In-Situ Monitoring and Process Control in Material Extrusion Additive Manufacturing: A Comprehensive Review

Abstract

Material extrusion additive manufacturing (MEAM) has emerged as a versatile and widely adopted 3D printing technology due to its cost-effectiveness and ability to process a diverse range of materials. However, achieving consistent part quality and repeatability remains a challenge, mainly due to variations in process parameters and material behavior during fabrication. In-situ monitoring and advanced process control systems have been increasingly integrated into MEAM to address these issues, enabling real-time detection of defects, optimization of printing conditions, reliability of fabricated parts, and enhanced control over mechanical properties. This review examines the state-of-the-art in-situ monitoring techniques, including thermal imaging, vibrational sensing, rheological monitoring, printhead positioning, acoustic sensing, image recognition, and optical scanning, and their integration with process control strategies, such as closed-loop feedback systems and machine learning algorithms. Key challenges, including sensor accuracy, data processing complexity, and scalability, are discussed alongside recent advancements and their implications for industrial applications. By synthesizing current research, this work highlights the critical role of in-situ monitoring and process control in advancing the reliability and precision of MEAM, paving the way for its broader adoption in high-performance manufacturing.

1. Introduction

Additive manufacturing (AM), commonly referred to as 3D printing, employs computer-aided design to fabricate objects through sequential layer deposition [1,2,3]. The ASTM F42.91 Additive Manufacturing Technology aims to advance industry knowledge, stimulate research, and facilitate the adoption of technology [4,5]. The Wohlers Report 2025 indicates that the global additive manufacturing (AM) industry expanded by 9.1%, achieving a valuation of USD 21.9 billion [6]. Among additive manufacturing (AM) technologies, Material Extrusion Additive Manufacturing (MEAM), specifically Fused Deposition Modeling (FDM) or Fused Filament Fabrication (FFF), stands out as the most popular due to its cost-effectiveness, user-friendliness, and wide range of commercially available materials [7]. In the MEAM process, a stepper motor-driven extruder feeds thermoplastic filament into a heated nozzle, melting and depositing it as thin cross-sectional layers onto a build platform, iteratively constructing the final object layer by layer [8,9,10,11,12]. Acrylonitrile butadiene styrene (ABS), polylactic acid (PLA), polyethylene terephthalate (PET), Nylon, Thermoplastic Polyurethane (TPU), polyvinyl alcohol (PVA), Woodfill, Bronzefill, Copperfill, Polyethylene terephthalate glycol (PETG) and polypropylene (PP) are the most common MEAM materials, with their distinct melting points primarily determining the mechanical and thermal properties of the resulting printed objects [13,14,15,16,17,18].

In-situ monitoring techniques, applied across all seven ASTM F42-defined additive manufacturing (AM) classes, provide real-time insights into the dynamic processes occurring during fabrication. By tracking print quality, detecting defects, minimizing material waste, and facilitating corrective actions, these methods enhance control and reliability throughout the AM build process [19,20,21,22,23,24,25,26,27,28]. MEAM has been employed across multiple industries, including aerospace [29,30,31,32,33,34,35], healthcare [36,37,38,39,40,41,42,43,44,45,46], automotive [47,48,49,50], construction [51,52,53,54], education [55,56,57,58,59,60], food [61,62,63,64], fashion [65,66,67], biomedical [68,69,70,71].

Nevertheless, components fabricated via material extrusion, when juxtaposed with those generated by alternative additive manufacturing methodologies, frequently demonstrate diminished reliability and an elevated propensity for failure [72]. These challenges arise from a multitude of factors, including nozzle clogging, filament runout, print jams, skipped steps, z-banding, over-extrusion, under-extrusion, inconsistent infill density and direction, micro-pores in the filament, fluctuations in print speed, and variations in print temperature, all of which collectively compromise the quality and consistency of the final printed components [73,74,75,76,77,78,79,80,81,82,83,84,85,86,87]. Consequently, it is imperative to diagnose and comprehend the operational status of a 3D printer to enable prompt modifications or substitution of degraded components, thereby ensuring consistent performance and part quality [88,89,90,91,92]. In the context of material extrusion additive manufacturing, in-situ monitoring entails the real-time observation and measurement of critical process parameters, including temperature, extrusion rate, layer deposition, and nozzle performance, through the utilization of integrated sensors, facilitating the immediate identification of defects, anomalies, or machine health issues during the printing process, and by enabling real-time adjustments and predictive maintenance, in-situ monitoring improves quality control, reduces post-production inspection, and minimizes material waste and processing time [93,94,95,96,97,98,99].

This paper aims to comprehensively review the existing research on in-situ monitoring of the MEAM process, evaluate the effects of these sensor-based monitoring systems, and discuss present and prospective research trajectories.

2. In-Situ Monitoring Techniques

In MEAM, researchers have utilized various sensors to assess process performance and ensure print quality throughout the AM process. The sensors enable continuous monitoring, support predictive maintenance, minimize defects, and ultimately lead to the more reliable fabrication of 3D-printed parts. These sensors include accelerometers, magnetometers, gyroscopes, optical cameras, rheometers, encoders, thermal cameras, piezoelectric sensors, temperature sensors, current sensors, laser scanners, acoustic emission sensors, etc. In the sections that follow (Section 2.1, Section 2.2, Section 2.3, Section 2.4, Section 2.5, Section 2.6 and Section 2.7), a summary of different sensor-based monitoring systems, experimental setups, and use cases is presented.

2.1. Rheological Monitoring

Colon et al. [100] introduced an experimental setup (Figure 1) tailored for material extrusion additive manufacturing, featuring a custom hot end equipped to measure multiple parameters simultaneously. Specifically, the apparatus recorded extruder motor torque (the rotational force driving the filament), filament infeed load (the force exerted on the filament as it entered the hot end), melt pressure (the pressure of the molten filament within the hot end), and melt temperature (the temperature of the melted filament). This integrated system enabled a detailed analysis of the pressure dependency chain, which referred to the interconnected relationships among these parameters and their influence on internal pressure dynamics. The study examined how various monitoring techniques, including sensor selection, influenced pressure estimates and how adjustments to process parameters altered these relationships. Additionally, an infrared (IR) array incorporated into the setup provided real-time tracking of the melt temperature, revealing how changes in process variables, such as temperature or feed rate, affected the thermal conditions of the melt. The findings underscore the critical role of comprehensive, in-situ monitoring in understanding and optimizing the extrusion process, offering valuable insights for improving control and consistency in additive manufacturing.

Anderegg et al. [101] established an innovative approach for evaluating temperature and pressure within the polymer flow field inside the hot end of a Fused Filament Fabrication (FFF) printer. A system was engineered to quantify temperature at multiple nozzle locations and assess pressure drop, minimizing flow field interference. Measurement was conducted using a piezoresistive pressure transducer, which detects pressure through changes in electrical resistance, and a custom thermocouple insertion device, ensuring precise temperature data. Internal conditions of FFF nozzles were successfully analyzed using these tools. The investigation focused on variations in conditions resulting from changes in filament feed rate and nozzle temperature within a fixed system geometry, yielding quantitative insights into the effects of process parameters on printing conditions.

Sousa et al. [102] developed an advanced configuration integrating a micro-counter-rotating twin-screw extruder, an in-line slit rheometer, a filament die, a miniaturized material feeder, and related downstream components for analyzing polymer behavior during material extrusion. The design of the in-line slit micro-rheometer minimized the delay between measurement initiation and result acquisition, while avoiding additional thermal or flow cycles—common in conventional rheological sample preparation—that might modify the initial material characteristics. Incorporation of small piezoelectric sensors, previously validated for accurately measuring steady melt pressures in polymer extrusion, enhanced the setup. Comparison of rheological data obtained in-line versus off-line facilitated direct evaluation of material properties under real-time processing versus traditional approaches.

Some limitations and gaps in rheological monitoring include [103,104]:

• Complexity of Integration into FDM Systems: Implementing rheological in-situ monitoring requires sophisticated equipment, such as pressure transducers, thermocouples, and modified nozzles, which can be challenging to integrate into standard FDM printers.
• Challenges in Data Interpretation: Rheological properties, such as viscoelastic moduli and complex viscosity, are intricate and influenced by factors like temperature, shear rate, and material composition.
• Difficulties with Real-Time Monitoring: FDM is a dynamic process characterized by rapid changes in temperature, pressure, and flow rates as the material transitions from a molten state in the nozzle to a solidified form on the print bed. Capturing accurate rheological data in real time under these conditions is technically challenging.
• Limited Understanding of Material Behavior: Viscoelastic properties are crucial for predicting printability; however, many materials used in FDM lack comprehensive rheological characterization.
• Inadequate Consideration of Environmental Factor: Environmental conditions, such as ambient temperature and humidity, can significantly affect the rheological properties of materials during printing.

2.2. Vibrational Pattern Monitoring

Isiani et al. [105] established a methodology utilizing three ADXL335 accelerometers to evaluate, track, and quantify the operational status of a Lulzbot Mini 3D printer through vibrational signal collection under varied printing conditions (as shown in Figure 2). Configuration featured ADXL335 units—3-axis accelerometers delivering signal-conditioned voltage outputs, measuring acceleration within a ±3 g range, and exhibiting 270 mV/g sensitivity—positioned at the nozzle, frame, and print bed, interfaced via a National Instruments NI 9202 module to LabVIEW for time-domain signal acquisition. Post-processing, analysis, and visualization of the captured signals were performed in MATLAB 2022. The initial assessment relied on the Fast Fourier Transform (FFT) for signal analysis, although constraints emerged in achieving thorough fault classification. The integration of FFT, spectrogram, Principal Component Analysis (PCA), and Support Vector Machine (SVM) forms an advanced approach, enabling detailed signal evaluation across various printing conditions and accurate fault identification and classification.

Zhao et al. [106] integrated a data-driven approach with monitoring and diagnostic processes for 3D printing, as reported. Optimization of model parameters occurred during training based on the generalization error, calculated separately from the model outputs, which increased computational effort. A random forest-based data-driven method emerged to assess and distinguish normal and filament jam states in FDM. Adjustment of model parameters relied on generalization error from random forest training, offering a straightforward yet effective training strategy. The enhancement of quality, efficiency, and performance in 3D printing has demonstrated significant application potential.

Tlegenov et al. [107] formulated a theoretical framework capturing the Fused Filament Fabrication (FFF) process, focusing on nozzle condition dynamics, particularly clogging susceptibility. Development encompassed both theoretical simulations and experimental validations to replicate nozzle clogging, detectable through vibration sensor outputs. A physics-based model emerged, delineating interactions between material flow, nozzle state, and operational variables, providing a structured representation of clogging mechanisms. Implementation of a monitoring strategy utilized vibration sensors to track nozzle condition in real time, leveraging frequency and amplitude shifts in vibrational signals to identify blockages. Validation involved conducting controlled experiments that simulated partial and complete clogs, correlating sensor data with physical nozzle states to establish a reliable detection method. The application of this approach provided a practical tool for preempting print failures, thereby enhancing process reliability in FFF by addressing a critical failure mode with precision and clarity.

Some limitations and gaps in vibrational pattern monitoring include [108]:

• Specificity to Certain Defects: Vibration monitoring excels at detecting mechanical issues, such as motor malfunctions or structural vibrations; however, it is less effective for other common FDM problems, including filament jams or poor layer adhesion.
• Insufficient Understanding of Material and Parameter Effects: The influence of printing materials (e.g., flexible vs. rigid filaments) and parameters (e.g., print speed, layer height) on vibration patterns is poorly understood. These factors can significantly alter vibrational signatures, yet few studies have systematically explored their effects.
• Limited Integration with Other Monitoring Techniques: Combining vibration monitoring with other in-situ methods, such as acoustic emission, thermal imaging, or visual inspection, could provide a more comprehensive view of the printing process.

2.3. Acoustic Sensing

Waheed et al. [109] implemented a configuration deploying multiple acoustic sensor arrays across a 3D printer to improve mechanical fault detection. Positioning occurred at critical sites—motors, gears, and belts—enabling capture of acoustic emissions, defined as sound waves produced during operation. Integration into a broader system incorporated accelerometers, vibration sensors, thermal cameras, and optical cameras for comprehensive printer health monitoring during printing. A combination of diverse sensor types facilitated a detailed assessment of operational performance, with each sensor type addressing distinct functional aspects.

Wu et al. [110] established real-time monitoring techniques to identify printing failures in Fused Deposition Modeling (FDM), employing acoustic emission (AE) as the primary sensing method. Utilization of AE capitalized on its capacity for delivering detailed data, affordability, non-invasive integration, and rapid processing suitable for continuous observation. Recording and analysis of AE signals transpired throughout the fabrication process via an AE data acquisition (DAQ) system, marking the initial application of this approach to FDM failure detection. The attachment of the AE sensor occurred on the hot bed’s upper surface using vacuum grease, ensuring stable signal capture, with sampling executed at a rate of 5 million samples per second. Processing through an onboard digital signal processing (DSP) module within the DAQ system enabled the extraction of critical AE signal characteristics across both time and frequency domains, facilitating the identification of anomalies such as layer detachment or material feed disruptions. Validation confirmed AE’s efficacy in pinpointing failure events, thereby enhancing FDM reliability by offering a robust and cost-effective monitoring solution adaptable to varying print conditions. Furthermore [111], the Acquisition and processing of acoustic emission (AE) signals initially transpired as discrete AE hits, followed by the extraction of multiple characteristics across time and frequency domains. The application of machine learning techniques, encompassing clustering and classification algorithms, formalized the diagnostic framework for identifying distinct failure modes. The selection of an unsupervised machine learning approach was made due to a limited prior understanding of process failures and their corresponding AE signatures, thereby minimizing subjective influence in pattern recognition. Analysis revealed key signal traits—such as amplitude, duration, and spectral peaks—linked to specific failure types, enhancing diagnostic precision. Implementation enabled the differentiation of failure mechanisms, such as material jams or layer defects, without reliance on predefined failure models, offering a robust, data-driven method adaptable to evolving AM process conditions. Validation confirmed the efficacy of mapping AE patterns to operational states, thereby advancing failure detection capabilities in real-time monitoring contexts.

Wasmer et al. [112] executed an exploratory investigation assessing in-situ, real-time monitoring feasibility for additive manufacturing (AM) processes, integrating fibre Bragg grating (FBG) sensors for acoustic detection and reinforced learning (RL) for data interpretation. Data acquisition occurred during an authentic fabrication sequence on a Concept M2 industrial system, utilizing a fiber laser (wavelength 1071 nm, spot diameter 90 μm, M2 value 1.02) in continuous operation mode, configured with a laser power of 125 W, a hatching distance of 0.105 mm, and a layer thickness of 0.03 mm. Variation in scanning velocities—spanning low, moderate, and high rates—yielded workpieces exhibiting three distinct quality tiers: substandard, intermediate, and superior. Implementation of the RL-based algorithm followed training, enabling quality classification with confidence levels ranging from 74% to 82%. Analysis of acoustic signals captured by FBG sensors revealed correlations between sound wave patterns and fabrication quality, providing a non-invasive, real-time method for differentiating process outcomes.

Li et al. [113] formulated an analytical framework (in Figure 3) to examine raster bead formation and interaction within the Fused Filament Fabrication (FFF) process, while also probing acoustic emission (AE) wave origins and their linkage to process characteristics and quality outcomes. The establishment of a physical model elucidated the effects of infill rate, printing speed, and material temperature on raster bead deformation and mutual compression, revealing the mechanics governing bead morphology. The deployment of high-sensitivity acoustic emission (AE) technology enabled the real-time tracking of deformation and compression dynamics under diverse printing conditions, capturing subtle acoustic signals associated with material behavior. Assessment extended to evaluating raster shape continuity across the printing sequence, clarifying how incremental changes influenced final product quality and performance through cumulative effects. The utilization of non-intrusive, responsive AE monitoring provided a viable strategy for online oversight and regulation of the FFF process, linking audible wave patterns to structural integrity. Validation of the model against experimental data highlighted the interconnections between process parameters, bead interactions, and quality metrics, thereby advancing our understanding of FFF optimization pathways.

Forster et al. [114] determined that in-line monitoring during the Fused Filament Fabrication (FFF) process yielded data categorized into two primary domains: evaluation of the 3D printer’s operational status and examination of the fabricated product’s attributes. Development of an advanced in-line monitoring approach focused on assessing product quality through multiple non-destructive evaluation (NDE) techniques, with results calibrated against Micro-computerised Tomography (Micro-CT) observations. The application of NDE methods—encompassing ultrasonic testing, thermography, and acoustic analysis—enabled the detection of internal defects, layer adhesion inconsistencies, and surface irregularities in real-time. Calibration with Micro-CT provided a high-resolution reference, ensuring accuracy in identifying structural anomalies. The extension of this methodology demonstrated adaptability to both composite materials and pure polymers FFF builds, broadening its utility. Validation confirmed capability to enhance quality control by delivering precise, transferable insights into product integrity across diverse material systems.

Yang et al. [115] explored monitoring of composite failure in additive manufacturing through analysis of acoustic emission (AE) peak amplitude distribution skewness, revealing a correlation with filament failure events. Observation indicated that AE skewness shifted with filament breakage, prompting an investigation into detection methods. Industrial approaches employed optoelectronic or mechanical contact switches to identify filament presence within a guide hole; however, these struggled to distinguish between static and moving filament states, often failing to detect breakage when residual filament remained after fracture. An examination of a 2016 MakerBot Industries innovation—an extruder with an encoder wheel—demonstrated the measurement of filament movement increments to mitigate under-extrusion and false detection. However, the complexity of integration limited its applicability to most commercial systems. Investigation shifted to fundamental breakage mechanisms, utilizing an AE sensor for its sensitivity to weak, transient signals indicative of filament rupture. Validation extended AE utility beyond breakage to encompass detection of clogging, material imperfections, and part quality deviations, establishing a versatile, non-invasive monitoring framework adaptable to diverse FFF process anomalies.

Some limitations and gaps in AE monitoring include [19,88]:

• Sensitivity to Environmental Noise: Acoustic sensors are susceptible to external noise, such as machinery operation or ambient sounds, which can interfere with the detection of printing defects.
• Complexity in Data Interpretation: The acoustic signals generated during FDM printing are complex and require advanced signal processing techniques, such as Fourier transforms or machine learning algorithms, to extract meaningful information. This complexity limits the practicality of acoustic sensing for real-time monitoring, as it demands significant computational resources and expertise.
• Challenges with Sensor Placement: The placement of acoustic sensors has a significant impact on data quality. Sensors too far from the sound source (e.g., the nozzle or print bed) may miss relevant signals, while those too close may be overwhelmed by operational noise.
• Scalability and Real-Time Application: While acoustic sensing is promising in controlled settings, its scalability to industrial or high-speed FDM applications is uncertain.

2.4. Filament Properties Monitoring

Jakub et al. [116] developed a filament winding measurement configuration for material extrusion, where uncontrolled process parameters prevented a detailed analysis of setup outputs (see Figure 4 and Figure 5). Establishment of a controlled monitoring system with feedback integration requires maintaining a consistent separation between the measurement output and the extruder, ensuring sufficient data capture on filament material properties for specific segments. Development addressed the demand for a real-time, cost-effective filament quality assessment tool, achieved through tracking capacitance variations and diameter changes across multiple sensor axes, encoding results to corresponding filament fragments. Evaluation enabled monitoring of material moisture, roundness, internal and external irregularities, compositional ratios, electrical permeability, Poisson’s ratio, and material classification. The demonstration introduced an initial proof-of-concept device, detailing its operational mechanics, and showcased applications including filament type identification across manufacturer variants, temporal moisture absorption tracking, and defect detection of varying severity—enhancing material extrusion process reliability through comprehensive, segment-specific quality insights.

Sgrulletti et al. [117] Integrated real-time optical and thermal monitoring into the Fused Filament Fabrication (FFF) process, embedding both sensor types within the printing head. Application extended to Polyamide 6 (PA6), a semi-crystalline thermoplastic commercially known as Nylon, noted for its robust mechanical properties suited to automotive, electronic, food packaging, and sports engineering applications. Modification of a standard 3D printer enabled the incorporation of optical and infrared cameras within the printing head, allowing for continuous observation of PA6 fabrication. Employing a statistical Design of Experiments approach identified and refined critical process parameters influencing FFF outcomes, targeting enhancements in mechanical strength and geometric precision. Analysis revealed optimal settings for temperature, extrusion rate, and layer thickness, correlating thermal and optical data with material behavior and structural integrity, marking an initial application of such dual-monitoring to PA6 printing for improved process control and product quality.

Some limitations and gaps in filament properties monitoring include [19,103,118]:

• Sensor Accuracy and Calibration: The precision of sensors used to measure filament properties, such as diameter, is critical for effective in-situ monitoring. However, achieving consistent accuracy across different filament types and colors remains a challenge.
• Speed of Monitoring: FDM is a continuous process, and the monitoring system must keep pace with the extrusion rate to provide real-time feedback. Current sensor technologies may not be fast enough to capture rapid changes in filament properties, particularly at higher print speeds, which limits their effectiveness in dynamic printing environments.

2.5. Print Head Monitoring

Isiani et al. [119] identified a deficiency in comprehending print head orientation effects during fabrication, noting limited exploration of belt and wheel parameter impacts on 3D printer performance. Execution of an experimental investigation (see Figure 6) assessed print head orientation and carriage condition within a 3D printing system, alongside evaluation of belt tension and wheel alignment influences. A collection of data under varied configurations revealed that orientation adjustments altered the consistency of layer deposition, while belt and wheel settings affected motion precision and vibration levels. The analysis provided a critical understanding of how these elements governed print accuracy, speed, and structural quality, addressing a previously underexplored aspect of process optimization. The extension of findings offered actionable insights for enhancing 3D printing efficiency and reliability through refined mechanical adjustments, thereby bridging a knowledge gap in fabrication dynamics.

Some limitations and gaps in gyroscopic-based monitoring include:

• Sensitivity to Vibrations and External Factors: Gyroscopes are highly sensitive to vibrations and external movements, which are common in FDM printing environments due to the presence of motors or surrounding equipment. This sensitivity can result in false positives or inaccurate data, making it challenging to differentiate regular operational movements from problematic ones.
• Scalability and Real-Time Application: Gyroscopes work well in controlled settings, but their scalability to industrial or high-speed FDM printing is uncertain due to the computational demands of real-time data processing and defect detection.

2.6. Thermal Imaging

Binder et al. [120] devised a methodology for tracking thermal profiles in the Fused Filament Fabrication (FFF) process, integrating thermal camera imagery with object models derived from G-code instructions governing printing operations. Calibration of thermal cameras initiated the approach, followed by synchronization of thermal images with point-based 3D object models, capturing temperature and point-type data at discrete intervals. Mapping of thermal distributions (see Figure 7) across the fabrication sequence enabled real-time insight into heat dynamics, correlating spatial temperature variations with G-code-defined geometry. Contribution to additive manufacturing process monitoring literature emerged, offering a structured framework for data aggregation suitable for training machine learning algorithms. Facilitation of automated detection of printing anomalies, adjustment of parameters like nozzle temperature, and assessment of product quality during fabrication underscored practical utility, laying the groundwork for intelligent Industry 4.0 advancements in 3D printing optimization and quality assurance.

Bauriedel et al. [121] recognized limitations in existing monitoring techniques for Fused Filament Fabrication (FFF), noting challenges including prolonged setup durations, reliance on expensive specialized equipment, and insufficient predictive capacity for final part characteristics. Development (as shown in Figure 8) focused on a streamlined, economical method that utilizes infrared (IR) imaging to oversee and regulate the FFF process, targeting the prediction of mechanical properties from thermal data. Implementation involved capturing temperature profiles of freshly extruded material across each layer, utilizing a straightforward and cost-effective configuration that avoided niche hardware requirements. Measurement of layer-specific temperatures enabled comprehensive thermal mapping throughout fabrication, allowing for a direct link between heat distribution and structural outcomes. Training a Machine Learning (ML) model using collected IR data enabled the forecasting of tensile strength, establishing a direct correlation between thermal history and mechanical performance. Validation demonstrated efficacy in delivering actionable insights into part quality, enhancing process control without complex integration, thus broadening accessibility for FFF optimization.

Guo et al. [122] demonstrated that in-situ thermography during Fused Filament Fabrication (FFF) processes captured dynamic thermal behavior, producing time-series thermal images (see Figure 9). The representation of infrared (IR) intensity in these images indicates heat-affected zone (HAZ) temperatures, providing data for training deep learning models, such as Long Short-Term Memory (LSTM), to predict temperatures and monitor processes in real-time. Identification of challenges arose from fixed IR camera positioning against a moving HAZ dictated by printing paths, resulting in periodic feature shifts in images unrelated to temperature variations. Recognition of these shifts as noise obscuring authentic thermal data necessitated preprocessing to eliminate interference before applying LSTM. The integration of an Autoregressive Integrated Moving Average (ARIMA) time-series model with a Stacked LSTM formed a pTS-LSTM framework, effectively filtering periodic noise and enhancing temperature prediction accuracy during FFF. Evaluation through case studies confirmed that pTS-LSTM outperformed traditional LSTM and Recurrent Neural Network, demonstrating reliability with lower-quality thermal inputs. Establishment of pTS-LSTM as a superior approach for thermal-image-based monitoring highlighted potential for improving real-time FFF process oversight, adapting to practical imaging constraints.

Macedo et al. [123] investigated the uncharted impact of residual thermal stresses on printed part strength in Fused Filament Fabrication (FFF), addressing a gap in comprehensive characterization. Determination of material strength relied on residual thermal stresses derived from a coupled displacement-temperature finite element analysis, integrating these stresses with microstructural stress distributions via a streamlined micromechanical model. Execution of experimental tests on printed specimens assessed mechanical properties across varying printing speeds, complemented by micrograph documentation. Development of a thermo-mechanical finite element model simulated the FFF process, computing temporal temperature fields, cooling rates, and residual thermal stresses. Correlation of experimental mechanical properties and microstructural patterns with calculated thermal and stress profiles elucidated underlying influences on strength. Validation linked cooling dynamics and stress accumulation to observed outcomes, enhancing understanding of thermal effects on FFF part integrity across process conditions.

Some limitations and gaps in Thermal Image monitoring include [124,125]:

• Material Variability: Thermal imaging may not work well for all materials, like carbon-filled ABS, where poor thermal activity can reduce accuracy in detecting defects.
• Data Interpretation: Using thermal data with machine learning can be hard to interpret, lacking the clarity of methods like Finite Element Method (FEM).
• Testing Challenges: It may not reliably predict outcomes in multisample tests, with only two out of four fracture events predicted correctly in some cases due to setup issues.
• Consistent Cooling Rates: Thermal imaging shows cooling rates remain consistent despite temperature changes (110 °C to 170 °C), limiting its ability to detect process variations affecting mechanical properties. Stavropoulos [92] et al. utilized thermal imaging to analyze the cooling rate, which correlates with the FFF process.
• Integration with Controls: More work is needed to integrate thermal data with printer systems for real-time adjustments.

2.7. Image Recognition and Optical Scanning

Werkle et al. [126] established an innovative camera-based technique adaptable across various printer models with minimal hardware modifications, facilitating error detection independent of specific 3D printer types (as shown in Figure 10 and Figure 11). Application of a Canny edge detector to top-view images captured the actual contour of the topmost workpiece layer, while simultaneous extraction of the target contour from G-code enabled direct comparison. Processing of visual data identified deviations between observed and intended outlines, providing a fresh approach to process oversight. Validation across diverse FFF systems confirmed generalizability, revealing capability to pinpoint misalignments, over-extrusion, or layer shifts in real time. Implementation offered a versatile, low-intervention monitoring solution, enhancing control precision by linking visual feedback to programmed expectations, adaptable to varying equipment configurations.

Xu et al. [127] explored the deployment of image recognition techniques within 3D printing, targeting enhanced material identification and classification. Examination traced foundational principles and the evolutionary trajectory of 3D printing, underscoring material selection’s pivotal influence on fabrication outcomes. Investigation detailed integration of image recognition, elevating precision in material differentiation, and refining printing parameter adjustments. A demonstration through a case study showcased the successful application in production settings, with data analysis affirming accuracy and process efficiency improvements. The assessment revealed the capability to distinguish material types—such as PLA versus ABS—via visual cues, optimizing extrusion settings in real-time. Validation confirmed positive impacts, strengthening material handling and quality control, offering a scalable approach to streamline additive manufacturing workflows [128,129].

Avro et al. [130] developed a structured deep transfer learning methodology utilizing a compiled dataset of Fused Filament Fabrication (FFF) process images to detect printing irregularities. Compilation of a dataset captured defective extrusion visuals, followed by application of image augmentation to produce diverse training and validation sets accommodating part shape and monitoring condition variability. Selection of pre-trained ImageNet convolutional neural network (CNN) architectures—VGG-16, VGG-19, and ResNet-50—facilitated feature extraction from enhanced images. Introducing an automated nozzle head detection technique, refined dataset images, improving focus on critical extrusion zones. Optimization during evaluation adjusted fine-tuning parameters, balancing performance with computational efficiency. Meanwhile, measuring computing time across models identified configurations that excelled in high accuracy while demanding low resources. Analysis revealed the capability to classify anomalies, such as under-extrusion or stringing, offering adaptable solutions for real-time FFF monitoring across various operational contexts.

Narayanan et al. [131] Implemented Support Vector Machine (SVM) and Convolutional Neural Network (CNN) methodologies to identify defects in Fused Filament Fabrication (FFF) processes. Development focused on processing visual and sensor data, leveraging SVM to classify anomalies through boundary separation and CNN to extract spatial features from fabrication imagery. Application-targeted detection of irregularities, such as layer shifts, warping, or incomplete extrusions, utilizing training datasets that reflect diverse failure modes. According to Khan et al. [132] and Paraskevoudis et al. [133] The optimization of SVM involved tuning kernel functions to distinguish normal versus defective states, while CNN adaptation refined convolutional layers to enhance pattern recognition accuracy. The evaluation demonstrated effectiveness in pinpointing defects across varying print conditions, with CNN excelling in complex image-based fault identification and SVM offering rapid classification from simpler data inputs. Integration of both approaches provided a robust framework, improving real-time defect monitoring and process reliability in FFF applications.

Some limitations and gaps in Image Recognition and Optical Scanning include [19,134,135]:

• Image recognition, which uses cameras to analyze print images, can be slow, taking minutes to process, and may not suit fast printing. It often misses minor defects, detecting only those that cover 5–10% of the area, and struggles with parts of different shapes, as it’s best suited for similar designs. Open-source solutions like OctoPrint or Klipper, when combined with AI plugins, can be utilized for image recognition and print monitoring. Additionally, other FDM brands can emulate and further enhance the integrated image recognition or vision-based monitoring systems found in Bambu Lab printers.
• Optical scanning, like laser methods, may not keep up with print speed for small objects, taking seconds per layer.
• Both need faster processing for real-time use and better detection of small defects. Optical scanning lacks 360° views, and both struggle with varied part shapes.
• Integrating with printer controls for automatic fixes is underexplored, and more studies on different materials are needed.
• Neither Image Recognition nor Optical Scanning is fully integrated with closed-loop control systems for real-time defect correction.
• Most studies focus on common materials like PLA or ABS. There is a gap in understanding how these monitoring techniques perform with advanced or unconventional materials, such as composites or flexible filaments.

3. Part Reliability and Quality Control

Investigation into monitoring and analyzing MEAM process emerged as a critical technology for ensuring real-time quality control and part reliability, enabling diagnosis of printing irregularities and defects essential for predictive maintenance. Beyond deployment of sensor-based monitoring, assessment of final product quality constituted a core function, addressing multiple influencing factors such as warping, z-banding, temperature variations, filament integrity, surface roughness, surface texture, infill density inconsistencies, and infill direction precision. Exploration of these elements, alongside additional techniques applied in research, underwent detailed evaluation across subsequent sections (Section 3.1, Section 3.2, Section 3.3, Section 3.4 and Section 3.5), encompassing capabilities, instrumentation, feasibility, objectives, and outcomes. Analysis revealed how monitoring pinpointed deviations like thermal gradients or material feed disruptions, while quality analysis linked process conditions to structural integrity, offering a comprehensive framework for enhancing FFF performance and consistency through systematic study and technological integration.

3.1. Mechanical Behavior of Thermoplastic Materials

Cojocaru et al. [136] examined factors influencing mechanical behavior of PLA components produced via Fused Filament Fabrication (FFF), encompassing filament production, geometrical configuration, process settings, equipment specifications, aging, post-processing, and testing protocols. Analysis concentrated on key process parameters—layer thickness, printing speed, printing head temperature, build plate temperature, build orientation, and raster angle—drawing from extensively studied literature variables. Identification of inconsistent or vague parameter definitions underscored a pressing need for standardized terminology, as ambiguity complicated comparative evaluation of prior studies. Observation of simultaneous parameter adjustments during experiments obscured isolation of individual effects, challenging precise attribution of outcomes. Revelation of frequent tensile specimen failures outside gauge lengths emphasized urgency for tailored testing standards, suggesting increased specimen counts to mitigate errors absent specific guidelines. Compilation of literature findings indicated reduced layer thickness enhanced interlayer diffusion, minimized air voids, improved surface finish, and elevated mechanical properties, whereas elevated printing speeds compromised surface quality due to incomplete lower-layer solidification. Detection of incomplete melting at lower printing head temperatures contrasted with unstable material flow at higher temperatures, while regulation of heat distribution through enclosed workspaces diminished build plate temperature gradients. Comparison revealed upright orientations (ZY, ZX) exhibited diminished strength relative to horizontal orientations (XY, XZ, YX, YZ), with mechanical performance declining as specimen angles increased relative to the build plate, clarifying process-property relationships for FFF optimization.

Plaza et al. [137] investigated geometric attributes of PLA polymer processed through Fused Filament Fabrication (FFF) additive manufacturing, utilizing a cost-effective, open-source 3D printer (as shown in Figure 12). Characterization focused on effects of build orientation, layer thickness, and feed rate on dimensional accuracy, flatness deviation, and surface texture, revealing non-linear experimental data necessitating artificial neural network modeling for geometric prediction. Determination highlighted build orientation and layer thickness as primary influencers of dimensional precision, governed by extruder movement length and layer stacking dynamics [138]. Assessment of build orientations identified on-edge configuration as optimal, achieving dimensional deviations below approximately 13 µm across all axes at a layer thickness of 0.06 mm and feed rates between 20 mm/s and 50 mm/s, indicating superior consistency. Observations in the X-Y axes showed that upright orientation minimized variability with reduced extruder travel. In contrast, flat orientation excelled along the Z-axis with fewer stacked layers, linking geometry to process mechanics. Evaluation of feed rate indicated a negligible impact across orientations, with layer thickness effects lacking a distinct trend. However, optimal outcomes were concentrated at feed rates up to 50 mm/s and layer thicknesses up to 0.12 mm, guiding parameter selection for enhanced FFF precision.

3.2. Deep Neural Networks in Quality Control

In neural network architectures [137,139], a consensus among researchers suggests that employing fewer hidden layers and neurons tends to yield optimal results (see Figure 13 and Figure 14). Consequently, artificial neural networks (ANNs) utilizing a feedforward backpropagation approach were refined, and the most effective performance was achieved with the Levenberg-Marquardt (LM) training algorithm, the hyperbolic tangent activation function, and a specific network configuration of 2 input neurons, 8 neurons in the first hidden layer, 4 neurons in the second hidden layer, and 1 output neuron (denoted as 2–8–4–1). This optimal structure was determined by minimizing the root mean square error (RMSE), a statistical measure that indicates the average magnitude of the errors in the set of predictions, thereby ensuring the highest accuracy of the model. After establishing this mathematical model, response surfaces were generated to individually illustrate the effects for each measurement direction (X, Y, Z) and build orientation (U, O, F). For each scenario, the evolution of dimensional deviations (ΔDx, ΔDy, ΔDz)—which represent the differences between the intended and actual dimensions—and their associated tolerances were analyzed in relation to the Fr and Lt processing parameters. This approach allows for a clearer understanding of how changes in these parameters influence the precision and quality of the output, making it easier to adjust and optimize the manufacturing process accordingly.

3.3. Optimization of Process Parameters

Mushtaq et al. [140] conducted an inclusive multi-objective optimization of critical performance parameters essential for industrial applications, including tensile strength, flexural strength, average surface roughness, print time, and energy consumption. Comprehensive investigations revealed conflicting optimized outcomes, such as maximizing flexural and tensile strengths while minimizing surface roughness, print time, and energy use (as shown in Figure 15 and Figure 16). Analysis identified layer thickness as the most influential factor in achieving the desired surface roughness and print time, primarily due to the staircase effect, whereas infill density emerged as the key determinant for attaining targeted mechanical properties. Print speed also impacted mechanical characteristics through polymer healing effects. Optimal printing parameters for acrylonitrile butadiene styrene were determined at a layer thickness of 0.27 mm, infill density of 84%, and print speed of 51.1 mm/s through numerical multi-objective optimization, resulting in a flexural strength of 58.01 MPa, tensile strength of 35.8 MPa, surface roughness of 8.01 µm, print time of 58 min, and energy consumption of 0.21 kWh. The discrepancy between predicted and experimental values ranged from 0.86% to 4.55% for the respective parameters, indicating satisfactory model predictive accuracy. Reducing print time and energy consumption demonstrated the feasibility of the fused filament fabrication approach in terms of power efficiency, fuel savings, and controlled carbon emissions. The mathematical models for acrylonitrile butadiene styrene closely aligned projected performance parameters with experimental results, providing reliable data for guiding optimal printing settings during product quality testing, thereby reducing reliance on trial-and-error methods.

3.4. Machine Learning Based Quantitative Analysis

Isiani et al. [105] conducted a quantitative evaluation (as shown in Figure 17, Figure 18 and Figure 19) of vibrational dynamics in Fused Filament Fabrication (FFF) across diverse printing states, employing Fast Fourier Transform (FFT) with three sensors to track frequency shifts. Detection captured transitions from unclogged to fully clogged nozzle conditions, identifying dominant frequencies at approximately 370 Hz and 270 Hz under standard operation, though complexity in printed prototypes obscured frequency distinction during complete clogging. Measurement revealed Sensor 1 (mounted on the nozzle) provided the sharpest, most consistent amplitude readings, while Sensor 2 (mounted on the frame) struggled to resolve minimal, unstable vibrations on the damped Lulzbot frame, indicating limited sensitivity to frame dynamics. Frame-damping mechanisms inherent to the Lulzbot platform attenuated vibrational transmission to structural components, compromising detection fidelity in secondary sensor placements. Observation of Sensor 3 (mounted on the print bed), positioned at the build plate, exhibited distinct clustering due to unidirectional y-axis motion, offering data applicable to quality control and wobble detection. Analysis underscored variability in sensor response tied to mounting location and printer mechanics, linking vibrational signatures to operational health and structural stability, enhancing diagnostic precision for FFF process monitoring. The experimental methodology employed frequency-domain feature extraction to correlate sensor responses with extrusion quality metrics, establishing a framework for real-time process monitoring. While extruder-mounted sensing provided the most direct process feedback, build plate vibration analysis offered complementary data streams for system diagnostics and motion path verification.

3.5. Statistical Evaluation of Surface Roughness

Buj-Corral et al. [141] examined the roughness surface of hemispherical PLA cups fabricated via Fused Filament Fabrication (FFF), utilizing Adaptive Neuro-Fuzzy Inference System (ANFIS) modeling for analysis (as shown in Figure 20 and Figure 21). Evaluation across specified parameter ranges revealed minimal disparity in roughness between outer and inner layers, with layer height exerting the strongest influence, followed by nozzle diameter, on arithmetical mean height (Ra) and maximum roughness profile height (Rz). Assessment of kurtosis (Rku) indicated a slightly flattened roughness distribution (Rku < 3), more pronounced externally than internally, driven predominantly by layer height, with higher values at 0.1 mm compared to 0.3 mm. Observation of skewness (Rsk) disclosed negative values across both surfaces, reflecting deeper valleys relative to peaks, with the most negative readings at 0.3 mm layer height and near-zero values at 0.1 mm, highlighting layer height’s role in texture distribution. Integration of ANFIS modeling clarified parameter impacts, linking finer layer settings to reduced roughness variability and distinct surface characteristics, enhancing understanding of FFF surface quality optimization.

Additionally, Selvam et al. [142] implemented multi-optimization strategies to determine optimal Fused Filament Fabrication (FFF) parameters enhancing surface quality of flat upper, lower, and side surfaces while minimizing printing duration. Utilization of Response Surface Methodology (RSM) and Particle Swarm Optimization (PSO), coupled with ANOVA statistical analysis, facilitated prediction of ideal settings, assigning weights of 0.25, 0.25, and 0.5 to surface roughness top/bottom (SRTB), surface roughness sides (SRS), and printing time (PT) via the Equal-weights approach. Calculation via RSM yielded printing speed at 76.98 mm/s, nozzle temperature at 220 °C, and layer thickness at 0.25 mm, producing SRTB at 3.23 µm, SRS at 1.51 µm, and PT at 32.15 min, whereas PSO predicted 125.6 mm/s, 221 °C, and 0.29 mm, resulting in SRTB at 3.84 µm, SRS at 1.72 µm, and PT at 23.47 min. Applying the Weighted Aggregated Sum Product Assessment (WASPAS) ranking method compared RSM and PSO outcomes, revealing PSO’s superior efficacy in addressing multi-objective challenges. Validation against experimental data confirmed predictions, with deviations below 4%, demonstrating reliability in optimizing surface finish and efficiency. The establishment of these findings supported the development of statistical models, reducing experimental iterations and refining FFF process control for enhanced production outcomes.

Generally, the limitations and gaps in part reliability and quality control include [19,143]:

• Need for Total Monitoring: There is a lack of systems capable of monitoring the entire print from all sides, which is essential for comprehensive defect detection.
• Lack of Research on Defect Formation Mechanisms: While defects like warpage and abnormal leakage can be detected using in-situ monitoring, the underlying mechanisms of their formation are not well-researched.
• Monitoring System Gaps: Current in-situ monitoring systems for FDM lack comprehensive development, particularly for next-generation additive manufacturing applications.
• Computation Time: There is a need for research to reduce computation time for real-time error detection and correction, as well as to develop more cost-effective monitoring solutions that are accessible to a wider range of users.

4. Conclusions and Recommendations

This review paper encompassed the application of sensors for in-situ monitoring in Material Extrusion Additive Manufacturing (MEAM). Synthesis of literature offered a comprehensive assessment of recent developments in sensor-based in-situ monitoring methodologies utilized across research efforts, detailing advancements in real-time observation capabilities. Examination highlighted the deployment of diverse sensing technologies—thermal, acoustic, optical, and vibrational—to track process dynamics, capturing data on material behavior, defect formation, and equipment performance during fabrication. The evaluation revealed progression in sensitivity and integration, enabling the detection of anomalies like flow irregularities or structural weaknesses, thus informing process optimization strategies. Compilation of these findings underscored the evolving role of sensor technologies in enhancing MEAM precision and reliability, providing a foundation for understanding current capabilities and guiding future innovations in monitoring techniques.

4.1. In-Process Fault Detection Systems

Investigation into in-situ monitoring within MEAM revealed significant advancements, yet a persistent gap remained in deploying real-time in-situ monitoring capabilities. Evaluation of existing literature indicated a predominance of post-process analysis over in-process assessment, limiting immediate insight into fabrication dynamics. Integration of machine learning appeared in select studies, enhancing data interpretation, but the prediction of machine operational status during 3D printing remained underdeveloped. The requirement for sustaining print quality consistency and part reliability underscores the necessity for continuous, real-time monitoring to capture transient anomalies, such as material flow disruptions or thermal irregularities. Addressing this deficiency involved shifting the focus toward systems capable of instantaneous data acquisition and analysis, enabling proactive adjustments to process parameters and thus fortifying MEAM reliability and performance across fabrication cycles.

4.2. Industrial Integration of In-Situ Monitoring Systems

Additionally, there is a gap in applying in-situ monitoring to enhance industrial processes and machine integration, particularly in detecting structural faults that are critical to component functionality. The assessment found that data-driven methods were transparent but limited for validation, favoring physics-based approaches that utilize thermodynamic and kinetic data for precise fault detection. These approaches promise reliable in-situ validation and improved process control.

4.3. Determination of Melted Plastic Volume

Determination of melted plastic volume within the hot end nozzle emerged as essential in rheological analysis for MEAM to enhance flow uniformity. Quantification utilized real-time measurement approaches, such as inline sensing or imaging, to assess the molten material volume during extrusion, addressing flow inconsistencies linked to thermal variations, viscosity shifts, and nozzle configuration. Correlation of volume data with pressure and temperature distributions stabilized extrusion behavior, reducing deposition irregularities such as excessive or insufficient material output. Refinement through this method facilitated the adjustment of process parameters, including feed rate and heating levels, ensuring consistent material delivery, thereby improving the repeatability and quality of fabricated components across MEAM processes.

4.4. Lack of Closed-Loop Control

The absence of closed-loop control in FDM presents significant challenges in ensuring consistent print quality and reliability. Without real-time feedback mechanisms, FDM processes are susceptible to defects such as layer misalignment, warping, and inconsistent extrusion, which can undermine the structural integrity and dimensional accuracy of fabricated components. These limitations are particularly critical in high-precision applications, including aerospace and medical device manufacturing, where repeatability and quality are paramount. To mitigate these issues, research efforts should be directed toward developing and integrating closed-loop control systems capable of monitoring critical process parameters, such as temperature, extrusion rate, and layer adhesion, and implementing automatic adjustments in real time. Such advancements would enhance the robustness of FDM, reducing print failures and elevating overall part quality to meet stringent engineering standards.

4.5. Minimal Integration of Sensor Modalities

A recent literature review reveals that most studies utilize data from only one sensor type, reflecting a minimal integration of sensor modalities. This reliance on single-sensor data limits the scope of insights, potentially obscuring critical system interactions that a multi-sensor approach could reveal. To advance the field and foster innovation, future research can prioritize the integration of diverse sensor modalities, thereby enhancing prediction and enabling more comprehensive analyses.

4.6. Underdeveloped Physics-Based Models

Current models fall short in accurately simulating critical aspects such as interlayer adhesion, viscoelastic relaxation of thermoplastic materials, and thermal-induced deformations during deposition. These limitations frequently result in discrepancies between predicted and actual part performance, compromising print quality and restricting FDM’s applicability in high-precision fields like aerospace and biomedical engineering. Consequently, targeted research is urgently needed to develop advanced models that comprehensively address these complex physical interactions.

4.7. No Standards of Reference for What Is a “Good” Print Outcome

This lack of universally accepted benchmarks complicates the evaluation of key quality metrics, including dimensional accuracy, surface finish, mechanical integrity, and adherence to design specifications. As a result, manufacturers and researchers are left without a consistent framework for quality control, process optimization, and cross-study comparisons, which in turn hinders the broader adoption of these technologies in precision-critical fields. The implications are significant, as the inability to reliably assess print quality undermines confidence in additive manufacturing’s potential for producing functional, load-bearing components.

4.8. Economic/Practical Feasibility

Despite significant advancements in engineering and applied sciences, insufficient attention has been devoted to the’ economic and practical feasibility of emerging technologies. While research often focuses on achieving breakthroughs in laboratory settings or optimizing theoretical models, critical factors such as scalability, cost-effectiveness, and integration into existing infrastructure are frequently overlooked. To address this, future research can shift toward comprehensive feasibility studies, incorporating tools like cost-benefit analyses, life-cycle assessments, and evaluations of implementation challenges. The use of life-cycle assessments helps analyze a material’s lifespan, as well as its efficiency and effectiveness when used in FDM machines. While minimizing the cost of FDM machines is important, the use of substandard materials is not recommended. By prioritizing these aspects, engineers and scientists can ensure that new technologies are technically impressive, economically sustainable, and practically deployable, paving the way for their successful transition from research concepts to impactful, market-ready solutions.