Int. J. Modelling, Identification and Control, Vol. 1, No. 2, 2006 101 Modelling of friction stir welding for robotic implementation Reginald Crawford,* George E. Cook and Alvin M. Strauss Welding Automation Laboratory, Vanderbilt University, Nashville, TN, USA E-mail: reginald.crawford@vanderbilt.edu E-mail: george.e.cook@vanderbilt.edu E-mail: al.strauss@vanderbilt.edu *Corresponding author Daniel A. Hartman NMT-10: Process Science and Technology, Los Alamos National Laboratory, Los Alamos, NM, USA E-mail: hartman@lanl.gov Abstract: A three-dimensional numerical model is used to simulate the Friction Stir Welding (FSW) process using the computational fluid dynamics package FLUENT. Two mechanical models: the Couette and the Visco-Plastic fluid flow models for Al-6061-T6 were simulated. The simulation results are compared to experimental data for Al 6061-T6 welded at high rotational (1500 5000 rpm) and travel speeds ranging from 11 to 63 ipm (4.66 26.7 mm/s) are presented. This paper examines the forces and torques associated with the FSW process with respect to considerations necessary for robotic implementation. It is shown that force control is an important requirement of robotic FSW. Keywords: Friction Stir Welding (FSW) modelling; high-speed friction stir welding; robotic friction stir welding. Reference to this paper should be made as follows: Crawford, R., Cook, G.E., Strauss, A.M. and Hartman, D.A. (2006) Modelling of friction stir welding for robotic implementation, Int. J. Modelling, Identification and Control, Vol. 1, No. 2, pp.101 106. Biographical notes: Reginald Crawford received his BS in Mechanical Engineering from Tennessee State University and MS in Mechanical Engineering from Vanderbilt University in 2005. Currently, he is a PhD candidate at Vanderbilt University. George E. Cook is currently a Professor of Electrical Engineering and Associate Dean for Research and Graduate Studies, School of Engineering, Vanderbilt University. He received his BE from Vanderbilt University, MS from the University of Tennessee and PhD from Vanderbilt University in 1960, 1961 and 1965, respectively, all in Electrical Engineering. Alvin M. Strauss received his BA from Hunter College and PhD from West Virginia University. Currently, he is a Professor of Mechanical Engineering at Vanderbilt University. He is also the Director of the NASA sponsored Tennessee Space Grant Consortium. Daniel A. Hartman received his BS and PhD from Vanderbilt University in 1995 and 1999, respectively. Currently, he is a technical staff member of Los Alamos National Laboratories manufacturing and materials joining department. 1 Introduction Friction Stir Welding (FSW) was invented and patented by Thomas et al. of the Welding Institute in Cambridge, UK. In FSW, a cylindrical, shouldered tool with a profiled probe is rotated and slowly plunged into the joint line between two pieces of sheet or plate material, which are rigidly clamped onto a backing plate in a manner that prevents the abutting joint faces from being forced apart. Frictional heat generated between the tool pin, shoulder and the material of the work pieces causes the latter to reach a viscoplastic state that allows traversing of the tool along the weld line. Currently, industries that use FSW are the aerospace, railway, land transportation, shipbuilding/marine and construction industries. These industries have seen a push towards using lightweight yet strong metals such as aluminium. Many products of these industries require Copyright 2006 Inderscience Enterprises Ltd.
102 R. Crawford et al. joining three-dimensional contours, which are not achievable using FSW heavy-duty machine tool type equipment with traversing systems, which are limited to only straight-line or two-dimensional contours. For these applications, industrial robots would be a preferred solution for performing FSW for a number of reasons, including: lower costs, energy efficiency, greater manufacturing flexibility and most significantly, the ability to follow three-dimensional contours. This paper examines the forces and torques associated with the FSW process with respect to considerations necessary for robotic implementation. It is shown that force control is an important requirement of robotic FSW. 2 Experimental procedure Experiments were performed on a Milwaukee #2K Universal Milling Machine fitted with a Kearney and Trekker Heavy Duty Vertical Head Attachment modified to accommodate high spindle speeds. The vertical head clamps the vertical sliding surface of the milling machine. A Baldor 20 HP, 3450 rpm motor controlled by a variable frequency drive is mounted on the shoulder of the head and drives the vertical spindle via a Poly-V belt. Plates of AL 6061-T651 aluminium, nominally 0.250 (6.35 mm) thick were friction stir welded. The samples were 3 (76.2 mm) wide by 18 (457.2 mm) long. The tool shoulder was flat with a 0.500 (12.7 mm) diameter. The pin was cylindrical with 10 24 threads per inch left-hand pattern. The pin length was 0.1425 (3.62 mm) and the diameter was 0.190 (4.826 mm). Heat sinks were cut into the far end of the tool shank near the shoulder to facilitate heat dissipation during welding. The tool was rigidly mounted on the tool holder using a twist lock system. The tool lead angle was set to 2. The tool depth was set to 0.145 (3.683 mm) A Kistler dynamometer (RCD) Type 9124 B was used for measuring the translational ( F x ), transverse ( F y ), axial force ( F y ) and the moment ( M z ) on the rotating tool for rotational speeds of 1500 5000 rpm and travel speeds ranging from 11 to 63 ipm (4.66 26.7 mm/s). Figure 1 shows a detailed schematic of the FSW process and the forces and torque associated with FSW. Figure 1 FSW process schematic 3 Modelling procedure FSW process modelling typically incorporates either a solid or fluid mechanics approach. Experimental results have been shown to correlate with models using either of the approaches. Owing to the moderately high temperatures associated with FSW (up to 480 C) (Sato et al., 1999), and the relatively low melting point of Al 6061-T6 (652 C); it is clear that the weld material surrounding the tool enters a temperature region where the material is not a true solid or liquid, though it has behavioural aspects of both. Understanding and accurately modelling this region will lead to an optimal three-dimensional model. 3.1 Mechanical Model: Part 1 North et al. (2000) experimentally correlated the material viscosity during FSW with the viscosity of a fluid intermediate between two concentric cylinders as first suggested by Couette in 1890 (White and Kudo, 1991). Figure 2 shows a schematic of the Couette Viscous flow model. The inner cylinder has radius r 0 and angular velocity w 0 whereas the outer cylinder has r 1 and w 1, respectively. Figure 2 Geometry and boundary conditions for the Couette flow model To apply this theory to FSW, take r 0 to be the radius of the tool pin and the set ω 0 Equal to the tool rotational speed. The outer cylinder radius r 1 is taken to be the radius of the tool pin plus the width of the third body region to a point in space where the material is solid and does not rotate, giving ω 1 = 0. M is the experimentally measured torque per unit depth of the tool pin. T a is the ambient temperature and q 0 is the heat generated by rotation of the tool. The material viscosity is found to be 2 2 ( 1 0 ) r r M µ = 4πrr ω ω ( ) 2 2 1 0 1 0 (1) The width of the weld region surrounding the tool pin is approximated using the model suggested by Arbegast and Braun (2000), which defines the weld region as the weld extrusion zone. The material viscosity is implemented into FLUENT as a constant viscosity.
Modelling of FSW for robotic implementation 103 3.2 Mechanical Model: Part 2 Seidel and Reynolds (2003) and Ulysse (2002) modelled the large plastic deformation involved in the FSW process by relating the deviatoric stress tensor to the strain-rate tensor. Though this model has been shown to correlate very well with experimental force data, it has not yet been shown that the model sufficiently correlates for High rotational Speed Friction Stir Welding (HS-FSW). With the Visco-Plastic fluid flow model, the Thermo-Mechanically Affected Zone (TMAZ) is assumed to be a rigid visco-plastic material where the flow stress depends on the strain-rate ( ε ) and temperature and is represented by an inverse hyperbolic-sine relation as follows: σ 1/ n 1 1 Z Q = = e sin h Z ε exp (2) α A RT where α, Q, A, n are material constants, R is the gas constant, T is the absolute temperature and Z is the Zener-Hollomon parameter (Sheppard and Jackson, 1997). The material constants were determined using standard compression tests. The material viscosity is approximated as σ e µ = (3) 3 ε Equation (3) is implemented into FLUENT as a user-defined function. 3.3 Governing equations The solver controls for the simulations were set to three-dimensional, segregated, laminar, implicit and steady incompressible flow. FLUENT uses this configuration to solve the conservation of mass, momentum (the Navier-Stokes equations) and energy equations. Gravitational and body forces are neglected as well as changes in potential energy. Heat transfer is assumed to obey Fourier s law of heat conduction. 3.4 Numerical model In this study, partial penetration FSW of AL 6061-T6 was considered. The sample/plate material is 3" long (76.2 mm), 2" (50.8 mm) wide and 0.25" (6.35 mm) thick. To reduce the size of the numerical model, the support table, located underneath the sample, is not included in the analysis. Therefore, heat transfer to the support table is ignored in this work. The sample/weld material model consists of 22,497 tetrahedron brick elements with 5152 nodes. The tool was assumed to be H-13 tool steel with constant density, specific heat and thermal conductivity and rotates counter-clockwise. The tool tilt angle was 2 and the depth was set to 0.145" (3.683 mm). The tool shoulder is flat with a 0.500" (12.7 mm) diameter and is 0.25" (6.35 mm) tall. To account for heat conduction from the tool/material interface up the tool, aft of the tool shoulder, a 1" (25.4 mm) diameter and 0.50" (12.7 mm) tall shank is included. The pin length is 0.1425" (3.62 mm) and the diameter is 0.190" (4.826 mm). Typically, the tool pin is threaded, usually with helical threads: left-handed threads for clockwise pin rotation and right-handed ones for counter-clockwise rotation. However, it was decided to model a smooth pin surface, as meshing the threads with our current FE mesher would have been very labourious. The tool consists of 14,300 tetrahedron brick elements with 3185 nodes (Figure 3). Figure 3 FE mesh of the welding model For convenience, the global reference frame is assigned coincidentally with the tip of the rotating tool. In other words, the tool rotates and the plate moves towards the tool. Therefore, at the flow domain inlet, the material incoming velocity is assigned the weld travel speed and an initial temperature of 27 C. The sample top, sides and bottom walls of the plate are also assigned the weld travel speed and an initial temperature of 27 C as well. The weld material exits through the flow domain outlet. The weld material, top, side and bottom walls of the plate are assigned temperature-dependent density, specific heat and thermal conductivity for Al6061-T6 and are detailed by Mills (2002). The tool pin bottom, sides, shoulder and shank are assigned a constant angular velocity. Currently, the VU FSW test bed is not instrumented for temperature measurement of the weld region. The increase in temperature is caused by frictional heat generated from the contact between the welding tool and the weld material. To account for heat generation, the model developed by Schmidt et al. (2004) which estimates the heat generation based on assumptions for different contact conditions for the combined tool pin bottom, sides and shoulder at the tool/material interface in FSW joints, is used in this work for specifying the temperature of the weld region. As the VU FSW test bed is instrumented for direct temperature measurement, it is believed that the accuracy of the forces and torque predictions for the Visco-Plastic mechanical model will improve. 4 Experimental and numerical results The primary requirements for an FSW robot is the ability to maintain contact pressure in the form of axial force, and be able to provide the welding torque (moment) necessary to stir the weld material.
104 R. Crawford et al. The axial force requirements are perceived to be the primary limitation to robotic FSW. With a precision heavyduty machine tool, such as a milling machine, this can be done by simply moving the rotating tool into the work piece until the desired axial force is obtained, and then initiating the traverse to make the weld. With the rigidity and precision of the machine tool, the axial force will be reasonably maintained over the extent of the weld. With a robotic implementation, one cannot assume such rigidity. Development of a working process model will greatly assist the ability of engineers to develop a good qualitative understanding of factors such as tool geometry, weld material and process parameters. Understanding these relationships is necessary to establish the optimum operating conditions for robotic FSW. It would be useful now to define the weld pitch (w p ), which is the ratio of rotational speed to travel speed and has units of revolutions per inch. The experimental and simulation results will be discussed relative to the effects owing to weld pitch variation. The weld pitch can be increased in one of two ways: 1 increasing the tool rotational speed or 2 decreasing the travel speed. Similarly, the weld pitch can be decreased by reducing tool rotational speed or increasing the travel speed or feed rate. 4.1 Axial force In Figures 4 6, it can be seen that during these experiments, the machine used for FSW can be called upon to deliver and consistently maintain an axial force of 1 12 kn depending on the tool dimensions and welding parameters. Figure 6 Axial force for experimental and simulations for TS = 27 ipm In Figures 4 6, we can see that both mechanical models correlate well with the experimentally measured axial force. Also evident is the trend of the axial force to decrease as the rotational speed is increased for a constant travel speed. By inspection of Figures 4 6, it can be seen that as the weld pitch increases, the simulation data for both models begin to converge with the experimental data. Figure 7 shows that for a constant rotational speed, as the travel speed is increased the axial force increases. The extent to which the increased rotational speed/decreased axial force and increased travel speed/increased axial force relationship holds true is not yet known. The understanding of this relationship is a key to widespread implementation of FSW capable robots. Figure 7 Axial force for experimental and simulations for RS = 1500 RPM Figure 4 Axial force for experimental and simulations for TS = 27 ipm Figure 5 Axial force for experimental and simulations for TS = 44.8 ipm In Figures 4 6, it is clear that the optimum operating parameters for robotic FSW will require high rotational speeds and low travel speeds. However, as the quite substantial axial force present during FSW is applied to the weld material, which becomes substantially more malleable the higher the rotational speed (or heat input rate) for a given travel speed, the weld material can be easily expelled from under the tool shoulder. There are two possible solutions that would allow a FSW capable robot to exploit the increased rotational speed/decreased force relationship and avoid the pitfalls of surface deformation caused by weld overheating. They are: 1 force feedback control and 2 a non-rotating shoulder FSW tool or floating shoulder tool as suggested by Talia and Chaudhuri (2004).
Modelling of FSW for robotic implementation 105 The floating shoulder tool has its practical applications; however, it offers a mechanical solution and is not a controls-based solution, and will not be discussed here. 4.2 Force feedback control Force feedback control would allow a FSW robot to adjust for cases where there is insufficient downward force caused by structural compliance of the robotic manipulator and also for high weld pitch parametric regimes where tool excess pressure will cause surface deformation in the form of excess flash. A force control scheme that has been successfully used for this application is shown in Figure 8 (Cook et al., 2004). At low weld pitches; however, the Couette flow model torque tends to deviate by a factor of 5 for a low weld pitch, but dramatically converges as weld pitch is increased. Figure 9 Welding torque for experimental and simulations for TS = 27 ipm Figure 8 Force feedback control implemented as an outer force control loop around the ordinary position control system of a robot manipulator (Cook et al. 2004) Figure 10 Welding torque for experimental and simulations for TS = 44.8 ipm An outer force control loop is closed around the inner position control loop of the robot manipulator as suggested by De Schutter and Van Brussel (1988). The programmed Z-axis position (with respect to the wrist frame) of the robot is modified as required to maintain the desired axial force set by the outer control loop. This approach is attractive, as it does not require access to the basic position control loop of the robot. Stability of this scheme will depend largely on the indentation characteristic of the rotating tool, as it acts against the plasticised weld zone material (Cook et al., 2004). Most force control schemes assume a linear elastic environment. However, in FSW the tool/work piece environment is non-linear, non-elastic and a function of the welding parameters, for example, tool rotation speed and travel speed. This has not been found to be a major problem, provided that the force control loop is made inactive during the start and stop portions of the weld. This is significant because the plunge force at weld start, for example, can rise initially to three to five times the weld value (Cook et al., 2004). 4.3 Welding torque From Figures 9 11, it can be seen that during these experiments, the machine used for FSW can be called upon to deliver and consistently maintain a torque of about 60 Nm, which depends greatly on the tool dimensions and welding parameters. By inspection of Figures 9 11, it can be seen that as the weld pitch increases, the simulation data for both models begin to converge with the experimental data. Figure 11 Welding torque for experimental and simulations for TS = 63.3 ipm One possible reason for this behaviour is that as the weld pitch increases, the weld material surrounding the tool experiences an increase in temperature, which leads to a decrease in yield strength of the weld material; thus, allowing the weld material to be more easily stirred. It should be noted that though the Couette flow model is for non-newtonian fluids, it does not factor temperature into the determination of the material viscosity as the Visco-Plastic flow model does. In general the torque follows the same trend as the axial force, where an increase in weld pitch has a corresponding decrease in torque. Similarly, a decrease in weld pitch has a corresponding increase in welding torque.
106 R. Crawford et al. 5 Conclusions and future work The experimental and numerical simulation data show that the increased rotational speed/decreased force relationship exists for rotational speeds ranging from 1500 to 4500 rpm, and for travel speeds from 11.4 to 63.3 ipm. At low weld pitches, the Couette flow model did not correlate as well with the experimental results as the Visco-Plastic flow model did. As the weld pitch increased, the experimental results and the Couette flow model began to converge. This implies that the Couette flow model is more predictive for very high weld pitches. Overall, the Visco-Plastic flow model was more accurate than the Couette flow model over the range of weld pitches. The Visco-Plastic flow model also converged with the experimental results as the weld pitch was increased. It was also observed that the present limitation to fully exploiting the increased rotational speed/decreased force relationship is the overheating and subsequent surface deformation of the weld material observed at high weld pitches. A possible solution to this barrier is the implementation of a force feedback control scheme. Force feedback control would allow an FSW capable robot to adjust for cases where there is insufficient downward force caused by structural compliance of the robotic manipulator and also for high weld pitch parametric regimes where tool excess pressure will cause surface deformation in the form of excess flash. The identification of the excess flash that can occur at high rotational speeds because of overheating of the weld material has clearly shown the need to instrument the VU FSW test bed to acquire temperature measurements. The axial force and torque continued to decrease as rotational speed was increased, showed that the upper bound for which the increased weld pitch decreased force and torque relationship holds true was not reached in this experiment. This implies that even higher rotational speeds can achieve a further decrease in axial force and torque during FSW. Acknowledgements Completion of this was made possible through support provided by an American Welding Society Fellowship Grant. Additional funding was provided by the NASA Space Grant Consortium of Tennessee. Dr. Author C. Nunes of the NASA Marshall Space Flight Center provided valuable expertise and guidance through private communication, which contributed to the completion of this work. References Arbegast, W. and Braun, G.F. (2000) Aluminum flow stress determinations using a gleeble system, Proceedings of 11th Annual Advanced Aerospace Materials and Processes Conference and Symposium, Seattle, WA, March. Cook, G.E., Crawford, R., Clark, D.E. and Strauss, A.M. (2004) Robotic friction stir welding, Industrial Robot, Vol. 31, No. 1, pp.55 63. De Schutter, J. and Van Brussel, H. (1988) Compliant robot motion, parts I II, International Journal of Robotics Research, Vol. 7, No. 4, pp.3 33. Mills, K.C. (2002) Recommended Values of Thermo-Physical Properties for Commercial Alloys, Cambridge, UK: Woodhead Publishing Ltd., pp.68 71. North, T.H., Bendzsak, G.J. and Smith, C. (2000) Material properties relevant to 3-D FSW modeling, Proceedings of the Second International Symposium on Friction Stir Welding, Gothenburg, Sweden, June. Sato, Y.S., Kokawa, H., Enomoto, M. and Jogan, S. (1999) Microstructural evolution of 6063 aluminum during friction-stir welding, Metallurgical and Material Transactions A, Vol. 30A, pp.2429 2437. Schmidt, H., Hattel, J. and Wert, J. (2004) An analytical model for the heat generation in friction stir welding, Modeling and Simulation in Materials Science and Engineering, Vol. 12, pp.143 157. Seidel, T.U. and Reynolds, A.P. (2003) Two-dimensional friction stir welding process model based on fluid mechanics, Science and Technology of Welding and Joining, Vol. 8, No. 3, pp.175 183. Sheppard, T. and Jackson, A. (1997) Constitutive equations for use in prediction of flow stress during extrusion of aluminum, Materials Science and Technology, Vol. 13, pp.203 209. Talia, G.E. and Chaudhuri, J. (2004) A Combined Experimental and Analytical Modeling Approach to Understanding Friction Stir Welding, Department of Mechanical Engineering Presentation, Wichita State University. Ulysse, P. (2002) Three-dimensional modeling of the friction stir-welding process, Machine Tools and Manufacturing, Vol. 42, pp.1549 1557. White, F. and Kudo, H.K. (1991) Viscous Fluid Flow, McGraw-Hill Publishing Company, p.112.