Quality control in welding is critical, as defects waste energy, material, and human effort. In recent years, deep learning models have been widely used for defect detection, but they have drawbacks in efficiency and real-time capability. Advanced control models require not only defect detection but also quantification of their severity. This work presents an image processing approach to detect and quantify undercut defect severity during welding. The process is monitored using a welding camera with a structured light pattern. A metric for quantifying the undercut defect is proposed, which measures discontinuity length in the pattern.

The development of additive manufacturing has made these technologies suitable for fabricating end products. This encourages companies to identify quickly parts in large databases for which switching from traditional manufacturing technologies to additive manufacturing is convenient. Typically, the manufacturing process selection is made by experts who weigh various parameters, but evidence suggests that intelligent systems could beneficially replace or aid this manual selection. One challenge in using manufacturing data for advanced analysis and machine learning is that it is usually unlabeled, and manual data labelling is expensive and time-consuming. This paper deals with the application of an enhanced unsupervised learning algorithm that automatically identifies parts suitable for additive manufacturing based on parts geometry as a preliminary step of process selection. One hundred randomly selected parts were evaluated by manufacturing experts through a survey and then clustered by the proposed algorithm. The comparison of the manual and algorithmic classifications, using unsupervised learning, regarding suitability for additive or traditional manufacturing is the main original contribution of this study. Overall, 78% convergence between most experts’ designations and the unsupervised learning algorithm is achieved. For those parts where expert opinions are substantially aligned, the algorithm showed a 90% convergence rate with human choices. These outcomes support the introduction of an intelligent system to perform a preliminary identification of suitable manufacturing processes based on part geometry, as it can be seen beneficial if compared with the time and cost spent when involving a pool of experts.

The contact tip to workpiece distance (CTWD) has a considerable effect on the current of the wire arc additive manufacturing (WAAM) process, which in turn affects the height of the deposited layer. An unstable CTWD may result in significant deviation in the height direction. Accordingly, it is necessary to monitor and regulate the CTWD for each individual layer. In practice, CTWD can be calculated from the gaps between the torch and the current layer edge. An image-based layer height estimation method is developed to ascertain the height of the newly deposited layer. A welding camera, affixed to the torch, is utilized to document the deposition process. A segmentation network is utilized to identify the perimeter of the newly deposited layer. Given that the camera is continuously focused on the molten pool, it is essential to ascertain the camera's position to calculate the layer height. This is achieved by synchronizing the robot position data and the images in ROS2 (Robot operation system 2) to locate the position of camera in the real world. To find out a better choice, three distinct deep learning-based segmentation algorithms, namely Unet3 +, YOLOv11, and PIDNet, are evaluated in terms of accuracy and efficiency. Then, the layer height estimation method is tested with a 10-layer thin wall. As a result, the proposed method can provide accurate height estimation. Among the segmentation algorithms, PIDNet using 256x256 resolution is considered as a better choice to balance the accuracy and efficiency.

A significant aspect of modern manufacturing involves the selection, modelling, and dynamic control of process parameters. This necessitates the development of advanced modelling methodologies capable of capturing intricate cause-and-effect relationships and interdependencies across systems and subsystems. The Dimensional Analysis and Conceptual Modelling (DACM) framework addresses this need by using directed graphs to represent causal relationships within complex systems. However, while the DACM framework is effective for modelling individual systems, it has limitations when connecting different parts of a system (subsystems) that need to work together. Moreover, the framework requires a mechanism to extract meaningful relationships between variables. To address these limitations, this study introduces new organs derived from bond graph theory, which perform specific functions, enabling effective modelling of interconnected systems. Expert knowledge, combined with the DACM framework, is used to represent systems as collections of functions and to assign appropriate variables to these functions. Relationships between variables are computed as power laws using LASSO and OLS regression algorithms. Unlike previous works, these algorithms rely solely on the fundamental dimensions of variables. In addition to variable assignment, expert knowledge is also used to validate the computed power laws, thus substituting experimental data with expert knowledge. The interaction between the derived dimensionless numbers is represented as a directed graph, capturing the causal relationships between the functions. The approach is demonstrated through a qualitative case study of Gas Metal Arc Welding, illustrating its potential for system modelling, analysis, and using explainable directed graphs for contradiction resolution.

Despite considerable progress over the past 3 years, some issues remain in obtaining printed polyetheretherketone (PEEK) parts printed by the material extrusion process (MEX). The fast crystallization of PEEK during printing results in distortion, poor dimensional accuracy, and low inter-filament adhesion. Optimizing the process requires a fine understanding of the material deposition and optimization of the features affecting the mechanical properties of the output. In this study, we modeled the polymer melt deposition on the substrate, heat transfer, and crystallization kinetics of PEEK. The isothermal and non-isothermal crystallization kinetics were determined for the processing range from the glass transition to the melting temperature. The non-isothermal crystallization of the polymer was applied to the process using the modified Nakamura non-isothermal crystallization equation. Moreover, the influence of the environmental temperature on the crystallization kinetics was determined for the deposition of the first layer on the substrate. This study determined the optimal process parameters to improve the mechanical properties of the manufactured parts, owing to a better interdiffusion of the polymer chains between the deposited layers and less porosity.

The integration of robotics and ML is driving innovation in digital manufacturing. Robotic welding benefits from precise defect detection and monitoring. Tack welds, crucial in pre-welding assembly, can affect final weld quality if improperly formed and unaccounted for while welding over them, making its rapid detection essential. Leveraging TinyML for on-device inference enables immediate tack weld detection using a welding camera, eliminating reliance on high-latency cloud processing. This study integrates TinyML-based tack weld detection on a Renesas EK-RA6M5 platform running micro-ROS. By leveraging convolutional neural networks and the Edge Impulse platform, a model to classify tack weld online is developed. During development, the model reached an estimated F1-score of 98.7

This study presents ongoing research on a novel vision-based structured light system for online weld geometry measurement in robotic welding. The approach integrates a Cavitar welding camera with illuminating laser projecting structured light patterns onto the weld surface, enabling online depth computation and 3D reconstruction of the weld bead. Iterative Closest Point (ICP) registration of computed scans against high-resolution reverse-engineering scans showed deviations ranging from 4.574 mm to 8.79e-07 mm, with an average of 0.746004 mm. These results indicate promising accuracy within industrial tolerances for applications like tack-weld detection. Future work will explore advanced point cloud processing to estimate weld bead parameters, such as leg and throat lengths, supporting closed-loop feedback control.

The evolution of manufacturing systems toward Industry 4.0 and 5.0 paradigms has pushed the diffusion of Machine Learning (ML) in this field. As the number of articles using ML to support manufacturing functions is expanding tremendously, the main objective of this review article is to provide a comprehensive and updated overview of these applications. 114 journal articles have been collected, analysed, and classified in terms of supervision approaches, function, ML algorithm, data inputs and outputs, and application domain. The findings show the fragmentation of the field and that most of the ML-based systems address limited objectives. Some inputs and outputs of the analysed support tools are shared across the reviewed contributions, and their possible combinations have been outlined. The advantages, limitations, and research opportunities of ML support in manufacturing are discussed. The paper outlines that the excessive specialization of the reviewed applications could be overcome by increasing the diffusion of transfer learning in the manufacturing domain.

Abstract Background Studies have revealed a lack of representation of skin of colour patients in academic sources of dermatologic diseases, including databases. This visual racism has consequently generated less comfort and confidence among the specialists in the care and attention of this ethnic group, including the opportunity of being correctly diagnosed. Objectives To investigate and uncover potential racial biases in the HAM10000 data set through an exploratory analysis of the dark skin tones representation, the identification of inaccuracies in its documentation, the recognition of relevant skin conditions absent for darker skin and the lack of ethnic diversity variables crucial for validating diagnosis across different skin tones. Methods An exploratory examination was conducted to investigate the occurrence of dark skin within the HAM10000 database (housed in a Harvard Dataverse repository), consisting of 10,015 dermoscopic images of skin lesions. A visual depiction encompassing the whole skin tones was generated by sampling four crucial data points from each image and applying the Gray World Algorithm for colour normalization. To confirm the accuracy of the graphical representation, dermatologists validated the pixel sampling process by analysing a randomly selected 10% of the images for each type of skin lesion. This visual representation was produced for the entire data set as well as for each skin lesion type. The study was further enhanced by comparing the skin lesion representation within the HAM10000 data set against documented prevalences of relevant conditions affecting dark skin. Results Less than 5% of the images came from dark‐skinned patients. Nevertheless, in about 4.9% of cases, our pixel sampling method might inadvertently capture shadows or dark spots resulting from the imaging device or the lesion itself rather than the individual's actual skin tone. In addition, there are inaccuracies in the data set's claims of diversity and comprehensive coverage, notably the underrepresentation of conditions prevalent in darker skin and the absence of ethnic diversity variables. Conclusions Visual racism is an issue that needs to be addressed in medical sources of information and education. Image databases and artificial intelligence models need to be nourished with information, including all skin types, to guarantee equal access to opportunities. Furthermore, any instances where conditions affecting people of colour are underrepresented must be meticulously documented and reported to highlight and address these disparities effectively. This is particularly important in dermoscopy imaging, where solely relying on image‐based racial bias analysis is limited. The alteration of the patient's actual skin tone by the dermatoscope's lighting complicates the accurate assessment of racial bias.

Significant efforts have been made to understand the intricacies of the welding process using numerical methods, machine learning, dimensional, and scaling analyses. Dimensional analysis (DA) is used for qualitative studies of weld pool spreading, heat transfer in welding, welding parameters, and detachment-droplet formation in welding. Nevertheless, DA have been used often in a rather conventional manner. This article proposes to combine DA principles with causally oriented graphical representations and functional analysis to augment the separate capabilities of those methods. The approach uses a physics-based functional model to decompose the welding phenomena into functions, with dimensionless numbers (π) representing aspects of those functions in form of mathematical relationships between the variables. These mathematical relationships are illustrated as a causally oriented graph. This graph is transformed into a system dynamic counterpart. The values of π numbers are estimated using a single example with the developed methodology. The π values in the model are the analogous of biases in Artificial Neural Networks (ANN). The current modelling approach has the advantage of exploiting supplementary sources of knowledge and consequently requires limited data in comparison to supervised machine learning (ML) algorithms used in the field. The proposed methodology is demonstrated with a case study of gas metal arc welding (GMAW) for mild steel. The developed model predicts the droplet formation in GMAW with high accuracy and offer multiple possibilities for extension and generalization to other welding and additive manufacturing processes.

Fused filament fabrication (FFF) is a material extrusion-based process and one of the most popular additive manufacturing processes, widely used for rapid prototyping and manufacturing of polymeric parts. Despite the simplicity of the process, the polymer undergoes complex rheological properties’ transformation in this process. Understanding the rheological properties of the polymer during deposition is of paramount importance to model and improve the quality of the final product. In this study, for the first time, a two-phase flow numerical simulation approach with a Level Set equation has been used to model the shear rate in the FFF process after polymer exiting from the nozzle. The rheological properties of the raw PEEK as the feedstock were measured at low frequency by parallel-plate rheometer, and at high shear rate using an extensional die. Influence of travel speed, inlet velocity, nozzle diameter, and layer height on the shear rate of the deposited bead has been investigated. Obtained results revealed that the inlet velocity, nozzle diameter, and layer height highly influence the shear rate of the bead after exiting from the nozzle, while the influence of travel speed on the shear rate is negligible. Increasing the inlet velocity leads to an increase in velocity field and consequently maximum shear rate during deposition. Polymer melt tends to relax stress and become at steady state after existing from the nozzle rapidly, thus, it is required to reduce the inlet velocity using bigger nozzle diameter or smaller diameter raw filament or increase layer height to reduce the induced shear rate during deposition. Finally, multiple equations have been proposed to predict the maximum shear rate during the deposition based on the printing parameters.

Cold metal transfer wire arc additive manufacturing (CMT-WAAM) has attracted attention in recent years due to its ability to print walls with higher dimensional accuracy than regular WAAM. To print near-net shape parts by CMT-WAAM, there is a need to define a set of height-related geometrical parameters (HGPs) that can capture, quantify, and compare the quality of the height of the produced parts. In the presenting study, a set of HGPs, namely, the average height error, maximum height variation, and average absolute slope are defined and assessed. Fifteen single-track multi-layer walls are printed to check the effect of process parameters on the defined HGPs. It is found that the stability and quality of the print cannot be guaranteed by checking the visual appearance of the single beads and at least five-to-ten-layer walls should be printed. It is also found that the travel speed and the wire feed speed have positive monotonic relationships with average absolute slope and maximum height variation, respectively. Correlations between process parameters and HGPs are modeled and optimized using multi-objective optimization, and a validation test is performed to check the validity of the developed models. Moreover, HGPs of walls printed using unidirectional and bidirectional path strategies are calculated and compared. Defined HGPs are able to quantify, capture, and compare the quality of height of a wall with only three parameters. The HGPs can be used in further studies to report and compare the quality of height of thin wall structures.

This study proposes a methodology for detecting anomalies in the manufacturing industry using a self-supervised representation learning approach based on deep generative models. The challenge arises from the limited availability of data on defective products compared with normal data, leading to degradation in the performance of deep learning models owing to data imbalances. To address this limitation, we propose a process that leverages the Gramian angular field to transform time-series data into images, applies StyleGAN for image augmentation of anomalous data, and utilizes a boosting algorithm for classifier selection in supervised learning. Additionally, we compared the accuracy of the classifier before and after data augmentation. In experimental cases involving CNC milling machine data and wire arc additive manufacturing data, the proposed approach outperformed the approach before augmentation, resulting in improved precision, recall, and F1-score for anomaly detection. Furthermore, Bayesian optimization of the hyperparameters of the boosting algorithm further enhanced the performance metrics. The proposed process effectively addresses the data imbalance problem, and demonstrates its applicability to various manufacturing industries.

The computational cost of modern simulation-based optimization tends to be prohibitive in practice. Complex design problems often involve expensive constraints evaluated through finite element analysis or other computationally intensive procedures. To speed up the optimization process and deal with expensive constraints, a new dimension selection-based constrained multi-objective optimization (MOO) algorithm is developed combining least absolute shrinkage and selection operator (LASSO) regression, artificial neural networks, and grey wolf optimizer, named L-ANN-GWO. Instead of considering all variables at each iteration during the optimization, the proposed algorithm only adaptively retains the variables that are highly influential on the objectives. The unselected variables are adjusted to satisfy the constraints through a local search. With numerical benchmark problems and a simulation-based engineering design problem, L-ANN-GWO outperforms state-of-the-art constrained MOO algorithms. The method is then applied to solve a highly complex optimization problem, the design of a high-temperature superconducting magnet. The optimal solution shows significant improvement as compared to the baseline design.

Wire arc additive manufacturing (WAAM) has gained prominence in its utilization in the manufacturing industry due to its ability to build large functional components at high deposition rates. Among the different metal additive manufacturing processes, WAAM has the potential for adoption in the industry due to the ease with which the system can be integrated into factory's robotic welding cells. The ability to develop 3D components using welding has opened possibilities for redesigning and envisioning new product designs. However, there are still challenges related to ensuring process quality with WAAM. Path planning strategies have tremendous effects on structural integrity, mechanical and microstructural properties of the components. The current research aims to experimentally investigate the effect of different infill strategies on the hardness of cuboidal parts in WAAM. The experimental work uses high-strength low-alloy steel as the material of choice. These steels are found in many high-stress applications, such as automotive, load-bearing structures, and low-temperature applications that require a high strength-to-weight ratio. The study reported herein comprises of testing three different infill patterns and their impact on the final part performance (geometric, microscopic defects, and Vickers' hardness). It was observed that all three strategies ensured a stable deposition process, yet with micro and macro defects. Lack of fusion defects and pores were identified in one of the infill strategies through microscopic evaluation. The hardness mapping showed uniform properties in separate planes for all printing strategies.

This paper aims to propose an online two-stage thermal history prediction method, which could be integrated into a metal AM process for performance control. Based on the similarity of temperature curves (curve segments of a temperature profile of one point) between any two successive layers, the first stage of the proposed method designs a layer-to-layer prediction model to estimate the temperature curves of the yet-to-print layer from measured temperatures of certain points on the previously printed layer. With measured/predicted temperature profiles of several points on the same layer, the second stage proposes a reduced order model (ROM) (intra-layer prediction model) to decompose and construct the temperature profiles of all points on the same layer, which could be used to build the temperature field of the entire layer. The training of ROM is performed with an extreme learning machine (ELM) for computational efficiency. Fifteen wire arc AM experiments and nine simulations are designed for thin walls with a fixed length and unidirectional printing of each layer. The test results indicate that the proposed prediction method could construct the thermal history of a yet-to-print layer within 0.1 seconds on a low-cost desktop computer. Meanwhile, the method has acceptable generalization capability in most cases from lower layers to higher layers in the same simulation, as well as from one simulation to a new simulation on different AM process parameters. More importantly, after fine-tuning the proposed method with limited experimental data, the relative errors of all predicted temperature profiles on a new experiment are smaller than 0.09, which demonstrates the applicability and generalization of the proposed two-stage thermal history prediction method in online applications for metal AM.

Automatically extracting knowledge from small datasets with a valid causal ordering is a challenge for current state-of-the-art methods in machine learning. Extracting other type of knowledge is important but challenging for multiple engineering fields where data are scarce and difficult to collect. This research aims to address this problem by presenting a machine learning-based modeling framework leveraging the knowledge available in fundamental units of the variables recorded from data samples, to develop parsimonious, explainable, and graph-based simulation models during the early design stages. The developed approach is exemplified using an engineering design case study of a spherical body moving in a fluid. For the system of interest, two types of intricated models are generated by (1) using an automated selection of variables from datasets and (2) combining the automated extraction with supplementary knowledge about functions and dimensional homogeneity associated with the variables of the system. The effect of design, data, model, and simulation specifications on model fidelity are investigated. The study discusses the interrelationships between fidelity levels, variables, functions, and the available knowledge. The research contributes to the development of a fidelity measurement theory by presenting the premises of a standardized, modeling approach for transforming data into measurable level of fidelities for the produced models. This research shows that structured model building with a focus on model fidelity can support early design reasoning and decision making using for example the dimensional analysis conceptual modeling (DACM) framework.

The material extrusion process (MEX), also known as the fused filament fabrication process, has attracted attention in the manufacturing industry. A major obstacle to further application of the technology is the lack of mechanical strength due to the weak interlayer strength and poor coalescence between the adjacent beads. Understanding the effect of printing parameters on the coalescence of the adjacent beads is a step toward the improvement of the process. In this study, a novel two-phase flow numerical simulation approach coupled with heat transfer simulation has been applied to the high-viscosity polymers to determine the coalescence in the MEX process. The influence of printing temperature, substrate temperature, and the temperature of the printing chamber, as well as material deposition strategy (unidirectional and bidirectional) on the coalescence of the beads, has been investigated by numerical simulation and validated by experimental study. The modeling approach is applied to Glycerol, Polyether ether ketone (PEEK) and Polylactic acid (PLA). The results show that increasing temperature points (substrate temperature, chamber temperature, and printing temperature), increase the coalescence between the beads in the MEX process. The heat transfer model reveals that the cooling rate of the deposited bead in the MEX process is relatively high, and hence, the time window for reaching the coalescence between beads/layers is short. The heat transfer model also indicates that deposition of the further layers and beads does not influence the coalescence. The coalescence in the bidirectional deposition is higher compared to the unidirectional all conditions remaining similar. Unidirectional deposition leads to a uniform coalescence between the beads. However, the coalescence is not uniform for bidirectional deposition. The main novelty of this research is to simultaneously model heat transfer, shear rate and coalescence for numerical simulation to study the effect of printing parameters on the coalescence in the MEX process. Since the modeling of coalescence is time-consuming, two empirical equations based on obtained results have been proposed to predict the coalescence for PLA and PEEK separately.

The material extrusion process is one of the most popular additive manufacturing processes. The presence of porosity in the MEX printed parts, which ultimately deteriorates the mechanical properties, is one of the main drawbacks of the MEX process. The porosity in the structure is related to the shape of the adjacent beads and overlapping during the material deposition. Due to the deposition nature of the MEX process, the porosity cannot be entirely removed from the printed parts. Understanding the influence of process parameters on material deposition and the rheological properties is crucial to improving the quality of the final product. In this study, the two-phase-flow numerical approach with the level-set equations has been used for the first time to model the material deposition on the moving platform in 3D. The influence of the viscosity and printing parameters, including travel speed, inlet velocity, viscosity, nozzle diameter, and layer height, on the width of the deposited bead has been investigated. The simulation results are validated against experimental measurements with an average error of 5.92%. The width measured by the experimental study shows good agreement with the results of the numerical simulation. The comparison between the results of the 3D numerical simulation and 2D simulation reveals that the 2D simulation is not appropriate and accurate enough to predict the geometry of the deposited bead with the given set of parameter settings. The key novelty of this research paper is the application of the level-set method in a 3D context for material deposition on a moving substrate.

For effective human–robot collaborative assembly, it is paramount to view both robots and humans as autonomous entities in that they can communicate, undertake different roles, and not be bound to pre-planned routines and task sequences. However, with very few exceptions, most of recent research assumes static pre-defined roles during collaboration with centralised architectures devoid of runtime communication that can influence task responsibility and execution. Furthermore, from an information system standpoint, they lack the self-organisation needed to cope with today's manufacturing landscape that is characterised by product variants. Therefore, this study presents collaborative agents for manufacturing ontology (CAMO), which is an information model based on description logic that maintains a self-organising team network between collaborating human–robot multi-agent system (MAS). CAMO is implemented using the Web Ontology Language (OWL). It models popular notions of net systems and represents the agent, manufacturing, and interaction contexts that accommodate generalisability to different assemblies and agent capabilities. As a novel element, a dynamic consensus-driven collaboration based on parametric validation of semantic representations of agent capabilities via runtime dynamic communication is presented. CAMO is instantiated as agent beliefs in a framework that benefits from real-time dynamic communication with the assembly design environment and incorporates a mixed-reality environment for use by the operator. The employment of web technologies to project scalable notions of intentions via mixed reality is discussed for its novelty from a technology standpoint and as an intention projection mechanism. A case study with a real diesel engine assembly provides appreciable results and demonstrates the feasibility of CAMO and the framework.