Combined Temperature and Force Control for Robotic Friction Stir … · 2017. 10. 20. · the...

Personal use of this material is permitted. Permission from ASME must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or

redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

Zinn - MANU-12-1357 – Page 1

COMBINED TEMPERATURE AND FORCE CONTROL FOR ROBOTIC FRICTION STIR WELDING

Axel Fehrenbacher1, Christopher B. Smith2, Neil A. Duffie1, Nicola J. Ferrier1,

Frank E. Pfefferkorn1, Michael R. Zinn1*

1Department of Mechanical Engineering, University of Wisconsin – Madison, USA

2Friction Stir Link, Inc., Brookfield, WI, USA

* corresponding author

e-mail: [email protected]

phone: 608/263-2893

ABSTRACT

Use of robotic friction stir welding (FSW) has gained in popularity as robotic systems can accommodate more complex part geometries while providing high applied tool forces required for proper weld formation. However, even the largest robotic FSW systems suffer from high compliance as compared to most custom engineered FSW machines or modified CNC mills. The increased compliance of robotic FSW systems can significantly alter the process dynamics such that control of traditional weld parameters, including plunge depth, is more difficult. To address this, closed-loop control of plunge force has been proposed and implemented on a number of systems. However, due to process parameter and condition variations commonly found in a production environment, force control can lead to oscillatory or unstable conditions and can, in extreme cases, cause the tool to plunge through the workpiece. To address the issues associated with robotic force control, the use of simultaneous tool interface temperature control has been proposed. In this paper, we describe the development and evaluation of a closed-loop control system for robotic friction stir welding that simultaneously controls plunge force and tool interface temperature by varying spindle speed and commanded vertical tool position. The controller was implemented on an industrial robotic FSW system. The system is equipped with a custom real-time wireless temperature measurement system and a force dynamometer. In support of controller development, a linear process model has been developed that captures the dynamic relations between the process inputs and outputs. Process validation identification experiments were performed and it was found that the interface temperature is affected by both spindle speed and commanded vertical tool position while axial force is affected primarily by commanded vertical tool position. The combined control system was shown to possess good command tracking and disturbance rejection characteristics. Axial force and interface temperature was successfully maintained during both thermal and geometric disturbances and thus weld quality can be maintained for a variety of conditions in which each control strategy applied independently could fail. Finally, it was shown through the use of the control process model, that the attainable closed-loop bandwidth is primarily limited by the inherent compliance of the robotic system, as compared to most custom engineered FSW machines, where instrumentation delay is the primary limiting factor. These limitations did not prevent the


implementation of the control system, but are merely observations that we were able to work around.

1. INTRODUCTION

Friction stir welding (FSW) was invented at The Welding Institute (TWI) in the UK in 1991 [1]. This relatively new, solid-state joining process differentiates itself from many other welding processes by not melting the workpiece. As a result, the joining process generates excellent joint properties, is energy efficient, environment friendly, and versatile.

The basic concept of FSW can be described as follows: a non-consumable rotating FSW tool with a specially designed shoulder and probe is pressed against the base metal surface, while a vertical downward force is applied (Figure 1). Due to friction between the rotating tool and the workpiece and plastic deformation of the workpiece, the temperature in the weld zone increases. The generated heat is usually not sufficient to melt the material, however, the workpiece is softened in the area around the probe and the deformation resistance (i.e., yield strength) of the base material decreases. The tool is traversed along the weld interface to mix the joining members in a forging action along the joining line to create a weld in the solid state. Friction stir welding results in intense plastic deformation and temperature increase in the weld zone, which leads to a significant microstructural evolution without typically causing phase changes [2,3].

Friction stir welding was initially applied to aluminum alloys but welding of other materials such as copper, titanium and magnesium alloys as well as steels and nickel alloys have been investigated [3]. Friction stir welding is also identified as a technology that can be used to join dissimilar alloys and metals. By maintaining the weld below the solidus temperature, minimal pre- and post-processing, excellent weld strength and ductility and environmentally friendly nature, the process enables cost reductions in many industrial applications and allows the joining of materials considered not weldable by fusion processes (e.g., highly alloyed 2XXX and 7XXX series aluminum). Friction stir welding has developed numerous potential applications in aerospace, automotive, railway, shipbuilding, construction and other areas [2,3].

Figure 1: Schematic of the FSW process.

2. MOTIVATION

During FSW, numerous parameters and conditions can vary that affect weld quality. Changing conditions include different mechanical and thermal properties of the workpiece (Figure 2 a) and changes in the thermal capacity of the workpiece (Figure 2 b). Furthermore, variations in the thickness of the workpiece (Figure 2 c) affect the relative position of the tool

Shoulder

Probe

FSW Tool

Trailing Edge

Leading Edge

Advancing Side

Retreating Side

Translation

Vertical force

Rotation


shoulder to the workpiece surface and can cause voids or extensive flash. The FSW process can be sensitive to these variations, which usually are present in a production environment; hence, there is a significant need to control the process to assure high quality. In addition, robotic FSW presents unique challenges not present in most custom engineered FSW machines. In particular, the inherent compliance of robotic manipulators makes it difficult to control plunge depth and often necessitates the use of force control or other strategies to maintain weld quality.

(a) (b) (c)

Figure 2: Schematics of (a) varying workpiece properties, (b) varying workpiece geometry and (c) varying workpiece thickness.

Workpiece variability, including changes in thermal constraints, material properties and geometry, as well as robotic manipulator compliance variability due to changes in robotic configuration, could be addressed through complex calculations beforehand, (i.e., an open-loop algorithm). However, this would only be effective if the variations (disturbances) to the FSW process could be well characterized a priori. This would add significant cost due to the metrology required for every workpiece and the need to calibrate the stiffness of the robotic manipulator over its complete workspace and would make the system inflexible. Workpiece variability can also be addressed through a closed-loop control strategy that allows flexible use under a wide range of conditions that do not have to be known beforehand.

Previous work showed that temperatures at the tool shoulder-workpiece interface can be measured in real-time and can be utilized as a feedback signal for closed-loop control of temperature, which can help maintain weld quality [4,5]. Cederqvist et al. [6] also implemented temperature control algorithms for the welding of copper canisters. A cascaded control algorithm with PI controllers was developed that determines the power input requirement and maintains the tool temperature within a process window. Ross and Sorensen [7] also employ a cascaded controller for temperature control for friction stir processing of low-carbon steel with the goal of increasing the tool life. Relationships between temperature and weld quality have been reported in the literature for FSW: Peel et al. found that weld properties are dominated by the heat input (temperature) in welding aluminum 5083 [8]. Gratecap et al. found a qualitative influence of weld temperatures on weld quality [9]. Simar et al. observed effects of the weld heat input (by varying the travel speed) on the microstructure and the mechanical properties of the weld [10].

Force control, during which the axial tool position (z-axis) is adjusted to maintain a constant axial welding force, has proven to be a successful technology. When developing a robotic FSW system, Smith [11] experienced problems with the robot’s inherent lack of stiffness. He found that welding with a force control algorithm produces higher quality welds than without force control. Von Strombeck et al. [12] also developed force control algorithms to overcome stiffness issues with robotic FSW. Zäh and Eireiner [13] proposed that constant welding conditions, i.e., a constant axial force, is required during welding to ensure a uniform weld quality. They implemented force controlled welding and presented welds with a constant axial force. Oakes et al. [14] developed a dynamic process model relating the travel speed, spindle

FSW ToolWeld

FSW Tool


speed and plunge depth to the axial force and implemented a force controller that tracks both constant and sinusoidal desired forces.

Both temperature and force control can help in maintaining the weld quality during the presence of inherent process disturbances. However, each control method has their limitations. One possible way to implement temperature control is to change the spindle speed (or travel speed) to adjust the energy input to the weld while welding at a constant plunge depth. This strategy would fail in situations where the workpiece thickness changes as illustrated in Figure 2 c because proper contact of the tool shoulder and the workpiece cannot be maintained. On the other hand, when welding under axial force control only, by adjusting the vertical tool position, the system might fail in certain situations. Welding over a workpiece of varying thermal capacitance affects the rate at which heat flows away from the weld zone, and can cause changes in stir zone temperature (Figure 2 a and b), affecting weld quality. In many cases, this change in temperature cannot be compensated for when welding solely under force control.

An example of a scenario in which simultaneous temperature and axial force control would be beneficial is when welding close to the edge of a workpiece or going around a corner (Figure 3). Heat generated by the FSW process is not able to conduct away from the weld zone as quickly as the tool approaches the corner (one direction of heat dissipation is removed), which causes the stir zone and region around it to overheat. Under force control alone, the resulting material softening can lead to the tool shoulder plunging well below the workpiece surface and creating unacceptable welds [15].

Figure 3: Schematic of welding around a workpiece corner.

For these reasons, it would be beneficial to combine the two control methods, so that during robotic friction stir welding both a certain temperature and a specified axial force is maintained. To the authors’ knowledge, such a combined control system has not been demonstrated on a robotic FSW system to date.

3. APPROACH

Closed-loop temperature and force control of robotic FSW requires real-time weld temperature and force measurements as well as a valid, calibrated process model from which the control design can be established. In our implementation, force measurements were provided by a 3-axis force dynamometer placed below the weld specimen while temperature measurements were acquired with a custom wireless tool-embedded temperature measurement system. The process model was developed through a set of characterization experiments. The following section details the approach in regards to each of these areas.

FSW Tool

Workpiece

Weld


3.1 Temperature Measurement

Ideally, weld zone temperatures are measured, which is not possible without significant effort. For this reason, another location, close to or on the boundary of the weld zone, must be measured. In this work, an approach is chosen that uses thermocouples embedded into the tool, together with a wireless system to transmit the data. This approach was originally reported in [4] but is repeated here for clarity.

In general, the FSW tool is made of a material of relatively low thermal diffusivity (e.g., highly alloyed tool steel), as compared to an aluminum workpiece material (which is friction stir welded most commonly). It is therefore desirable to place the thermocouples as close to the tool-workpiece interface as possible to minimize the time delay associated with heat flow through the tool. In this work, we are making use of through holes to enable direct contact of the tip of the thermocouples with the workpiece material - in this case an aluminum alloy that has a very high thermal diffusivity.

Two 0.8 mm diameter through holes were fabricated using electrical discharge machining (EDM) into the tool shank (Figure 4). One 7.1 mm deep hole exits on the shoulder, 3.4 mm from the outer edge of the shoulder. Another, 17 mm deep hole was made that exits on the side of the probe (location of thread of the tool), 1.2 mm from the bottom of the probe, in order to obtain temperatures further down in the weld. The two through holes are located at the same angular position.

The smallest possible off-the-shelf, sheathed, ungrounded, type K thermocouple was chosen to reduce the temperature response time (sheath diameter 0.25 mm, part no. TJ36-CAXL-010U by Omega Corp.). The two thermocouples were inserted into the through holes and secured with high temperature thermocouple cement (maximum service temperature 1426 °C). The thermocouple sheaths are in direct contact with the workpiece material during welding as there is no thermocouple cement between the tip of thermocouple assembly and workpiece material.

Since the tool is rotating at high speed, a wireless data transmission system is used to transmit the temperature measurements in real-time (i.e., without significant delays) to a stationary data acquisition (DAQ) and control system. Figure 5 provides a schematic of the overall wireless DAQ system, illustrating the main components. Figure 6 and Figure 7 show a photograph of the instrumented tool holder and a close-up view of the FSW tool with the embedded thermocouples, respectively. For the various spindle speeds used in this study (1700 rpm to 2900 rpm) and a sample rate of 333 Hz, the system can capture 7 to 12 temperature measurements per rotation of the tool (angular resolution of 31 to 52 degrees). More detailed information can be found in [4].

The instrumented tool holder has been successfully operated for a weld length of over 7 m using a single set of thermocouples without failure. Another, similar, tool has been used flawlessly for a weld length of over 22 m. A failure that would occur during a weld could be detected by additional algorithms. In that case the thermocouple or tool would have to be replaced.


Figure 4: Schematic of through hole locations for the thermocouples on the FSW tool (not to scale, section view). The thermocouples are exposed at

the tool-workpiece interface.

Figure 5: Schematic illustrating the main

components of the wireless DAQ system used for FSW.

Figure 6: Photograph of assembled instrumented tool holder for FSW.

Figure 7: Close-up view of FSW tool showing the exposed thermocouples at the shoulder-workpiece

and probe-workpiece interfaces.

3.2 Force Measurement

Welding forces were measured using a 3-axis force dynamometer (Kistler model 9265A), which is cooled during and in between experiments using a constant temperature bath at room temperature. The spindle torque is recorded through a serial interface (Modbus communication protocol) from the spindle motor servo drive (Control Techniques model FM-3DN). Custom data acquisition, logging and control software (developed using National Instruments LabVIEW™) is running on a laptop and communicates with the robot controller through a serial interface. Compared to the robot controller, the software on the laptop offers a powerful programming environment, fast processing, various interfaces to connect other devices, and real-time plotting, to name a few. Based on the measured temperature and axial force, the control software determines spindle speed and vertical tool position adjustments, which are sent to the robot. Custom code was added to Friction Stir Link’s StirWareTM software (programmed in ABB’s Rapid), which runs on the robot controller and receives and processes the adjustments. The setup is able to manipulate both spindle speed and vertical tool position adjustments simultaneously. The spindle speed and vertical tool position commands were limited to protect the experimental hardware from damage (1700 rpm to 2900 rpm and 7 mm to 18 mm, for spindle speed and tool position, respectively).

Force measurements are sampled at 2500 Hz. The sampling rate for the temperature measurements is 333 Hz. Spindle torque is available at approximately 5 Hz. Spindle speed and vertical tool position commands are sent at 20 Hz to the robot controller, where they are updated

RotationAxis

ProbeShoulder

37°44°

3.4 mm

1.2

mm

FSW Toolwith two

Thermocouples

R8Tool Holder

SignalConditioning Transmitter Receiver

StationaryDAQPower Supply

(Battery)

Rotating Assembly

Stationary MagnetNS

Hall EffectSensor


at approximately 10 Hz. The schematic of the signal flow of the measurement and control system is shown in Figure 8.

Figure 8: Signal flow schematic of experimental testbed.

3.3 Dynamic Process Model

The closed-loop controller developed in this work is model-based and, hence, a dynamic process model needs to be established. Other reasons for obtaining a process model are the ability to better understand the process dynamics, to get a basic insight into the physics governing the process and to help explain the limitations of the system. The process is represented by a multi-input-multi-output (MIMO) system including two inputs: the commanded spindle speed ω*tool(t) and the commanded vertical tool position z*tool(t) and two outputs: the interface temperature tint(t) and the axial force fz(t).

The closed-loop controller is designed in a way that cross-coupling effects are decoupled, i.e., nulled (see section 3.4). For the purpose of model development it is assumed at this point that no cross-coupling exists. This allows the development of two individual models. First, a model relating the commanded spindle speed to the interface temperature is introduced. Next, a model relating the commanded vertical tool position to the axial force is discussed. Finally, a combined MIMO model, using the two previous models with additional coupling terms, is presented.

Prior system identification work of the welding process indicated that a first order model with pure delay (gain Kp,T, time constant τp,T and delay ∆Tp,T) could be used to represent the dynamic relation between commanded spindle speed and measured shoulder interface temperature for FSW on a CNC mill [4]. The 1st order system model is given in equation (1) and (2) in the time and frequency domain (Laplace domain), respectively. In that work, the process model was identified using frequency domain techniques. While providing good process model and parameter identification, frequency domain identification can be time consuming. In this work, we rely on the earlier process model identification and focus on parameter identification,


assuming that the same process model type is also suitable for the robotic setup. As such, time domain identification, via examination of time history response to step inputs, is used to determine the model parameters that describe the relationship between commanded spindle speed and measured shoulder interface temperature. Equation (1) represents the system behavior in the time domain and equation (2) in transfer function notation in the Laplace domain.

The relation between commanded vertical tool position z*tool(t) and measured axial force fz(t) is modeled using a simple analytical, lumped parameter model of the system as illustrated in Figure 9. The single degree of freedom (DOF) model consists of one mass m, two springs with stiffness k and two dampers with damping factor c. The springs represent the flexibility of the robotic manipulator and the workpiece while the dampers add damping to the system behavior. Only displacements and forces perpendicular to the workpiece surface (i.e., parallel to the tool axis) are considered in this model. ztool(t) represents the actual vertical position of the tool. The differential equation that governs the physics of the simple model is given in equation (3). The force fz(t) is measured using a dynamometer and is related to the actual vertical tool position as shown in equation (4). A higher DOF model with more degrees of freedom could be chosen, but the described simple model is believed to capture the main dynamics of the system. The model is not intended to predict actual numerical values, but rather used to gather qualitative information about the system and to guide control system design.

Figure 9: Schematic of simple analytical 1‐DOF lumped parameter model

)()(2)(2)(2

2

tkztzdt

dckztz

dt

dctz

dt

dm tooltooltooltooltool

(3)

)()()( tkztzdt

dctf tooltoolz (4)

Using equations (3) and (4) and applying a Laplace transform yields the transfer function relating the measured axial force and the commanded vertical tool position:

)()()( ,,intint, TptoolTpTp TtKttttdt

d (1)

sT

Tp

Tp

tool

Tpes

K

s

sT,

1)(

)(

,

,int

(2)


kcsms

kcs

sZ

sF

tool

z

22

)(

)(

)(2

2

(5)

Equation (5) can also be expressed in terms of time domain parameters using a gain Kp,F, a damping ratio ζp, a natural frequency ωp,n and a time constant τp,z associated with a zero of the transfer function. These parameters can then easily be correlated with experimental data.

2,,

2

2,,

2

)1(

)(

)(

npnpp

zpFp

tool

z

ss

sK

sZ

sF

(6)

Step response tests have been performed previously by Zhao et al. to identify nonlinear process models for the use in axial force control via plunge depth adjustments [16]. We are also estimating the model parameters by examination of the time history response to step inputs. During some welds, oscillations of the tool and robot arm were observed at approximately 0.6 to 0.8 Hz due to the low stiffness of the robotic manipulator. The occurrence of these oscillations is dependent on the orientation of the robotic end effector, the heat input (low travel speeds are in favor of the oscillations) and other factors. The frequency of the oscillations was not correlated with the spindle speed or its harmonics and thus is assumed to be a resonance frequency of the system and was used to determine the undamped natural frequency of the model. We suspect that the observed oscillations reveal the resonance frequency of the system (1st mode) but this cannot be independently verified. The observed oscillations may not be the dominant mode in regards to our simplified model, i.e. may not be a mode aligned with the direction of the applied force. The simplified mechanical model used in this work is restricted to forces and displacements in the vertical direction only. We have chosen to use this assumption to estimate the mechanical system model parameters, as this results in a conservative estimate of the model resonance frequency (i.e. lower in frequency) and thus a controller designed based on this model will result in a more robust controller. Stiffness and mass properties vary as a function of the configuration of the robotic manipulator and were determined for the particular configuration used in this work.

Both inputs of the MIMO model are assumed to affect both outputs, resulting in a dynamic cross-coupling of the system as illustrated in the block diagram in Figure 10. The cross-coupling terms are introduced to capture the dynamic coupling between the thermal and mechanical process models. The coupling terms are cast as input-equivalent disturbances and are directly related to the controlled process outputs. For example, the measured force, Fz, was found to have an effect on the interface temperature, Tint, equivalent to a scaled change in spindle speed. Here it is likely that plunge force directly affects the torque applied to the spindle (acting as a disturbance torque to the spindle speed controller) which in turn results in a change in the spindle speed. In this case, the spindle speed closed-loop controller dynamics are significantly faster that the thermal process model dynamics. As such, the coupling between force and temperature can ignore the spindle speed controller dynamics.

The governing equations of the system shown in Figure 10, expressed in the Laplace domain, are shown in equations (7) and (8). The combined MIMO model contains the two individual models as described above and two cross-coupling gains Kp,Z and Kp,Ω.


)(stool sT Tpe ,

1,

,

s

K

Tp

Tp

)(int sT

)(sZ tool

)(sFzsT Fpe ,

ZpK ,

,pK

2,,

2

2,,

2

)1(

npnpp

zpFp

ss

sK

Figure 10: Block diagram of multi-input-multi-output dynamic process model with dynamic cross-coupling.

)()(1

)( ,,

,int

, sFKses

KsT zptool

sT

Tp

Tp Tp

(7)

)()(2

)1()( int,2

,,2

2,, , sTKsZess

sKsF zptool

sT

npnpp

zpFpz

Fp

(8)

The parameters of both the temperature and the force model were estimated empirically for the given setup on the robotic FSW system using step response tests and the System Identification Toolbox in MATLAB™. Assuming linearity enables the use of the principle of superposition for parameter estimation, i.e., one input was held constant while the other was varied in steps, observing both outputs. The estimated model parameters based on experiments are summarized in Table 1.

As noted above, the model structure for the temperature model was derived from [4], whereas the model structure for the force model was developed using the mechanical model discussed in this section. The model parameters were determined by performing step response experiments. Example plots of experimental data of those step response tests are illustrated in Figure 11, which shows sections of two welds: (a) Changing the spindle speed in a step from 1900 rpm to rpm and observing both outputs and (b) changing the vertical tool position from 9 mm to 10.5 mm and observing both outputs. By examining the plots, the first and second order model structure of the temperature and the force model, respectively, can be seen. The figure also shows that for our setup and parameter ranges, there is no significant effect of the spindle speed on the axial force, except a change in the amplitude of the axial force oscillations due to the tool rotation.

The process time delays are comparable to each other (400 ms and 300 ms) and are a result of communication delays between the control system and the robot, internal delays in the robot and delays resulting from the wireless transmission of the temperature measurements. It was found that the vertical tool position has a significant effect not only on the axial force but also on the interface temperature ( NrpmK p /5.0, ), i.e., a change of 1 N in axial force has

the same effect on the interface temperature as a variation in spindle speed by 0.5 rpm. On the other hand, the spindle speed affects the interface temperature, but the influence on the axial


force is negligible ( CmmK Zp /0, ). This may not be in agreement with previous research

[17,18], but is valid on the given welding machine, setup, workpiece alloy and input parameter range investigated here. Although no definite explanation can be named at this point, the flexibility of the robot manipulator is believed to be a reason for the insensitivity of the axial force with respect to the spindle speed. Unlike stiffer welding setups (e.g. those using c-frame gantry systems), where plunge depth is held constant by the stiff supporting structure, a robotic welding system does not maintain constant plunge depth but, instead, acts as a compressed spring. Variations in material height (e.g. due to the softening of the material during welding) do not have a significant effect on measured axial force because the resulting deflections are modest as compared to the effective compression of the robotic manipulators supporting structure. In order to help investigate this phenomenon further, a position sensing system could be added to measure the actual vertical tool position during welding.

Table 1: Estimated model parameters

Parameter Estimated Value Unit

Temperature

TpK , 0.04 °C/rpm

Tp , 0.04 s

TpT , 0.4 s

Cross-Coupling

ZpK , 0 mm/°C

,pK 0.5 rpm/N

Force

2,,

npFpK 500 N/mm

zp, 0.07 s

ζp 0.2 - ωp,n 0.7 Hz

FpT , 0.3 s

(a) (b)

Figure 11: Example welds used for model parameter estimation for (a) changing the spindle speed from 1900 rpm to 1700 rpm and (b) changing the vertical tool position from 9 mm to 10.5 mm.

50 50.5 51 51.5 52

500

520

540

Time [s]

Inte

rfac

eT

em

p.[C

]

50 50.5 51 51.5 521500

2000

2500

3000

3500

Time [s]

Axi

alF

orc

e[N

]

66.5 67 67.5 68 68.5520

530

540

550

Time [s]

Inte

rfa

ceT

em

p.[C

]

66.5 67 67.5 68 68.53000

3500

4000

4500

5000

Time [s]

Axi

alF

orc

e[N

]


Using the parameter estimates found above, the open-loop frequency response plots of the model are generated and are shown in Figure 12. The bandwidth of both processes is similar and both systems experience significant phase lags at higher frequencies due to the time delays. The peak in the magnitude of the force model is due to the underdamped characteristics of the system.

(a) (b) Figure 12: Frequency response plot of open-loop process model of (a) output interface temperature relative

to commanded spindle speed and (b) output axial force relative to commanded vertical tool position.

3.4 Closed-Loop Controller

Based on the open-loop process model found in section 3.3, a closed-loop controller is designed in the frequency domain. Two integral controllers are chosen for simplicity and to eliminate any steady-state errors. This choice is strongly suggested from the open-loop frequency response of the system model. With the significant time delays, which introduce large phase lags, there is little hope of achieving closed-loop bandwidths above the break (or resonant) frequencies. The use of the integral controllers, with cross-over frequencies placed below the break (or resonant) frequencies, provides good phase margin and sufficient gain stabilization (i.e., attenuation) of the resonant mode) to ensure a robust controller. Each controller is introduced independently assuming no cross-coupling. The temperature controller is discussed first, then the force controller. Finally, both controllers are combined including static decoupling.

Successful closed-loop control of the shoulder interface temperature has been previously implemented on a CNC mill [4]. An integral controller with gain Kc,T was used, which determines spindle speed adjustments. The same measurement and control system is transferred to the robotic FSW testbed in this work. Figure 13 shows the block diagram of the interface temperature control system, for which either the shoulder or probe interface temperature can be used as the feedback signal. The controller gain (20 rpm/°C) was determined utilizing the developed process model, resulting in a gain margin of 13 dB and a phase margin of 70 degrees and an estimated closed-loop bandwidth of 0.2 Hz. This results in a reasonable tradeoff between sufficient stability and good performance. Figure 14 shows magnitude and phase of the compensated open-loop system )(/)(int sEsT T and Figure 15 shows magnitude and phase

of the compensated closed-loop system )(/)( intint sTsT .

10-2

10-1

100

101

102

10-3

10-2

10-1

Frequency [Hz]

Mag

nitu

de [C

/rpm

]

Kp,T

(2 p,T)-1

10-2

10-1

100

101

102

-360

-270

-180

-90

0

Frequency [Hz]

Pha

se [

deg]

Phase lag due totime delay

10-2

10-1

100

101

102

102

103

Frequency [Hz]

Mag

nitu

de [

N/m

m]

Oscillatory ModeKp,F

p,n -2

p,nK

p,F p,z

2

10-2

10-1

100

101

102

-360

-270

-180

-90

0

Frequency [Hz]

Pha

se [

deg]

p,n

Phase lag due totime delay


)(stool sT Tpe ,

1,

,

s

K

Tp

Tp

)(int sT

)(sET)(int sT

s

K Tc,

Figure 13: Block diagram of closed-loop interface temperature control system adjusting spindle speed.

Figure 14: Simulated compensated open-loop frequency response of interface temperature control

system.

Figure 15: Simulated closed-loop frequency response of interface temperature control system.

Closed-loop control of axial force by vertical tool position adjustments has not been implemented on our previous testbed (CNC mill) due to system limitations (i.e., lack of ability to adjust the vertical tool position in real-time), but has been employed previously by other researchers [11–14]. Figure 16 shows the block diagram of the axial force control system implemented in this work. The same controller structure was used as in the temperature control system presented above. The controller gain (0.001 mm/N) was determined utilizing the developed process model, resulting in a gain margin of 7 dB, a phase margin of 84 degrees and an estimated closed-loop bandwidth of 0.1 Hz. Figure 17 shows magnitude and phase of the compensated open-loop system )(/)( sEsF Fz and Figure 18 shows magnitude and phase of the

compensated closed-loop system )(/)( sFsF zz .

)(sZ tool )(sFz

)(sEF)(sFz

s

K Fc,sT Fpe ,

2,,

2

2,,

2

)1(

npnpp

zpFp

ss

sK

Figure 16: Block diagram of closed-loop axial force control system with vertical tool position as control

input.

10-2

10-1

100

101

102

10-4

10-2

100

Frequency [Hz]

Mag

nitu

de [C

/ C

]

Gain Margin 13 dBC

10-2

10-1

100

101

102

-360

-270

-180

-90

0

Frequency [Hz]

Pha

se [

deg]

Phase Margin 70 deg

10-2

10-1

100

101

10-2

10-1

100

Frequency [Hz]

Mag

nitu

de [C

/ C

]

B

10-2

10-1

100

101

-360

-270

-180

-90

0

Frequency [Hz]

Pha

se [

deg]

B


Figure 17: Simulated compensated open-loop frequency response of axial force control system.

Figure 18: Simulated closed-loop frequency response of axial force control system.

In order to combine the interface temperature and axial force control systems, both controllers are implemented in parallel using the same gains as found previously. Because the two processes are dynamically cross-coupled, as shown in section 3.3, a decoupling structure is added, shown in the block diagram of the overall control system in Figure 19. As a consequence of placing the closed-loop bandwidth of the system below the process model break frequency, the controller only needs to decouple the static gain of the process coupling

model, simplifying the implementation. The static decoupling gains used are rpmmmK Zp /0,

and mmrpmK p /250,

. The former gain is chosen to be zero because no significant effect of

spindle speed on the axial force was observed. The latter decoupling gain improves the performance of the control system because the undesired effect of the vertical tool position on the interface temperature is nulled through this feed-forward element.

sT Tpe ,

1,

,

s

K

Tp

Tp

)(int sT

)(sFzsT Fpe ,

ZpK ,

,pK

)(stool

)(sZ tool

)(sET)(int sT

s

K Tc,

Controller

)(sEF)(sFz

s

K Fc,

Process

,pK

ZpK ,

Static Decoupling

2,,

2

2,,

2

)1(

npnpp

zpFp

ss

sK

Figure 19: Block diagram of combined closed-loop interface temperature and axial force control system using both spindle speed and vertical tool position as control inputs.

10-2

10-1

100

101

102

10-4

10-3

10-2

10-1

100

Frequency [Hz]

Mag

nitu

de [

N/N

]

Gain Margin 7 dB

C

10-2

10-1

100

101

102

-360

-270

-180

-90

0

Frequency [Hz]

Pha

se [

deg]

Phase Margin 84 deg

10-2

10-1

100

101

10-2

10-1

100

Frequency [Hz]

Mag

nitu

de [

N/N

]

B

10-2

10-1

100

101

-270

-180

-90

0

Frequency [Hz]

Pha

se [

deg]

B


4. EXPERIMENTAL PROCEDURE

Welding experiments were performed on a robotic FSW system (Friction Stir Link, Inc. RobostirTM system using an ABB IRB 7600 robot with an S4C+ robot controller). The spindle motor’s (Control Techniques model MH-8500) maximum power rating was 9.95 kW, the rated torque was 31.6 Nm and the maximum speed was 3000 rpm. On the given setup, a gear ratio of 1:2.83 was utilized between the motor and the spindle, resulting in a maximum spindle speed of 8500 rpm and maximum torque of 11.2 Nm. The tool travel angle was held constant at 3 degrees by using an angled fixture with the spindle orientated vertically. A FSW tool made of H13 tool steel with a concave shoulder and a threaded, conical probe with three flats was used. The tool shoulder diameter was 15 mm, the probe diameter was tapered from 7.0 mm to 5.2 mm and the probe length was 4.7 mm (measured from the outer edge of the shoulder). The tool rotation direction was always counterclockwise. The tools were machined in house on a 5-axis mill-turn center (Mori Seiki NT1000/W). 170 mm long bead-on-plate welding was performed on extruded 6.35 mm thick 6061-T6 aluminum workpieces (102 mm by 203 mm). All welds are partial penetration welds. An 8 mm thick low carbon steel backing plate was used under the workpieces unless otherwise noted.

5. RESULTS AND DISCUSSION

5.1 Process Dynamics

Figure 20 shows the measured interface temperatures at the shoulder and probe location and the measured forces during a weld under position control with a constant spindle speed of 2000 rpm, a constant travel speed of 300 mm/min and a step change in commanded vertical tool position from 10.5 mm to 11.0 mm in the first half of the weld traverse. The temperature measurement approach chosen in this work captures the dynamics of the process very well, because the thermocouple sheaths are in direct contact with the aluminum workpiece at the tool-workpiece interface. The measured temperatures are not constant, but rather oscillating as the tool traverses under constant operating conditions as seen in Figure 20. The frequency of these oscillations is found to match the frequency of the spindle rotation, i.e., the thermocouple is capturing temperature variations through 360 degrees of the tool rotation. For the weld shown in Figure 20, the amplitude of the shoulder and probe interface temperature is 3 °C and 6 °C, respectively, and the amplitude of the axial force is 750 N. More information about the temperature dynamics captured with this measurement strategy can be found in [19], and a calibration of the measurements is presented in [20]. The amplitudes of the oscillations observed in the three welding forces (also at the same frequency as the spindle rotation rate) are larger than those observed with the same workpiece/tool combination on a CNC mill setup [19], which is one indication of a higher flexibility of the robotic system compared to the stiffer CNC mill.


Figure 20: Measured interface temperatures and forces when welding 6.35 mm thick 6061-T6 with a constant

spindle speed of 2000 rpm, constant travel speed of 300 mm/min and a step in commanded vertical tool position from 10.5 mm to 11.0 mm (at 48 s).

The closed-loop control system is only enabled during the weld traverse, hence all subsequent figures will only show the weld traverse for simplicity.

5.2 Closed-Loop Control of Interface Temperature

In order to verify the control system’s performance, command tracking experiments were performed (Figure 21), in which the desired probe temperature was varied sinusoidally with a 10 °C amplitude and a frequency of 0.2 Hz. As mentioned in section 3.2, the spindle speed limits were 1700 rpm and 2900 rpm, and the limits of the tool position were 7 mm and 18 mm. It can be seen that the spindle speed is adjusted by the controller and the measured probe interface temperature also various sinusoidally at the same frequency. Because the system operates right at the closed-loop bandwidth, there is a significant phase lag of approximately 90 degrees. This matches the result from the simulation of 83 degrees (Figure 15). Due to the higher communication delays (400 ms) present on the robotic system (the delay on the CNC mill was found to be only 85 ms), the phase lag of the control system implemented on the CNC mill is only approximately 40 degrees (same workpiece and command trajectory). The closed-loop performance could be increased using more sophisticated controller structures if needed, but the performance of the given system is likely to be acceptable for many applications. The magnitude ratio of the measured temperature compared to the commanded temperature is approximately 0.8 on the robotic system (matching the result from the simulation of 0.74, Figure 15) and approximately 1.3 on the CNC mill. Disturbance rejection properties of the temperature control system have been demonstrated previously on the CNC mill and can be found in [4].

20 30 40 50 60 70 800

200

400

600

Time [s]

Inte

rfa

ceT

em

pe

ratu

res

[C]

Plu

nge

star

ts

Tra

vers

est

arts

Cha

nge

inve

rtic

al

tool

posi

tion

Too

lret

ract

s

Shoulder

Probe

6061 solidus

20 30 40 50 60 70 80

0

2

4

6

Time [s]

Fo

rce

s[k

N]

Lateral

Traverse

Axial


Figure 21: Command tracking of closed-loop interface temperature control system using a sinusoidal

command with 10 °C amplitude and frequency of 0.2 Hz. Constant commanded vertical tool position 10.5 mm. Constant travel speed of 360 mm/min. Feedback of probe interface temperature.

5.3 Closed-Loop Control of Axial Force

The performance of the axial force control system was evaluated and is shown in Figure 22 for command tracking of steps in desired force magnitude of 500 N. It can be seen that the vertical tool position is adjusted appropriately and that the system follows the commands well with no steady-state error, no overshoot and a closed-loop time constant of approximately 0.9 s.

Figure 22: Command tracking of closed-loop axial force control system using step commands of 500 N

magnitude. Constant spindle speed 2300 rpm. Constant travel speed of 360 mm/min.

In order to evaluate the disturbance rejection properties of the axial force control system, ramp disturbances were introduced into the system by adding a vertical offset to the programmed end point of the weld. In Figure 23 it can be seen that the axial forces increase or decrease significantly (positive and negative offset, respectively) when no axial force control is utilized. This can lead to a significant degradation in weld quality as seen in the photographs of the weld surfaces (Figure 24). When employing axial force control, the vertical tool position is adjusted by the controller, which leads to proper tool shoulder-workpiece contact, a constant

35 40 45 50 55

4

5

6

Time [s]

Axi

al F

orc

e [k

N]

Measured

Desired

35 40 45 50 55

10.5

11

11.5

Time [s]Ve

rtic

al T

oo

l Po

sitio

n [m

m]

F-c

ontr

olen

able

d Command


axial force and a substantial improvement in weld quality while encountering the ramp disturbance.

(a) (b)

Figure 23: Disturbance rejection of closed-loop axial force control system using an (a) 3 mm positive and (b) 5 mm negative ramp disturbance. Constant spindle speed 2300 rpm. Constant travel speed of 360 mm/min.

Plunge depth too deep

Plunge depth too shallow: Voids

(a) (b)

Figure 24: Photographs of weld surfaces related to Figure 23. Top: No control, bottom: With axial force control. (a) 3 mm positive ramp disturbance, (b) 5 mm negative ramp disturbance.

5.4 Combined Closed-Loop Control of Interface Temperature and Axial Force

The combined temperature and force control system is first evaluated in terms of command tracking (Figure 25). Initially a constant axial force of 4.5 kN is commanded with a weld travel speed of 300 mm/min. After approximately 20 seconds, the axial force command is decreased and increased in discrete steps over a 7 second interval. During the complete weld the desired interface temperature is held constant. The vertical tool position is automatically decreased, then increased to achieve the desired axial force. Changing the vertical tool position also affects the temperatures, so the controller increases and then decreases the spindle speed in order to compensate for the changed heat generation through different vertical tool positions. At the end of the weld, the desired interface temperature is increased in steps while the axial force command was held constant, which caused the controller to increase the spindle speed and hold the vertical tool position approximately constant (negligible effect of spindle speed on axial force).

30 35 40 45 50 55

4

6

8

Time [s]

Axi

al F

orc

e [k

N]

No Control

With ControlDesired

30 35 40 45 50 557

8

9

10

11

Ve

rtica

l To

ol P

osi

tion

Co

mm

an

d [m

m]

No Control

With Control

35 40 45 50 550

2

4

6

Time [s]

Axi

al F

orc

e [k

N]

No Control

With ControlDesired

35 40 45 50 5510

12

14

16

Ve

rtic

al T

ool

Po

sitio

nC

om

ma

nd

[mm

]

No Control

With Control


(a) (b)

Figure 25: Command tracking of combined closed-loop interface temperature and axial force control system. (a) Temperature part, (b) force part (same weld). Constant travel speed of 300 mm/min. Feedback of probe

interface temperature.

In order to further test the control system for simultaneous command tracking, both sinusoidal temperature and force commands were used (Figure 26). The desired probe interface temperature varied at 0.1 Hz with an amplitude of 15 °C and the desired axial force varied at 0.1 Hz with an amplitude of 500 N and a phase shift of 90 degrees. It can be seen that the combined control system performs well by simultaneously adjusting spindle speed and vertical tool position to track both the temperature and force commands. The phase lag for the temperature is approximately 60 degrees (44 degrees from the simulation, Figure 15) and about 50 degrees for the axial force (53 degrees from the simulation, Figure 18).

(a) (b)

Figure 26: Simultaneous command tracking of combined closed-loop interface temperature (a) and axial force (b) control system (same weld). 0.1 Hz sinusoidal commands for both interface temperature (15 °C amplitude) and axial force (500 N amplitude). Constant travel speed of 300 mm/min. Feedback of probe

interface temperature.

35 40 45 50 55 60 65520

540

560

580

600

620

Time [s]

Inte

rfa

ceT

em

pe

ratu

re [ C

]

Shoulder

Pin

Desired (Pin)

35 40 45 50 55 60 65

1800

2000

2200

2400

Time [s]

Sp

ind

le S

pe

ed

[rp

m]

T

-con

trol

enab

led

Command

35 40 45 50 55 60 65

3

4

5

Time [s]

Axi

al F

orc

e [k

N]

Measured

Desired

35 40 45 50 55 60 6510

12

14

Time [s]Ve

rtic

al T

oo

l Po

sitio

n [m

m]

F-c

ontr

olen

able

d

Command

45 50 55 60 65

520

540

560

580

600

Time [s]Inte

rfa

ce T

em

pe

ratu

re [

C]

Shoulder

Pin

Desired (Pin)

45 50 55 60 651500

2000

2500

3000

Time [s]

Sp

ind

le S

pe

ed

[rp

m]

Command

45 50 55 60 65

3

4

5

Time [s]

Axi

al F

orc

e [k

N]

Measured

Desired

45 50 55 60 659

10

11

Time [s]Ve

rtic

al T

oo

l Po

sitio

n [m

m]

Command


The combined temperature and force control system is also applied in a scenario with a ramp disturbance (5 mm negative ramp) as described in section 5.3 while simultaneously commanding a desired interface temperature (Figure 27). The vertical position of the tool is automatically adjusted throughout the weld in order to maintain a proper tool shoulder-workpiece contact, resulting in a constant axial force. Meanwhile, the desired interface temperature is changed in 10 °C steps, resulting in the controller adjusting the spindle speed to alter heat generation.

(a) (b)

Figure 27: Command tracking of temperature (a) and disturbance rejection of force (b) of combined closed-loop interface temperature and axial force control system (same weld). Step commands in interface

temperature of 10 °C magnitude. 5 mm negative ramp disturbance for axial force. Constant travel speed of 300 mm/min. Feedback of probe interface temperature.

A final scenario to evaluate the combined control system is the application of a step disturbance to the temperature by welding over two backing plates of very different thermal diffusivities as illustrated in Figure 27 and simultaneously applying a ramp disturbance to the force (3 mm negative ramp disturbance). To simulate a change in workpiece geometry or properties, welds were performed over a set of two different backing plates of significantly different thermal diffusivities. Titanium and copper with dimensions of 102 mm x 76 mm x 12.7 mm were selected for their significantly different thermal diffusivities but similar stiffness. Backing plates of various materials have also been utilized for thermal management during welding in order to improve the hardness of the resulting weld, particularly in the heat affected zone (HAZ) [21,22].

Figure 28: Schematic of different backing plates to affect thermal boundary conditions.

The desired probe interface temperature was held constant at 550 °C and the desired axial force was held constant at 4 kN. The results are shown in Figure 29 and it can be seen that when no temperature or force control is used, the interface temperature decreases by over

30 35 40 45 50 55 602

3

4

5

Time [s]

Axi

al F

orc

e [k

N]

Measured

Desired

30 35 40 45 50 55 6010

12

14

16

18

Time [s]Ve

rtic

al T

oo

l Po

sitio

n [m

m]

F-c

ontr

olen

able

d

Command


80 degrees over the copper backing plate as compared to the titanium backing plate due to the higher thermal diffusivity and hence heat loss out of the weld zone. The temperature starts to decrease approximately 2 s (i.e., 10 mm) before the backing plate interface is crossed, which indicates that the heat flow out of the weld zone is already affected by the copper backing plate at that location. The axial force decreases throughout the whole weld due to the weld end point intentionally programmed 3 mm above the workpiece. When employing the combined control system, the controller increases the spindle speed initially and increases the vertical tool position throughout the weld. This successfully maintains the desired temperature and force as compared to the scenario when no control is used and weld quality is substantially improved as seen in the weld surfaces in Figure 30. There is a brief drop in probe interface temperature just before the backing plate interface, which is due to the limited response characteristics of the control system (see section 5.5). The probe temperature is dropping by 27 °C, even when using the control system, however, the probe temperature is only outside a ± 10 °C window for 1.7 s. This brief drop is not believed to have any significant effect on the weld quality since the temperature stayed at all times within a temperature window previously determined to result in acceptable weld quality [5]. The chosen setup with the different backing plates also represents a rather extreme change of thermal boundary conditions, which is rarely found in practice. After crossing the backing plate interface, the commanded spindle speed continues to rise and then levels off, because most heat is being conducted from the weld zone into the backing plate at the center of the copper backing plate. Towards the beginning and end of the copper backing plate, there is less heat conducted out of the weld zone due to the material boundaries to titanium and air with much lower thermal diffusivities than copper.

(a) (b)

Figure 29: Simultaneous disturbance rejection of combined closed-loop interface temperature (a) and axial force (b) control system (same weld). Welding over different backing plates (step disturbance for

temperature) and 3 mm negative ramp disturbance for axial force. Constant travel speed of 300 mm/min. Feedback of probe interface temperature.

40 45 50 55 601

2

3

4

5

Time [s]

Axi

al F

orc

e [k

N]

Ti Cu

No Control

With Control

Desired

40 45 50 55 6010

11

12

13

14

Ve

rtic

al T

oo

l Po

sitio

nC

om

ma

nd

[mm

]

Ti CuNo Control

With Control


Plunge depth too shallow and temperature too low: Voids

(a) (b)

Figure 30: Weld surfaces when welding over different backing plates and applying a 3 mm negative ramp disturbance for axial force. (a) No control, (b) with combined temperature and force control.

5.5 Limitations

The performance of the robotic FSW closed-loop temperature and force control is directly related to the achievable closed-loop bandwidth. The higher the bandwidth, the faster the closed-loop system can respond to commands or reject disturbances. As such, it is important to understand which system characteristics are responsible for the limitations in control bandwidth. The major limitations on the given system are time delays and the low stiffness of the robotic manipulator.

Time delays are caused by communication delays between the laptop that runs the control laws and the robot controller, internal delays in the robotic FSW system (e.g., communication to spindle motor drive) and transmission delays of the measured temperatures. These delays are inherent to the given system and could be slightly reduced if needed by employing more advanced hardware and software. We are not attempting to modify the process dynamics, our goal is rather to reject disturbances and model uncertainties. For this reason, an integral controller were chosen – where the integral gain was selected to achieve a cross-over frequency below the process break frequency. A proportional term could have been included, but only when using a modest gain due to the delay. The closed-loop bandwidth must very likely remain below the process break frequency to maintain stability with a high level of control robustness (against model parameter variation). In the interest of simplicity, only the integral term was kept. In order to better cope with these delays, an advanced controller structure could be implemented, e.g., a Smith predictor (assuming the delay does not change significantly). The effect of the time delay on the open-loop process model relating commanded vertical tool position to axial force is illustrated in the frequency response plot in Figure 31 (a). It can be seen that the magnitude is not affected by the delay, but that a frequency dependent phase lag is added. A time delay always reduces the system stability and limits the achievable bandwidth of a system.

This approach is initial work in establishing combined temperature and force control on robotic friction stir welding systems. The authors believe that the general model structure should be appropriate for most robotic systems or other systems with a lack of compliance. As in many process control applications, some form of model parameter identification would be required for other test beds with alternate FSW machines, tooling, fixtures, workpieces, etc., either through explicit experiments or indirectly through tuning of control gains. As such, the structure and approach are general to these types of systems with some tuning required.

The industrial robot used in this work is designed for a variety of manufacturing tasks, most of them require higher tool center point (TCP) velocities and lower loads as compared to most FSW applications. The inherent robot flexibility causes deflections in the serial robot linkages during FSW, resulting in position errors. E.g., in order to achieve a plunge depth of approximately 5 mm, the tool position must be commanded to be approximately 11 mm below the workpiece surface. The robot’s compliance was estimated to be in the order of 1 mm/kN, as


compared to the CNC mill which has a compliance of 0.05 mm/kN [23]. Figure 31 (b) shows the effect of varying stiffness on the open-loop process model relating commanded vertical tool position to axial force. It can be seen that for a higher stiffness, the natural frequency of the robotic manipulator increases. As a result, the phase lag associated with the resonance is pushed to a higher frequency and the maximum possible stable closed-loop bandwidth is increased.

(a) (b) Figure 31: Frequency response plot of open-loop process model of axial force relating to commanded

vertical tool position when (a) varying the time delay and (b) varying the stiffness.

In order to further examine the effect of time delays and compliance, both delay and stiffness were varied and the maximum achievable crossover frequency was computed such that the closed-loop system maintains stability. Figure 32 plots the maximum achievable crossover frequency as a function of time delay and natural frequency. As shown in Figure 31, the natural frequency is directly related to stiffness. It can be seen that the time delay in combination with the natural frequency (stiffness) affects the maximum crossover frequency and hence closed-loop bandwidth and system performance. For low natural frequencies (as experienced on the robotic system), the low stiffness is the dominating limitation and the time delay only plays a minor role. On setups with a high stiffness (such as a CNC mill), the time delay is the mayor limitation. For systems with low delays the stiffness has an important effect on the performance whereas for higher delays the delay becomes dominant. On the current setup (shown in Figure 32 in the lower right hand corner), both the delay and the compliance should be reduced in order to improve the performance of the control system.

10-2

10-1

100

101

102

40

100200

50010002000

Frequency [Hz]

Ma

gn

itude

[N/m

m]

10-2

10-1

100

101

102

-360

-270

-180

-90

0

Frequency [Hz]

Ph

ase

[de

g]

Increasing delayNo delay

Delay = 0.0833s

Delay = 0.1667s

Delay = 0.25s

Delay = 0.333s

10-2

10-1

100

101

102

40100

1000

10000

Frequency [Hz]

Ma

gn

itude

[N/m

m]

Increasing stiffness

10-2

10-1

100

101

102

-360

-270

-180

-90

0

Frequency [Hz]P

ha

se [d

eg

]

Increasing stiffnessStiffness = 0.5*k

Stiffness = 1*k

Stiffness = 2*k

Stiffness = 5*k

Stiffness = 10*k


Figure 32: Maximum achievable crossover frequency (Hz) when varying the stiffness (i.e., natural frequency)

and time delay.

Loop update rates are also a limitation. In our system, internal interrupts in the robot controller and bus communication to the spindle motor servo drive are executed at approximately 10 Hz and are not able to run faster because communication to external devices has lower priorities than motion and safety tasks in the given robot control architecture. This time discretization contributes additional equivalent delay equal to approximately one half of the effective sample rate. In addition, the response of the actuators (spindle motor and robotic arm drive system) is limited, as well as the rate of change of the actuators due to restricted power being available to these devices. The outputs of the actuators are also limited to a certain range based on their design specifications.

6. SUMMARY

We have demonstrated the successful development and evaluation of a closed-loop control system for robotic friction stir welding (FSW) that controls plunge force and tool interface temperature by varying spindle speed and commanded vertical tool position. The simultaneous control of temperature and plunge force is necessitated by the unique characteristics of robotic friction stir welding – primarily the significant increase in compliance as compared to most custom engineered FSW machines. The system developed was implemented on an industrial robotic FSW system equipped with a custom real-time wireless temperature measurement system and plunge force sensor. The combined control system was shown to possess good command tracking and disturbance rejection characteristics over a range of operating scenarios. It was demonstrated, through the use of the control process model described here, that the attainable closed-loop bandwidth of the robotic system is primarily limited by the inherent compliance of the robotic system, as compared to most custom engineered FSW machines, where instrumentation delay is the primary limiting factor. The developed system was successfully implemented albeit the described limitations and is believed to have many potential applications for production in industrial settings.

7. ACKNOWLEDGMENTS

Partial support of this work by the Department of Mechanical Engineering and the College of Engineering at the University of Wisconsin - Madison, the Wisconsin Alumni Research Foundation Technology Development RA, a Wisconsin Innovation & Economic

0 0.05 0.1 0.15 0.2 0.251

1.5

2

2.5

3

3.5

4

Time Delay [s]

Na

tura

l Fre

qu

ency

[Hz]

Maximum Crossover Frequency [Hz]

1

1.5

2

2.5

3

3.5

4

CNC mill

Robot usedin this work


Development Research Program (IEDR) and the U.S. National Science Foundation under contract CMMI-0824879 is gratefully acknowledged. The Mori Seiki NT1000/W is an equipment loan from the Machine Tool Technology Research Foundation. The authors would like to thank Lee Cerveny of Friction Stir Link, Inc. and Edward Cole, Joshua Schmale, Madhu Vadali, Eric Meunier and Justin Morrow at the University of Wisconsin - Madison and Amir Assi and Ingrid El Helou at the American University of Beirut (Lebanon) for their help.

8. NOMENCLATURE

CNC Computer numerical control Cu Copper DAQ Data acquisition DOF Degree of freedom EDM Electrical discharge machining FSW Friction stir welding MIMO Multi input multi output TCP Tool center point Ti Titanium c Damping constant [N/(m/s)] EF Axial force error [N] ET Temperature error [°C] fz, Fz Measured axial force [N] F*z Desired axial force [N] k Stiffness constant [N/m] Kc,F Force controller gain [mm/N] Kc,T Temperature controller gain [rpm/°C]

Kp,F Process gain (force) [N(rad/s)2/mm] Kp,T Process gain (temperature) [°C/rpm] Kp,Z Cross-coupling gain [mm/°C] Kp,Ω Cross-coupling gain [rpm/N] m Mass [kg] tint, Tint Measured interface temperature [°C] T*int Desired interface temperature [°C] ztool ,ztool Actual vertical tool position [mm] z*tool ,z*tool Commanded vertical tool position [mm] ∆Tp,F Process delay (force) [s] ∆Tp,T Process delay (temperature) [s] ζp Damping ratio [-]

p,z Time constant (zero, force) [s]

p,T Time constant (pole, temperature) [s] ωB Closed-loop bandwidth [Hz] ωC Crossover frequency [Hz] ωp,n Natural frequency [rad/s] ω*tool, Ω*tool Commanded spindle speed [rpm]


9. REFERENCES

[1] Thomas W. M., Nicholas E. D., Needham J. C., Murch M. G., Temple-Smith P., and Dawes C. J., 1991, “Friction Stir Butt Welding,” GB patent no. 9125978.8.

[2] Mishra R. S., and Ma Z. Y., 2005, “Friction Stir Welding and Processing,” Materials Science and Engineering, R 50, pp. 1–78.

[3] Threadgill P. L., Leonard A. J., Shercliff H. R., and Withers P. J., 2009, “Friction Stir Welding of Aluminium Alloys,” International Materials Reviews, 54(2), pp. 49–93.

[4] Fehrenbacher A., Duffie N. A., Ferrier N. J., Pfefferkorn F. E., and Zinn M. R., 2011, “Toward Automation of Friction Stir Welding Through Temperature Measurement and Closed-Loop Control,” J. Manuf. Sci. Eng., 133(5), pp. 051008–12.

[5] Fehrenbacher A., Duffie N., Ferrier N., Pfefferkorn F., and Zinn M., 2012, “Closed-Loop Control of Temperature in Friction Stir Welding and its Effect on Weld Quality,” Paper in preparation.

[6] Cederqvist L., Garpinger O., Hagglund T., and Robertsson A., 2012, “Cascade control of the friction stir welding process to seal canisters for spent nuclear fuel,” Control Engineering Practice, 20(1), pp. 35–48.

[7] Ross K., and Sorensen C., 2011, “Investigation of Methods to Control Friction Stir Weld Power with Spindle Speed Changes,” Friction Stir Welding and Processing VI (TMS Annual Meeting & Exhibition).

[8] Peel M., Steuwer A., Preuss M., and Withers P. J., 2003, “Microstructure, Mechanical Properties and Residual Stresses as a Function of Welding Speed in Aluminum AA5083 Friction Stir Welds,” Acta Materialia, 51, pp. 4791–4801.

[9] Gratecap F., Racineux G., and Marya S., 2008, “A Simple Methodology to Define Conical Tool Geometry and Welding Parameters in Friction Stir Welding,” 7th International Friction Stir Welding Symposium, Awaji Island, Japan, TWI, Published on CD.

[10] Simar A., Brechet Y., de Meester B., Denquin A., and Pardoen T., 2008, “Microstructure, local and global mechanical properties of friction stir welds in aluminium alloy 6005A-T6,” Materials Science and Engineering A, 486(1-2), pp. 85–95.

[11] Smith C. B., 2000, “Robotic Friction Stir Welding Using a Standard Industrial Robot,” 2nd International Friction Stir Welding Symposium, Gothenburg, Sweden, TWI, Published on CD.

[12] Von Strombeck A., Schilling C., and dos Santos J. F., 2000, “Robotic Friction Stir Welding - Tool Technology and Applications,” 2nd International Friction Stir Welding Symposium, Gothenburg, Sweden, TWI, Published on CD.

[13] Zäh M. F., and Eireiner D., 2004, “Friction Stir Welding Using NC Milling Machines,” Welding and Cutting, 3(4), pp. 220–223.

[14] Oakes T., and Landers R. G., 2009, “Design and Implementation of a General Tracking Controller for Friction Stir Welding Processes,” IEEE, Piscataway, NJ, USA, pp. 5576–81.

[15] Smith C. B., and Hinrichs J. F., 2004, “Personal communication, Friction Stir Link, Inc., Waukesha, WI.”

[16] Zhao X., Kalya P., Landers R. G., and Krishnamurthy K., 2009, “Empirical Dynamic Modeling of Friction Stir Welding Processes,” Journal of Manufacturing Science and Engineering, 131(2), p. 021001 (9 pp.).

[17] Melendez M., Tang W., Schmidt C., McClure J. C., Nunes A. C., and Murr L. E., 2003, “Tool forces developed during friction stir welding,” UTEP, USC, NASA Marshall.

[18] Cole E. G., 2009, “Investigation of Weld Material and Process parameter Influence of Forge Force in Friction Stir Welding,” MS Thesis, University of Wisconsin-Madison, Madison, Wisconsin.


[19] Fehrenbacher A., Schmale J., Zinn M., and Pfefferkorn F., 2012, “Measurement of Tool-Workpiece Interface Temperature Distribution in Friction Stir Welding,” Paper in preparation.

[20] Schmale J., Fehrenbacher A., and Pfefferkorn F., 2012, “Modeling and Calibration of Friction Stir Welding Temperature Sensors for Measurement of Tool-Workpiece Interface,” Paper in preparation.

[21] Reynolds A. P., and Tang W., 2012, “Thermal Management for Production of Very High Strength alloy 7050 Friction Stir Welds,” 9th International Friction Stir Welding Symposium, Huntsville, AL, USA, TWI, Published on CD.

[22] Upadhyay P., and Reynolds A. P., 2012, “Effect of backing plate thermal property on friction stir welding of 1” thick AA6061,” 9th International Friction Stir Welding Symposium, Huntsville, AL, USA, TWI, Published on CD.

[23] Shultz E. F., Cole E. G., Smith C. B., Zinn M. R., Ferrier N. J., and Pfefferkorn F. E., 2010, “Effect of Compliance and Travel Angle on Friction Stir Welding with Gaps,” Journal of Manufacturing Science and Engineering, 132(4), p. 041010 (9 pp.).

Date post:	13-Oct-2020
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times

Combined Temperature and Force Control for Robotic Friction Stir … · 2017. 10. 20. · the...

Documents