooling or lubrication (C/L) conditions of machining are rarely taken into consideration in the works reviewed in this manuscript. Including C/L conditions in the input presents a unique conundrum, as they are categorical data. And to use them as variables for a regression model, a workaround has to be found that can represent the variables without hindering the modeling. They are often used as dummy variables to solve the problem.