Quality Assurance in Resistance Spot Welding: State of Practice, State of the Art, and Prospects

Abstract

Resistance spot welding is a process with high variability regarding the quality of the produced joints. This means that key performance indicators (KPIs) such as geometrical and mechanical features as well as failure modes can deviate from the initial design even if the same process parameters are used. The industry has developed quality assurance programs and quality control methods for tracking these KPIs; however, most of them are based on offline or/and destructive practices. On the other hand, state-of-the-art approaches have made online quality assessment feasible and proved its necessity if a 100% quality rate is required. However, limited attention has been given to “closing the loop” and providing feedback for preventing and correcting process anomalies that cause quality variations in real time. In this study, the main gaps between the state of practice and the state of the art are discussed in the context of quality assurance for resistance spot welding. Finally, the role and importance of digital twins by taking into consideration the entire welding ecosystem in quality assurance are discussed in order to form the prospects for the road ahead.

1. Introduction

Spot welding refers to several joining applications that have one feature in common. This feature is the joining of two or more workpieces at a single point by melting or mixing the metal locally. In laser spot welding [1], the incident laser beam penetrates and melts, successively or simultaneously, the thin metal workpieces rapidly and with high precision. The same goes for the gas tungsten arc spot welding [2], which, however, may employ a filler material and may not be as precise. The big advantage of these methods is that one-sided access is required for welding. Friction stir spot welding [3] is another application mainly used in research and development applications which, unlike the previous methods, does not require the involved materials to be melted. Similarly, ultrasonic spot welding [4] creates solid-state welds and is mainly used for joining thin metal components. Resistance spot welding (RSW) is by far the most used and researched spot welding method (Figure 1). Although it is considered conventional, several factors such as (i) the non-consumable electrodes, (ii) the absence of shielding gases and/or flux, (iii) its simple design, (iv) its simple operation, (v) the high welding speeds and (vi) the adaptable characteristics of the process make RSW suitable for automating welding in high-volume and/or high-rate production scenarios such as the “body in white” manufacturing stage in the automotive industry [5].

Controlling the RSW process in real time is not a trivial task, as the power control takes place in the primary coil of the transformer and the power delivery to the spot takes place in the secondary coil. This means that even if dynamic information of the joint is available during the process, such as the nugget evolution, fine adjustments on the secondary coil would induce huge fluctuations to the primary coil [6]. Two types of power sources are most commonly used in RSW: (i) the single-phase AC and (ii) the three-phase medium-frequency DC power. The main drawback of the first power source type is the low control frequency, while the main drawback of the second power source type is the negative phenomena that are induced in the welding transformer due to the frequent switching of the welding current direction [7]. According to that, even if dynamic information about nugget growth and formation is available, fine adjustments during the process are not guaranteed. Beyond that, external disturbances and errors related to the involved materials (surface roughness, contamination, poor fitting) and the welding machine’s conditions (electrode wear, misalignment) may also affect the process variability. This means that welds made with the same process parameters and with materials from the same batch may present different quality features [6].

To overcome these challenges, quality assurance programs (Figure 2) describe the steps that need to be implemented in order to ensure that a weld will perform according to its design or intended use. The implementation of these steps is called quality control and includes procedures and methods that can be utilized to execute a quality assurance plan for which the weld is compared to the design requirements (applicable code), specifications, standards, drawings, etc. While for different applications different quality assurance programs exist, the steps for executing quality control are the same. As an example, measuring the seam width, regardless of application, can be achieved using the same measuring equipment. The most common way of performing quality control is either by preventing a defect, correcting an out-of-spec joint feature (process control), or identifying the parts that are defective or not aligned with the design requirements. The industry norm is to keep the interference between inspection (if it is performed offline) and production to a minimum. Thus, for joined parts, samples are collected during production according to a specific standard, i.e., MIL-STD 105D, in order to ensure that the statistical significance of measurements is maintained with respect to production characteristics (e.g., batch size) [8].

Typically, depending on the application, legacy destructive inspection methods such as chemical, metallographic, and mechanical tests or offline non-destructive testing methods based on vision, radiography, and mechanical signals are used on samples or specimens for inspecting the joint’s features. These tests are carried out by certified personnel according to standardized procedures for defining whether a joint is defective, fulfills the design requirements, or is finally acceptable.

While for many applications offline inspection can be considered adequate, especially if the designed product includes a single or a few welds, for products that have a lot of joints, the effects of process variability, are multiplied (e.g., body-in-white [9], battery assembly for electric vehicles, etc. [4]). In the automotive industry, although destructive testing may be applied on samples, the quality of the process is monitored also by collecting and comparing process variables such as the clamping force, current, and voltage to dynamic thresholds or to fixed threshold values [10]. To this end, state-of-the-art systems [11] and approaches exist and aim, beyond implementing a non-destructive inspection [5] of the process, to integrate knowledge and quality assurance aspects into them by using technologies and frameworks from the industry 4.0 pool such as artificial intelligence, cyber–physical systems, and the internet of things [12]. This enables quality assessment [5] or even the real-time control of the process [13].

This study reviews the aspects that are related to weld quality and defects as regards the RSW process in Section 2 and Section 3. Following that, a concise walkthrough of the most common approaches and practices used in the manufacturing industry for offline inspection is given. The second part of this review analyzes representative state-of-the-art systems and approaches for quality assurance in RSW as found in the literature for the last 15 to 20 years in Section 4 and Section 5. Finally, the last part of this study concerns a discussion on the prospects for quality assurance and the role of current and future technological concepts in manufacturing.

2. Quality Indicators

The main key performance indicators (KPIs) or otherwise the measurable values used to determine the quality of a joint without considering a specific application can be classified into three main categories. The 1st one concerns the joint’s fusion zone features, the 2nd one concerns the joint’s mechanical performance, and the 3rd one concerns the joint’s failure modes. Of course, depending on the application, dedicated KPIs may be used, e.g., the electrical resistance of the joint. The above are briefly described in the following subsections.

2.1. Joining Zone’s Geometrical Features

One of the most important aspects of the joint is the nugget diameter and the nugget thickness (Figure 3). These two have been linked to the mechanical strength [14] and fatigue [15] of the joint. In addition, some studies have indicated that the heat-affected zone (HAZ) is also another aspect that could be linked to the mechanical performance of the joint [16]. Finally, indentation features, caused at the electrode–workpiece interface, have been correlated to the strength of the joint [17,18].

2.2. Mechanical Performance

The tensile–shear load–displacement curves are the most widely used metric for evaluating the spot weld’s mechanical performance [20]. Similar metrics include the load–displacement curves resulting from cross-tension load tests [19]. Fatigue life is also another metric for evaluating the mechanical performance of the joint, and it is very important in the characterization of dissimilar joints [21]. Finally, the microhardness of the weld zone is another important metric, as it typically reveals hardness differences between the base metal, the heat-affected zone, and the nugget [22,23] (Figure 4).

2.3. Failure Modes

Unlike the measurable aspects described previously, failure modes are qualitative metrics for classifying the mechanical properties of the joint. They describe the different modes upon which the weld zone can fracture under load. The most widely recognizable failure modes are interfacial and pullout. In the interfacial mode, failure occurs through the nugget, while in pullout mode, failure occurs as the complete (or partial) withdrawal of the nugget from one workpiece. In between these modes, several more have been identified in different studies (Figure 5), although without any consistency [24,25]. Finally, it has been stated that the load-carrying capacity and the energy absorption capability of those joints that fail under the interfacial mode are much less than those that fail under the pullout mode [26].

2.4. Application-Specific KPIs

While the majority of KPIs aim to provide a metric for the mechanical performance of a joint, this is not always the case. RSW is widely adopted for the assembly of batteries in e-mobility applications. This fact makes the measurement of the electrical resistance of the joints a requirement in order to assess their electrical performance [27]. However, standardization of these procedures is still under development.

3. Defects in RSW

In welding, defects are imperfections that can be caused by a bad process design but also by uncontrollable external factors. A defect does not always imply that a joint must be rejected. In the following paragraphs, the most common ones concerning the RSW process are described (Figure 3).

3.1. Expulsion

Expulsion can be described as the ejection of molten metal from the weldment area. It can be located at the electrode–workpiece interface or at the faying surface of the involved materials (Figure 6).

Expulsion is a common problem during the BIW stage with a rate of occurrence up to 60%, requiring the car bodies to be cleaned and polished before reaching the paint shop due to stuck metal on the chassis part surfaces [28].

The root cause of expulsion is the result of the force applied by the liquid nugget onto the solid containment being equal to or greater than the force applied by the electrode [29]. The above condition can be true when excessive heat is induced at the weld spot (isovolumetric heating), which can happen due to the selection of non-optimal process parameters, such as low electrode force, high amplitude current, and short welding duration. Working close to the expulsion limit is very common to the industry in order to achieve a large nugget and achieve high production volume.

In special cases of welding aluminum and copper alloys, the risk of expulsion is significantly higher compared to steel due to the abrupt change in the electrical conductivity when the materials change from solid to liquid [30,31].

While expulsion at the interface between the electrode and workpiece may only affect the electrode life and surface quality, expulsion at the workpiece’s interface can potentially affect the mechanical quality of the joint. The ejected material can cause liquid metal deficiencies in the fusion zone, which in turn can create discontinuities such as cavities during solidification. Severe expulsion at both interfaces can lead also to weld thinning, which means forming a significantly undersized nugget [28].

Expulsion can also cause damage or wear the equipment such as the welding machine, fixtures, sensors, and safety fences, especially in big automotive plants with a lot of expulsion incidences. This can decrease the service life of the equipment and increase the production costs due to prolonged downtimes and additional maintenance efforts.

3.2. Shrinkage Void

The way that the welded spot is cooled can favor the creation of voids (Figure 7). In more detail, the solidification of the nugget starts from the periphery and moves toward the center. This happens at different rates in different directions, which are parallel and perpendicular to the imaginary axis of the electrodes. More specifically, the cooling rate in the direction parallel to the electrode axis is more significant compared to the cooling in the direction parallel to the workpieces’ interfaces due to the high thermal conductivity of the copper electrodes and the fact that they are actively cooled. As a result, the growth of dendrites, in the direction of the electrodes, move faster, thus obstructing the inter-dendritic feeding during the final stages of solidification owing to dendrite coherency. This causes a shortage in liquid feeding to the nugget which along with the metal contraction is speculated to cause the so-called shrinkage voids typically near the center of the fusion zone (cavity) [32,33].

While the main cause of shrinkage voids is the immediate cut-off of the power delivery to the spot, their probability of occurrence can be increased if a number of conditions are true [34]. Thus, the tendency to form shrinkage voids is highly favored by the use of advanced high-strength steels along with inappropriate welding programs that do not aim to increase the electrode force but the welding current and time instead [33]. In addition, an increase in the sheet thickness, nugget size, and the number of sheets results in an increase in the cavity volume [32].

The void (cavity) size depends on the geometrical feature of the nugget such as its diameter and thickness. The tensile shear strength of the joint decreases drastically with an increase in cavity size as the local stresses are increased; however, if the cavity is located away from the pulling interface, it has minimal effect on the tensile shear strength [32].

Figure 7. Schematic representation of shrinkage void (left), macrograph of a shrinkage void artifact (Reprinted from Ref. [35]) (right).

Figure 7. Schematic representation of shrinkage void (left), macrograph of a shrinkage void artifact (Reprinted from Ref. [35]) (right).

3.3. Cracking

Typically, we can identify two types of hot cracking that are most commonly encountered in RSW. These are solidification cracking and cracking due to liquid metal embrittlement.

Solidification cracking (Figure 8) is due to the inhomogeneous deformation occurring during welding, which creates non-uniform stresses and strain fields. During the process, the resulting stresses at the fusion zone in their majority are compressive; however, due to the rapid cooling, tensile stresses evolve in the direction parallel to the faying surface. This along with the fact that the grain boundaries are perpendicular to this direction causes cracks along the thickness direction at these points [34]. Liquation cracking, as with the solidification cracking, is intergranular. It occurred in the partially melted zone, which is right outside the fusion zone. It is caused by the presence of low-melting eutectics, which causes continuous intergranular liquation. These liquid films have no strength to resist thermal stresses during solidification [36].

Liquid metal embrittlement is a phenomenon that can occur in specific solid/metal systems and usually under the action of tensile stresses. In RSW, the involvement of zinc-coated advanced high-strength steel induces the risk of surface cracking. These cracks most commonly appear at the periphery of the weld zone during cooling and are favored by short hold time and misaligned electrodes [37].

Figure 8. Schematic representation of different crack types (left), liquation metal embrittlement at the electrode–workpiece interface (right) (Reprinted from Ref. [38]).

Figure 8. Schematic representation of different crack types (left), liquation metal embrittlement at the electrode–workpiece interface (right) (Reprinted from Ref. [38]).

4. Practices and Norms in Manufacturing

In the industry, quality assurance for RSW is typically achieved using offline destructive inspection methods on samples and by utilizing empirical rules to select the process parameters depending on the involved materials, machinery, and design requirements. Destructive methods/techniques are carried out to qualify a welding process, a welder, or a welding operator in the context of quality control (QC), which is based on quality assurance (QA) programs and standards managed by organizations such for example the American Welding Society [39] and European Welding Federation [40]. In their majority, destructive inspection methods incorporate statistical process control by using control charts to detect and track abnormalities on product batches and to help identify non-conforming welding trends so appropriate attention is to be applied to process design and achieving stability [41]. Destructive examination is carried out offline either before the process, on the parts that are going to be welded, or after the process.

Destructive inspection methods for RSW are applied to directly measure the mechanical strength, fatigue, hardness characteristics, and failure modes of the joint. The techniques used for doing that include tensile shear tests and cross-tension tests (Figure 4 and Figure 9) [19].

Beyond mechanical testing, metallographic tests are also conducted in the context of process optimization and analysis and not for tracking-quality purposes (e.g., identifying defects and determining the fusion zone’s geometrical features or grain boundaries) [43].

Destructive testing means that the joints or the corresponding assemblies that undergoing inspection have to be scrapped. However, this is not always an option for inspecting production batches, as it is time-consuming and requires special machinery that is not portable and easy to use. Initially, the research community and then the standardization institutions established a set of methodologies to correlate the fusion zone geometrical features (e.g., nugget diameter) and defects with the mechanical performance of the joint, creating also rejection and acceptance rules.

Non-destructive methods and techniques for many applications are the norm to determine the geometrical features of the fusion zone and identify defects such as cavities pores and cracks. In RSW except for the visual inspection, which can be utilized for identifying surface defects (expulsion, indentation, cracks) and for measuring features such as the HAZ, ultrasonic testing is probably the most used one, especially in the automotive industry. Ultrasonic testing is carried out and interpreted using A-scan modules or advanced phased ultrasonic arrays [11] to depict the partial or complete morphology of the nugget. However, ultrasonic testing is used on a limited number of samples as it requires additional preparation and physical access to the part. Other methods such as radiographic testing [44] are also used for inspecting RSW joints, although these methods cannot be easily validated outside the research community, which utilizes them in the context of process model validation and the development of quality assessment approaches or process optimization approaches.

On the other hand, ultrasonic testing has been integrated into robot systems for carrying out quality assessment tasks. These tasks are realized as an additional step in the production line [45,46]. Its utilization is limited to applications where access to the joint by the robot end-effector is possible and when short times between production steps are not required. In the same way, thermal vision monitoring has been also utilized for implementing active lock-in thermography [47,48]. The utilization of this system concerns battery welding applications only.

As regards process control, the majority of the available welding systems are implementing two types of control strategies: namely, constant current control (CCC) or constant power control (CPC) [49]. In the first type of control, the current variations are controlled during welding. In the second type, both the current and voltage variations are controlled, or otherwise, the power that is delivered to the spot. The CCC is the most common one, and it is the simplest one to implement with the selected welding current being used as a setpoint for the closed-loop controller of the welding system.

The quality of the joints is typically controlled by correlating these setpoint values, which are presented as process parameters, with the quality features of the joints depending on the welding design and the application’s standards and requirements. A common case that makes use of the above-mentioned strategies concerns the creation of joints with predetermined desired geometrical features, such as the nugget diameter [13]. In several studies, the nominal value of the welding current has been linked to the diameter of the nugget, which in turn has been linked to the mechanical strength of the joint [50,51,52]. Thus, by ensuring that the welding current is properly regulated to this value, the mechanical quality of the joint can be regulated within a predetermined range.

To conclude, the practices and norms for quality assurance in resistance spot welding still use, in the majority of cases, destructive testing to determine the quality features of the joints. This is especially true when the mechanical performance of the joints needs to be assessed. Recently, in high-volume industrial applications, non-destructive methods such as ultrasonic testing have been established for measuring the geometrical features of joints on samples. Finally, quality assurance through process control is achieved by regulating a dynamic value, such as the welding current and in turn linking this value to a quality feature.

5. State-of-the-Art Methodologies

Although in the previous section (Section 4), the industry norms and legacy practices regarding quality assurance were described, in this section, the state-of-the-art approaches are classified into three categories. By “state of the art”, the authors are referring to approaches that are not yet available, either as a commercially available solution or as standardized or commonly accepted practices in the industry for assuring, tracking, or improving the quality of RSW.

5.1. Inspection

The majority of the state-of-the-art approaches are based upon a common principle: the establishment or exploitation of an otherwise unknown relation between the process’s intrinsic variables, emissions (e.g., thermal signature), or the measured weld features by using sensors and/or auxiliary signals (e.g., nugget thickness/diameter) with a quality indicator or a defect of the joint. Typically, this is achieved by utilizing supervised learning methodologies [53] (Figure 10). Of course, techniques such as radiographic imaging and spectroscopy can directly identify defects and provide metallographic information; however, they are not considered in this section as they are well established in the welding industry as regards offline testing.

The first class of the above-mentioned approaches is based on dynamic electrical features captured during the process. As such, in [54], the dynamic resistance signal has been calculated from in-process voltage and current (Rogowski coil) measurements and used to extract hand-crafted temporal features. These were used along with three weld quality labels as defined by the mechanical testing of corresponding joints (cold weld, good weld, and expulsion) to train a random forest classifier which reached 94% accuracy and even boosted to 98%, aggregating the process parameters with the feature vectors. In the same way, in [55], the authors by using a radial basis function neural network and features extracted from the dynamic resistance signal were able to predict the nugget diameter and weld strength. Monitoring the clamping force along with the current and voltage during the process enabled the extraction of simple features. These features were classified into three classes (98% accuracy), each one representing different quality levels (unfused joint, normal joint, and expulsion), by using a kernel extreme machine model that was optimized using a particle swarm optimization algorithm [56]. In another approach [57], by utilizing once again the dynamic resistance signal and a linear regression model with two variables, the authors highlighted the importance of feature engineering regarding the elimination of redundant information and the value of statistical feature extraction methods like principal component analysis.

A different type of online non-destructive inspection method is based on monitoring the displacement signals of the electrodes. For example, in [58], handcrafted features were extracted from the dynamic displacement signals of the electrodes, which were then matched with a feature map (Chernoff’s face) and used to train a Hopfield associative memory neural network with five classification outputs or otherwise quality levels. Similarly, in [59], a linear regression model is used for predicting the nugget thickness, while in [60], the electrode displacement signal is used for identifying shunt welds.

Looking at the mechanical emissions of the process, in [61], the prediction of an expulsion event is implemented by performing a seven-layer wavelet decomposition on the upper electrode force’s signal. By selecting the wavelet with the highest energy, two features (kurtosis, peak-to-peak) were extracted and used to train a simple neural network. Against the “norms”, a physics-based approach in [62] integrates an acoustic sensor into the electrode and extracts a number of ultrasonic signal’s eigenvalues in the frequency and time domain. These are then used to develop a quality assessment application that can predict the nugget’s diameter and thickness with an error of less than 7% and an overall accuracy of 96% for the quality.

Moving on to the machine-vision-based approaches, a small robot paired with an optical camera given the position of a weld takes pictures of the weld’s surface and extracts geometrical features [63]. Then, it uses them to determine eleven fuzzy input functions of a fuzzy inference system model. By considering also historical data, the weld’s topography can be generated with high accuracy and/or detect various defects within a short time window. Within the same scope, another approach by using a deep learning model for object detection enables the localization and assessment of the welds. In [64], in-process and post-process (using an induction heating source) monitoring is enabled by the utilization of an infrared camera. A database is loaded with thermal signatures of good welds made with different materials and process parameters. The measured thermal signatures are compared with the nominal good ones for rejecting or accepting the joint. Again, by using an infrared camera in [5] (Figure 11), a number of machine learning classifiers are trained and compared, concluding that the maximum infrared intensity and cooldown profiles offer the best class separability between “OK” and “Not OK” welds. In addition to that, a supplementary approach in the same study succeeds in achieving a perfect score on the classification of expulsion using only two features extracted from the recorded videos of RSW runs.

5.2. Process Optimization—Open Loop Methods

While standards offer guidelines to ensure the quality of joints, this is accomplished to a certain degree and usually for specific materials and RSW application variants. Thus, another way to implement quality assurance is the selection of the optimal parameters for a specific RSW application (Figure 12).

Response surface methodology is a widely adopted methodology for process optimization, offering an easy and fast way to select the optimal process parameter for specific applications. In the context of RSW, it has been used extensively to optimize various applications using as explanatory variables the nominal process parameters (e.g., welding current, clamping force) and as target variables, single quality indicators such as tensile shear strength and custom indicators which aggregate multiple ones [65]. Additionally, multi-objective approaches have been also used along with analysis of variance for making the final selection(s) of process parameters [66]. For establishing a relation between the explanatory variables and target variables, both machine learning models such as artificial neural networks and low-order polynomials have been used by the research community to create or fit a response surface to the available data. Data have been usually collected by some kind of design-of-experiments methods such as factorial, central composite, or Taguchi’s designs [67].

On the other hand, physics-based numerical or analytical modeling has been also used both in process optimization as well in dynamic process analysis. As such, cooling times and temperature field evolution can be observed and used for defining the microstructural evolution of joints [68].

Finally, process optimization has been incorporated for determining the process window for which the defect probability is minimized. In [69], for example, the authors determine the optimal way of delivering the welding current (pulsed or constant waveform) and the optimal value of clamping force for reducing expulsion.

5.3. Process Control—Closed-Loop Methods

Most of the common control approaches (Figure 13) presented in the literature are based on control strategies already integrated to an extent into industrial equipment and probably are already implemented in less delicate forms in industrial applications. In [13], the authors have developed, based on a CCC controller and a nugget estimator, a nugget diameter control system. The system can control and produce a joint of the requested nugget diameter with a small error. A comparison between the CCC and CPC strategy has been carried out in [48] with the CCC showing a better performance overall and being able to deliver larger nuggets using the same amount of energy.

Apart from nugget size regulation, in [71], the authors presented a novel control strategy able to prevent expulsion. Based on the fast regulation of current and the comparison of the estimated nugget size to a threshold, for the cases of under-the-edge proximity and initial gap conditions, the proposed study demonstrated the prevention of expulsion compared to multi-pulse welding control strategies.

6. Discussion

6.1. State of Practice and State of the Art: How They Affect Each Other

The current state of practice, as already presented in Section 4, lags compared to the current state of the art. On the other hand, state-of-the-art studies are drifting away from the actual requirements of the industry. Non-destructive testing such as ultrasonic testing has been well established in the automotive industry for identifying nugget geometrical features [72,73,74]. However, it is applied to samples, meaning that real-time production management is not an option; while keeping the quality of the products at a certain level, additional stages are required. Beyond that, this offline schema cannot be feasible for serving the concept of a digital product passport [75] without incorporating exhaustive quality check procedures. Thus, any form of online quality inspection would only add to the overall production flexibility and quality tracing. Finally, methods such as radiographic testing are not only difficult to implement but involve health risks for the operator if their usage is excessive [76].

On the other hand, as for those factors that cannot be directly measured or correlated to a quality feature or a defect through a known physical mechanism, there is a good chance that it might be thanks to supervised learning. In the literature, a number of studies have already proved that the process emissions include information regarding the joint formation and the generation of defects. However, not many studies have been focused on revealing internal defects in RSW such as cracks or voids [77] compared to similar applications for other welding applications [78].

Looking only in a certain direction does not reveal the whole story. The gap between the industry and academia does not always exist due to the limited adaptation of new methodologies by the industry but also due to the fact that the proposed innovations do not resolve major industry problems or are quite difficult and costly to integrate into production.

Many studies reviewed in the previous section indeed propose a method for quality assessment in real time but do not compare it with the ability to predict quality using the corresponding process design parameters [5]. In addition, many of these studies were based on monitoring system implementations that could hardly be integrated “as is” into automated welding production as they required placing sensors that were operating under very specific conditions [79] or for example requiring specialized welding machine components [62]. Finally, most of the studies based their success on machine learning methods. While this is desirable, in a real-life scenario, it must be considered that the amount of data required for developing a machine learning model is typically big, and thus it comes with a very high cost both for collecting these data and labeling them.

Process optimization and the in-process control domain, things, as presented in the corresponding sections (Section 5.2 and Section 5.3), have not evolved radically except for some specific applications. For process optimization, innovations included new information regarding the selection of process parameters for applications using non-common materials [80] but included limited information on how combinations of variables affect the quality indicators and the probability of errors. In the case of in-process control, the industry can benefit from the literature; however, this is mainly through small developments on the existing control strategies that may already be applied in advanced applications found in the automotive industry.

6.2. A Realistic Scenario for Online Quality Assessment in Production

In this subsection, the authors discuss the feasibility of implementing advances found already in the state-of-the-art into the industry. Among the studies reviewed in the previous sections, the ones concerning expulsion are the ones that present the greatest potential. This is because this specific KPI seems to have been studied extensively in a great number of works mainly regarding its online quality assessment but also its optimization (minimization) and real-time control. This claim, as concerns online quality assessment, is supported by the results of these studies [5,54,56] where the authors achieved a perfect classification score, while in [61], perfect score was almost achieved. Most importantly, these studies presented monitoring systems that can be installed into real-life production scenarios without interfering with the process and at a relatively low integration cost. In [5], the system’s integration in a production scenario was even evaluated, while in [54], it was quite obvious that integration could be achieved in robotic applications. On the other hand, for [56,61], as the monitoring setup includes pressure sensors, the feasibility may be limited to stationary applications, but this depends on the design of the welding machine. The majority of the available studies for expulsion’s online assessment [81,82,83,84] based their success on the classification of features extracted from the dynamic signals of the current, voltage, and pressure. Finally, another aspect that makes online expulsion assurance a real thing is the available studies concerning cases that involve different materials and welding infrastructure.

6.3. Digital Twins and the Welding Ecosystem

The quality assurance approaches reviewed in this study concerned the actual process. However, the quality of a joint is dictated also by other factors such as the involved machinery, materials, and in general nearly all the items and functions of a welding ecosystem. Electrodes for example which are part of the welding machine and come into contact with the workpiece are worn out during use. This has a great impact on the final quality of the joint [85]. In the same manner, the surface condition of the involved materials greatly affects the electrical contact resistance, which is an important parameter during the RSW process [86]. Finally, sensor drift and performance is an aspect also to be considered when making decisions on quality [87].

Combining these entities into one intelligent entity has been already envisioned under the concept of digital twins [88]. This kind of digital twin encompasses the twinning of individual systems and the establishment of links between them to affect each other’s internal states. A digital twin can be characterized by three main interconnected functional requirements: prediction, monitoring, and diagnosis [89].

While for other manufacturing applications, a significant number of studies have been conducted on digital twins [90], for RSW, only a few dedicated approaches exist. In [91], a digital twin model is used to obtain the temperature distribution of the workpiece at different timestamps and determine changes in the size of the nuggets. This makes the nugget formation process clear and intuitive, and the numerical simulation based on the sensors’ data can be synchronized to the real process. In another study [92], a digital twin of coaxial one-side resistance spot welding RSW was developed to predict the transient temperature field on the coaxial one-sided joint.

7. Conclusions and Prospects

In this review article, the authors analyzed the current state of practice and state of the art concerning the art as regards quality assurance in resistance spot welding. While this review does not include each and every study available in the literature, it thoroughly includes the most representative ones, capturing in this way the average trends and highlights for each domain of quality assurance and quality control. The findings of this review indicate that indeed there is a significant gap between academia and industry, but this gap regards mainly the quality assessment of the joints and more specifically the low number of existing online quality assessment applications in the industry. This means that the current practices cannot provide 100% quality tracking for all the joints of an assembly, or for all the assembled parts, despite that several state-of-the-art approaches exist and could be adopted.

In addition, the gap in quality assessment concerns also the available methods for exploiting complex relations between process variables and monitoring signals. In industry, these are limited to simple algorithms and statistical tools, while several state-of-the-art approaches have proved the advantage of machine learning for being able to capture more complex relations between process variables and monitoring signals.

Process control and process optimization on the other hand are two domains for which no radical innovations have been proposed, and the few ones existing are mainly improvements on top of commercial welding systems and strategies found in the industry.

All the above-mentioned facts, along with the potential variation of other elements that comprise the resistance spot welding ecosystem and process (materials, machines, etc.) and affect the final quality of the joint, make the zero-defect manufacturing concept hard to implement.

Toward this direction, holistic approaches have been proposed through the concept of digital twins; however, their implementation has been achieved only to a limited extent as regards resistance spot welding applications compared to other manufacturing applications.

Based on the previous points, the authors conclude that although the path toward improving quality assurance in industrial resistance spot welding applications, as always, can be favored based on the current state-of-the-art approaches, this is solving part of the problem. Multilevel approaches that take into account a greater portion of the welding ecosystem are more likely to provide a single platform for developing and implementing strategies and pave the way toward zero-defect manufacturing.

This has already been studied in the context of other manufacturing applications. In [93], the proposed digital twin for robust control of the additive manufacturing process based on the optimization of linear matrix inequalities provides a superior alternative to the classic proportional integral derivative (PID) controller due to the uncertainty management of external factors such as the material properties. A digital twin framework has been developed and tested for the case of laser welding of aluminum car doors [94]. The digital twin updates iteratively an entity, called process capability space, during the tasks of new product introduction, which is able to simulate the dimensional, geometric, and weld quality features of parts and assemblies, isolate the root cause of quality defects, and suggest corrective actions for automatic defect mitigation. This enables closed-loop In-process quality improvement with the results of the case study showing an impressive right-first-time rate of >96%.

Similar studies have indicated the need for multilevel approaches as described previously and also highlighted the advantages that come along when holistically approaching quality assurance. Thus, given the reviewed advancements in RSW, the digital twins concept provides the medium for integrating the welding ecosystem into a quality assurance strategy.