Novel Correlations Between Process Forces and Void Morphology for Effective Detection and Minimization of Voids During Friction Stir Welding

Friction stir welding (FSW) is a solid-state welding technique invented by The Welding Institute (TWI) in 1991 [1]. With process temperatures below solidus, FSW avoids fusion-related defects such as hot cracking, spatter, etc., and thus improving the weld quality in comparison to traditional fusion welding processes. FSW also permits the joining of difficult welds and dissimilar alloys [2–4]. Furthermore, no requirement of any filler material, absence of fumes, etc. ensure that FSW has a lower environmental impact. The fundamental mechanics underlying the process is the intense plastic deformation due to high strain rates that lead to material bonding and dynamic recrystallization in the stir zone. These characteristics, combined with lower residual thermal effects, make FSW an energy-efficient method of creating high-quality joints for many widely used engineering metallic materials like aluminum alloys, magnesium alloys, and stainless steel [3]. However, sub-surface defects and material heterogeneities resulting from the FSW process often necessitate post-weld inspection to ensure the quality of weld obtained. In this context, in-process void detection techniques can potentially help in optimizing the process conditions and thereby reduce expensive and time-consuming post-process inspection of welds. Current in-process defect detection techniques rely on approaches that try to directly correlate the part-scale welding quality to defect formation, without a fundamental understanding of the underlying mechanics and materials physics that modulate defect evolution.

In this work, we seek to demonstrate an in-process numerical technique that uses process force signals to detect volumetric void formation and connect the variations in the force signals to interactions between the FS tool probe and the underlying material voids. While the spectrum of possible types of defects during FSW is quite broad, in this work we solely focus on the underlying non-optimal material plastic flow that results in material voids. In this spirit, our reference to a defect in this manuscript implies the underlying material void. This work is a continuation of the FSW experimental studies performed by our group in Ref. [31], which indicated that volumetric voids formed with tools, including three flats on the probe, distort the oscillatory component of the process force signal. The oscillatory process force signal at the FS tool rotational frequency was described by identifying the opening and filling of a cavity in the wake of the FS tool probe. However, a comprehensive correlation of the probe–void interactions and various distortions of the oscillatory component of the process force signal cannot be established with limited experimental data. In this work, a finite element (FE) method-based numerical approach for modeling FSW was developed for identifying and characterizing the process forces and their correlation with the underlying material flow and defect (void) structures. This model is employed to numerically extract the process force signals and the corresponding volumetric void distributions in the vicinity of the probe. We focus on the interactions of the FS tool geometry and FS tool motion with the surrounding material undergoing plastic deformation and deduce novel insights into various correlations of FS tool motion and void formation. Through this approach, we demonstrate that we can identify FS tool-related process conditions that can be optimized to minimize void formation. This can potentially lead to the development of an in situ force-based void monitoring method that links to the underlying material plastic flow during the FSW process.

In the following subsections of this introduction, we provide a literature survey of the state of material flow modeling, in situ monitoring of voids, and related numerical approaches that are popular in modeling of FSW. Beyond the introduction, we present the experimental and numerical methods used in this work, followed by a detailed description of the results and their discussion.

Extensive research has been published hypothesizing various mechanisms to understand the material flow and mechanics underlying FSW. Cross-sectional imaging has regularly revealed an extruded band of material in the horizontal plane of the FSW setup. This intermittence is also observed in the FS tool force signals recorded during the process. The origins of oscillatory force signal and the intermittent banded structure were first discussed by Arbegast [5] and then expanded upon by Boldsaikhan et al. [6]. These authors hypothesized periodic cavity opening and filling around the FS tool probe with each FS tool revolution. Zaeh and Gebhard [7] investigated the material transport around the FS tool by the periodic action of the FS tool. Material transportation around the FS tool happened by the interaction of FS tool probe and workpiece material which causes intermittent material flow. Variation in shear bands by measuring a single band was measured by Fonda et al. [8] which led them to propose that the oscillatory motion of the FS tool because of the FS tool eccentricity (runout) or deflection causes weld banded structure. The intermittency has also been attributed to multiple process dynamics such as the stick-slip friction condition between FS tool and workpiece [9], FS tool runout [10], and probe features [11,12]. Overall, developing an understanding of the physics underlying void formation has been one of the centerpieces of research in the FSW scientific community. Intermittent material flow is critical to bolster this understanding, and recent studies have investigated the role of insufficient plastic flow within the periodic motion of the FS tool to volumetric sub-surface voids [13,14].

The potential for sub-surface void formation during FSW partly contributes to the need for post-process weld quality evaluation for defects. Nondestructive evaluation ethods such as eddy current testing [15] and ultrasonic testing [16,17] have been explored in the past; however, the cost and time involved with these methods are significant and this often acts as a dampener to their extensive adoption in the context of FSW. If one assumes that the oscillatory property of the FS tool reaction force signal is directly linked to the plastic flow of material around the probe, then a breakdown in the material plastic flow would result in a distortion of this process force signal. Over the years, various researchers have suggested in situ volumetric void detection methods as alternates to expensive post-process inspections, and as a mechanism to enable accurate in-process corrective action. Along these lines, if distortion in the process forces signal can be captured and correlated to the occurrence of volumetric voids, the need for costly inspections can be eliminated and replaced with accurate in-process monitoring. Mishra et al. [18] have shown in-process monitoring and process control in FSW by mainly focusing on force and torque measurements. Several different techniques based on process force and torque signals have been discussed in the literature [19–24]. One general observation that can be made about existing literature on this topic is that most prior studies do not provide a fundamental understanding of what is physically happening during the process and what creates the distortion in forces/torques signals at the time of void formation. More recently, the use of machine-learning algorithms combined with frequency analysis of force signals to link distortions in the measured signals to the occurrence of voids is gaining popularity, including some of our recent work [31]. While these techniques are important black-box approaches, they are limited by the need for extensive training of the algorithms with experimental data that is often sparse, and limiting the in-process monitoring technique to the vicinity of the solution space of known process variations such as probe geometry, welding machine stiffness, workpiece material, etc. Currently, there is an absence of a fundamental understanding of what is happening within the measured force signal under process conditions leading to volumetric void formation, and a lack of understanding of the correlation between force signal variations and the underlying plastic flow and void structures. This current study seeks to address this limitation and is discussed in detail in the subsequent sections.

Existing research has strongly demonstrated that numerical modeling is a powerful tool in terms of gaining a fundamental understanding of the material plastic flow inherent to FSW. Some of the relevant computational studies are listed below.

To model the material flow and predict deformation during the FSW process, the numerical methods of choice are computational fluid mechanics (CFD) and computational solid mechanics (CSM). Depending on the intensity of the plastic flow that needs to be modeled, one chooses a CFD (Eulerian approach) or CSM (Lagrangian approach) based technique. However, relatively recent computational frameworks like arbitrary Lagrangian–Eulerian (ALE) or coupled Eulerian–Lagrangian (CEL) formulations integrate both these approaches to result in a framework that is neither fixed in space (Eulerian) nor attached to the material (Lagrangian), and are widely used to predict plastic flow and volumetric void formation.

Schmidt and Hattel [25] simulated the plasticized stir zone and weld voids using the ALE framework. To avoid the high computational cost of the ALE formulations and the incapability of CFD frameworks towards predicting plasticity and void formation, the CEL framework is being widely used in recent literature, and hence is also adopted in this current study. CEL enables better modeling of the material flow, formation of plasticized stir zone, and different types of voids that might occur during FSW. Al-Badour et al. [26] developed a CEL formulation to explore the effect of coefficient of friction and process parameters on weld quality and void formation. Zhu et al. [27] have used the same framework to investigate the probe geometry and its effect on material flow during FSW. They showed that the probe feature has a substantial impact on weld quality. Dialami et al. [28,29] created a numerical framework based on ALE and CEL stages, and by adding a particle tracing strategy, this framework could simulate defects like joint line remnants. Ajri and Shin [30] have simulated different voids including cavity, tunnel, groove-like, and excess flash formation using the CEL framework. Furthermore, the dependence of void formation on temperature distribution, equivalent plastic strain, and material flow was investigated using this model.

The objective of this study is using a high-fidelity finite element modeling framework to gain a fundamental understanding of the correlation between the process forces and the underlying void structures. In the context of intermittent material flow, we show that the resultant interaction force is due to an eccentric motion of the FS tool in the workpiece material. The interaction of probe features and the workpiece material leads to the intermittent material flow that in turn causes oscillatory force signals.

In this study, friction stir welding was performed by a three-axis computer numerical control (CNC) milling machine (HAAS TM-1). A three-axis piezo-electric force dynamometer (Kistler, model 9265) was used to measure the forces applied to the workpiece by the FS tool in a three-axis coordinate system defined by the axis of the FS tool (Z), the direction of travel (Y), and perpendicular to travel in the plane of the workpiece surface (X). Signals from the dynamometer were guided to the charge amplifiers to read the forces by the data acquisition (DAQ) system (National Instruments, BNC-2090A, PCI-6014, PCIe-6320). The average steady-state forces during the stirring phase are the only force that is analyzed in this study. The friction stir tool was made of hardened H13 tool steel consisting of a 15 mm diameter concave shoulder and a threaded probe that tapered from 7 mm to 5 mm in diameter with three 0.635 mm deep flats. Toolholder resulted in a natural true runout for the FS tool equal to 64 µm, when it is rotating freely in the spindle and measured by a dial indicator at the tool shoulder. The origin of the FS tool runout is due to the tool and holder manufacturing process [31]. All workpieces were made of 6061-T6 aluminum alloys with approximate dimensions of 200 × 100 × 6 mm³. All 120 mm long friction stir welding tests were performed with a 3 deg tilt angle, using a mild steel backing plate with a thickness of 6.35 mm. The stiffness of the milling machine plus the FS tool and tool holder in the direction of interest (X and Y directions) was measured by the small incremental movement of the machine table in both directions to maintain contact between the friction stir tool and the dynamometer fixture. The force data are recorded by the dynamometer while the FS tool deflection was measured by a dial indicator that was fixed to an isolated static structure. The measured stiffness of the system in X and Y directions are 6150 kN/m and 8300 kN/m, respectively.

To get a consistent plunging depth for all tests, a defined preload was used by the gage block at the trailing edge of the tool shoulder and the workpiece. The plunging phase was achieved by applying a constant plunging speed of 25 mm/min and a shoulder plunging depth of 0.2 mm. After the plunging phase, dwelling time was set at 0.2 s to increase the temperature by rotating the FS tool at its place, and then the tool was moved in the traverse direction with a 3-deg tilt angle and constant speed. Table 1 shows all the experimental details of the current study. All tests were performed on a single block of aluminum and in the extrusion-direction of the plates while the initial temperature was equal to the surrounding temperature (25 °C). Each experiment for a given set of parameters was performed three times to establish repeatability.

To model the FSW process and the associated material deformation, contact mechanics and thermo-mechanics, the coupled Eulerian–Lagrangian (CEL) framework available within the abaqus [32] explicit finite element package, were adopted. The details about the CEL framework, material model (Johnson–Cook model), frictional contact model, and modeling assumptions and boundary conditions have been presented in another publication by the authors [33]. The specific goal of the current research is to use this numerical framework to extract and examine the reaction force between the probe and workpiece and to correlate the force variations with the underlying sub-surface void morphology (absence or presence of volumetric voids and their morphology). The 3D numerical model geometry, material properties, and boundary conditions were chosen to closely model the experimental setup and conditions described in Sec. 2.1. In addition to the usual material properties of the workpiece, measured properties (FS tool runout and system setup stiffness described in Sec. 2.1) characteristic of the FSW setup were also incorporated into the FE model.

In this model, the workpiece is defined as an Eulerian domain whereas the FS tool was modeled as a Lagrangian domain. As an example of the FE discretization, Fig. 1(a) shows the 3D Eulerian domain with a volume of 20 × 20 × 7 mm³, which is meshed with 23,016 thermally coupled Eulerian elements (EC3D8RT) with an element size of 0.5 mm and having four degrees-of-freedom per node (displacement and temperature). The Eulerian domain is made up of two main regions: the blue part (material region) was assigned to the aluminum alloy with a thickness of 6 mm and the red zone (empty region) with a thickness of 1 mm. While there is no material in the red zone in the reference configuration, this zone permits flash formation above the surface of the workpiece during welding.

The meshed FS tool and tool dimensions are presented in Fig. 1(b). The FS tool is modeled as a Lagrangian rigid body with 23,522 four-node thermally coupled tetrahedron (C3D4T) elements with an element size of 0.6 mm. The FS tool was simplified with straight flats at a zero-tilt angle instead of a conical shape at 3 deg tilt angle during the simulation. This simplification was done because of capturing the probe in-plane position and extrusion action during the process. To justify this simplification, the probe circumference circle diameter was considered equal to the average diameter of the conical probe in the experiment which was 6 mm. The plunging and dwelling phases of the process were not modeled to minimize the computational time-steps needed by the explicit solver, but instead, the simulation begins with the FS tool starting at an initial plunge depth. To ensure that there are no spurious numerical measurements from the initial few time-steps when the FS tool is achieving mechanical equilibrium with the surrounding material of the workpiece, numerical results were only investigated after the tool reaction forces reached a steady-state condition.

All of the FS tool movements were assigned to the tool reference point to control the tool motion accurately. FS tool runout was applied by offsetting the body of the tool from the tool reference point around which the tool rotated at 1000 rpm. The body was offset 32 µm to achieve the true runout of 64 µm that was measured in the experiment (Fig. 2(b)) [31]. Elastic compliance of the setup was modeled by adding linear spring elements between the reference point of the FS tool and fixed points in the X, Y, and Z directions (Fig. 2(a)). Since the FS tool eccentric motion occurs in the horizontal X–Y plane, the spring element in the Z direction was set to an infinite value to isolate the tool deflection in the welding plane. The X and Y direction spring stiffness are set to the predetermined values from the experiments (6150 kN/m and 8300 kN/m, respectively). The welding phase is simulated by setting the inflow and outflow velocities of the domain equal to the travel speed of the process (Fig. 2(b)). Zero material velocity constraints were applied on all the other exterior surfaces to prevent material from deforming out of the domain.

We now present the numerical simulation results and a discussion of the force dependence on various process parameters. To do so, we first discuss the graphical representation of the reaction forces on the FS tool in the polar coordinate system. This representation is ideally suited to compare the force magnitude variations as a function of the FS tool angle and as a function of the void location and morphology in relation to the tool rotation.

The oscillatory reaction forces on the FS tool as it stirs the material around it are a result of the interactions of the near-rigid probe with the intermittent plastic flow of the material around the probe. Distortion of the oscillatory reaction forces during the formation of volumetric voids (material defects) needs a fundamental understanding of the process mechanisms that drive the plastic flow around the probe.

Since the rotating probe encounters different regimes of plastic flow as it moves between the advancing and retreating sides of the weld, we hypothesize that reaction forces (caused due to the interaction of the probe, particularly the leading edge of the probe with the surrounding material) can be used as an important signature to estimate the weld quality around the probe. Figure 3 shows the process coordinate system where the X force is acting on the probe surface normal to the welding direction, the Y force is acting along the welding direction, and the Z force is the vertical force acting on the FS tool. To plot the polar plot of the resultant force, an angular encoder was used to measure the FS tool angular position within the process, and this is described in our earlier publication [31]. The resultant force acting in the X–Y plane is the net effect of the forces applied by the material to the probe. It is important to note that while the total force magnitude would also involve the Z component of the force, it is found that the magnitude of the radial in-plane forces is more useful in correlating with the occurrence of voids than the entire force magnitude.

Figure 4 shows a polar plot of the resultant radial force for a full revolution of the FS tool with both welds with voids (1000 rpm, 600 mm/min) and welds without voids (1000 rpm, 200 mm/min) conditions. For both cases, the plot is elongated along the welding direction. Furthermore, for the welds with voids case, there is a significant deviation (indicated by arrows) from the ellipsoidal shape observed in the welds without voids case. These two cases already demonstrate that interesting variations to the reaction forces can be expected as a function of the travel direction and void occurrence. We seek to expand on this observation and demonstrate the dependence of the force variations on different process parameters. The goal is to establish the in-plane radial force measure as an important metric for detecting voids, and thereby characterizing weld quality.

As part of the numerical studies, in addition to the force measurements, we monitor the evolution of void formation. To help with the visualization of void formation, a horizontal section (A–A) of the model is used to elucidate the FS tool probe motion as shown in Fig. 5. This cross section is extracted from the numerical model after the process forces reach a near steady-state. This cross section is used to understand the interaction of the probe with the surrounding material or volumetric voids within the weld. Within abaqus, a view cut with a Eulerian volume fraction of 0.5 is extracted as field output to obtain this cross section and the possible void geometry inside.

Before venturing into the estimation of numerical correlations between the forces and the internal voids, we try to provide some validation of the numerical force measurements. Towards this goal, we experimentally measure the force signal in the X direction for one full rotation of the FS tool for a given welding condition (1000 rpm, 600 mm/min) and compared this with the numerically obtained value of this force component. The resulting comparison of the experimental and numerical forces values is shown in Fig. 6. As can be seen, the overall force profiles (peak magnitudes and temporal evolution) match significantly. The finer differences in the force profiles are to be expected as the numerical model is only an approximation of the experimental setup, and the FS tool stiffness, material properties, and process conditions are only closely approximated and not exactly represented as in any numerical model of this complexity. This comparison gives us confidence in the overall representation of the process conditions, rate and temperature-dependent elasto-plastic material properties, and FS tool–workpiece interactions, and demonstrates the utility of the numerical model to simulate the physics of probe–void interaction and resulting void formation.

The fundamental understanding of probe–void interaction causing force signal distortion is important for the development of force-based void detection. This study tries to explain probe–void interaction through an inspection of the radial force response, represented as a polar plot in the X–Y plane, in the case of welds with voids and welds without voids (Fig. 7). Both welds with voids (1000 rpm, 600 mm/min) and welds without voids (1000 rpm, 200 mm/min) cases for the three-flat FS tool with no runout were considered. Because the FS tool does not have any eccentricity, each peak of the probe of the tool has a similar contact condition with the surrounding material. When there is no volumetric void, the peaks of the probe are in contact with the surrounding material during the entire rotation of the FS tool, resulting in a near circular force plot (Fig. 7(a)). However, when there is void formation, due to excessive process conditions, the force plot shows three distinct regions of deviation from a circular force profile (Fig. 7(b)). These regions are highlighted with arrows in Fig. 7(b). We attribute each of these three deviations in the force plot shape to the lack of interaction between each of the three peaks of the probe and the volumetric void in the surrounding material. When the peaks of the probe interact with the void, there is a drop in the force value, as the void offers reduced reaction to the probe due to the absence of material that needs to be being displaced and forged.

This correlation between the drop in the radial force value and the underlying void is clearly demonstrated in Figs. 8 and 9. In both figures, we show three different snapshots of the void location with respect to the probe and the corresponding radial force vector. While Fig. 8 is the observation from a non-voiding case, Fig. 9 is the observation from a voiding case. The near circular pattern of the force plot indicates consistent contact between the probe peaks and the surrounding material, without any significant force deviation that is characteristic of probe–void interaction. However, when a volumetric void is formed, the force plot shows significant deviation (indicated by arrows in the force plot in Fig. 9) from a circular pattern at three distinct points. By comparison with the void location shown in the top row of Fig. 9, we can confidently deduce that each of the three depressions in the force plot is due to the interactions of the three probe peaks with the void in their wake. Detailed videos, showing the probe–void interactions seen in the numerical simulations, can be found in the Supplemental Material on the ASME Digital Collection.

This important correlation between the radial force and the corresponding interaction of the probe with the underlying void suggests that one could use the deviations in the radial force signal to identify potential void formation. Besides the detection of the void, the magnitude of the force deviation can also potentially indicate the void morphology (size and shape). More detailed correlations of the void morphology and force magnitudes will be addressed in a subsequent publication.

In addition to the above observations, we note that the FS tool profile, FS tool runout, and stiffness of the system are also very important to force-based void detection. For example, having a sharper probe peak causes more deflection of the probe into the volumetric void and creates a larger force plot depression. Additionally, FS tool runout makes one peak of the probe more eccentric than the other peaks. It changes the contact between peaks of the probe and the surrounding material that drives the plastic flow. Furthermore, a stiffer FS tool may not be as responsive, because it cannot easily deflect into the volumetric void. These effects of the FS tool runout and system stiffness will be examined in the following sections.

Welds with the voids and without voids process regimes were simulated with a 64 µm runout of the FS tool to examine the force variations and material flow around the probe. This condition produces the most eccentricity of the probe peak, and hence the contact conditions between each peak of the probe and the surrounding material will be different. Figure 10 shows the force plot for both the welds with voids and welds without voids with a FS tool runout. Clearly, the probe–void interaction is different than the earlier case with a no-runout FS tool. In the non-voiding process condition, the peaks of the probe maintain full contact with the surrounding material during the entire rotation of the FS tool. This is the reason the force plot does not have any localized depressions (Fig. 10(a)). Relative to this plot, the force plot for the voiding case has two distinct depressions in the force plot (shown with arrows in Fig. 10(b)). Furthermore, the overall shape also has significant variation from a circular shape. On closer examination (refer to videos of the simulation provided as Supplemental Material on the ASME Digital Collection), this perturbation of the force plot and these two distinct regions of depression in the force magnitude occur due to the reduced contact between the probe and the surrounding material (due to the presence of a volumetric void), in addition to the perturbed motion of the FS tool caused due to the runout as compared to earlier cases. In comparison to the FS tool with no runout, the amplitude of the force plot for both the welds with voids and welds without voids is moderately stretched in the welding direction.

This correlation between the drop in the radial force value and the underlying void for this case is demonstrated in Fig. 11. Figure 11(a) shows the instance when the most eccentric peak of the probe moves into the void on the retreating side of the weld. Because the FS tool is nominally deflected toward the advancing side of the weld due to the average process force in the X direction [31], having a void in the retreating side causes relaxation of the loaded spring that in turn distorts the forces signals and produces a depression in the force plot. Figure 11(b) shows the second probe–void interaction. In this interaction, the most eccentric peak of the probe momentarily deflects towards the void in the advancing side of the weld due to the lack of the contact force that constrains the eccentric motion of the FS tool. These simulation results of the probe–void interaction for the case with runout show good agreement with the experimental force signals and FS tool deflections for this setup described in our earlier publication [31]. Detailed videos, showing the probe–void interactions seen in the numerical simulations, can be found in the Supplemental Material on the ASME Digital Collection.

In the previous sections, we mentioned that the FS tool probe can deflect into the volumetric void regions and produce perturbations in the force plot. This alludes to a potential role of the effective elastic stiffness of the overall setup as it affects the dynamic motion of the FS tool and its ability to swing away from the central axis of rotation. The stiffness of the setup results from the elastic stiffness of the milling machine plus the FS tool and the tool holder, as described in Sec. 2.1. As some industrial friction stir welding setups are structurally stiffer than the system used in this work, it is important to model the effect of the elastic stiffness of the setup to see how a stiffer setup performs. In general, a stiffer setup may lead to reduced FS tool deflection into the volumetric void.

Accordingly, welds in a voiding scenario with no FS tool runout were simulated with two different stiffness setups. One with the stiffness of our experimental milling machine (referred to as the regular setup) and another with a stiffness that is 10⁶ higher than that of the regular setup to model a near-rigid setup. Figure 12 shows the force plot for both the stiffness conditions. As the figure shows, like the regular setup (Fig. 12(a)), the stiffer setup (Fig. 12(b)) also shows three primary depressions in the force plot (indicated with arrows). However, the force magnitudes are much larger due to the increased stiffness (more force is needed to produce a given deflection of the FS tool), and the overall force profile is seen to have multiple perturbations from the relatively more smoother force profile for the regular setup. These simulations demonstrate that, in addition to the dependence of the force values on the setup stiffness, the methodology outlined in this paper to use force plots to detect voids is still reliable even in cases with near-rigid FSW system setups that can be found in industrial applications.

Additionally, there were many investigations for the FS tool wear and tool life, but we know relatively little about the effect of system stiffness on FS tool wear and tool life. Some experimental studies have shown that low stiffness of the system increases tool life in machining processes [34,35]. Given our current results and the similarities (for tool and workpiece force interactions) between FSW and machining processes, it is possible that tool life would be better in a less stiff system for friction stir welding. As can be seen in Fig. 12, higher reaction forces along with an increased vibrational pattern of the forces in the case of the stiffer system can potentially increase the wear of the tool. However, more detailed studies are needed to conclusively establish this correlation between the system stiffness and tool life.

In this work, we introduce polar plots of in-plane radial reaction forces on the FS tool as a strong candidate for in situ detection of void formation. We consider different process conditions that produce welds without voids and welds with voids, and study the influence of probe–void interaction, FS tool runout, and system stiffness on the evolution of the reaction force profiles and their correlation with underlying void occurrence. In each case, we provide insights into the potential causes for perturbations in the force profile and relate important distinct variations of the force profile with the underlying probe–void interactions. In summary, the following conclusions can be drawn from this study:

• Experiments and simulations show a strong correlation between void formation and perturbations to the in-plane reaction forces on the FS tool. This is due to the FS tool probe deflecting into the void region that then reduces the effective contact between the peak of the probe and the surrounding material. This, in turn, reduces the process forces that appear as a distinct depression in the force plot.

• Comparing the cases of a FS tool without any runout and a tool with runout, we observe that the force plot changes from near circular to an elliptical shape due to a significant difference in the plastic flow of material around the FS tool probe. Specifically, FS tool runout changes the interaction between the probe peaks and the surrounding material, and instead of having three primary instances of probe–void interaction (as observed for a FS tool without runout), only two instances of probe–void interactions are observed. Hence, considering the effect of FS tool runout is important for in situ force-based void monitoring.

• The effect of system stiffness on the probe–void interactions was also investigated. Results show that FSW setups with higher system stiffness are as sensitive as regular CNC milling machines based FSW setups for producing distinct perturbations and depressions in the radial force profile. However, for the higher stiffness setups, the force magnitudes are much larger, and the overall force profile is seen to have multiple perturbations as compared to the relatively smoother force profile of the regular setup.

The authors gratefully acknowledge the financial support for this work from the U.S. National Science Foundation (NSF Grant No. CMMI-1826104), the Department of Mechanical Engineering at the University of Wisconsin-Madison, and colleagues in the Multiscale Metal Manufacturing Processes Lab.

There are no conflicts of interest.

The authors attest that all data for this study are included in the paper.