2014 UKSim-AMSS 8th European Modelling Symposium

Productivity Increase through Joint Space Path Planning for Robot Machining

Agus Atmosudiro, Matthias Keinert, Ali Karim, Armin Lechler, Alexander Verl

Akos Csizar University of Stuttgart Graduate School of Excellence advanced Manufacturing Engineering Stuttgart, Germany [email protected]

Department of Mechanical Engineering Institute for Control Engineering of Machine Tools and Manufacturing Units Stuttgart, Germany [email protected]

bigger workspace compared to standard machine tools. However, the accuracy and productivity of robot machining are lower compared to the attainable outcome of machine tools [2], [3]. Researches in the area of robot machining have been focusing mostly on methods to improve the machining accuracy. This paper on the other hand attempts to enhance the productivity of robot machining and presents the challenges to maintain the accuracy already attainable to date.

Abstract—Machine tools realize tool movements with high accuracy mainly due to highly developed computerized numerical controls (CNCs). As articulated industrial robots are used more and more for machining, robot controllers (RC) have to be equipped with additional path planning capabilities, similar to machine tools. A RC is very similar to a CNC from a software and hardware point of view, but with one major difference, the RC has an additional transformation stage, the transformation from Cartesian space to joint space. Machining with robots is a field intensely researched in the last years. CNC systems for robots are commercially available, furthermore, more and more CAM systems have extensions for machining with robots. Most of these offer a simulation of the machining process using a robot model, in order to solve the inverse kinematic problem and, additionally, to take into consideration axis motion limits (maximum angular amplitudes) and singularities. Moreover, path planning for machining robots is done in exactly the same way as for machine tools, with the mentioned additional transformation stage. This paper describes the advantages and challenges which result from the integration of the kinematic transformation in the path planning stage. Keywords-Robot machining; planning; CNC; CAD-CAM

I.

joint

space-based

II.

path

INTRODUCTION

The path planning of any modern machine tool is accomplished by the Computerized Numerical Control (CNC) which is responsible for two main tasks: synchronizing the motions of the axis and securing the Tool Center Point (TCP) to precisely follow a programmed contour. The CNC software interprets the G-code generated by Computer Aided Manufacturing (CAM) systems and calculates the trajectory reference values (positions, velocities and accelerations with regard to axes dynamics) for machining the desired parts. Even though machine tools offer high accuracy and good surface quality, there is a necessity for more cost-effective solutions in the case of large work pieces with high manufacturing tolerances [1]. Both industry and academia are actively pursuing the use of articulated industrial robots, hereafter referred to industrial robots, for machining due to the low investment costs and 978-1-4799-7412-2/14 $31.00 © 2014 IEEE DOI 10.1109/EMS.2014.46

RELATED WORKS

Currently, average modern CNC machine tools have an accuracy of up to ±1μm [3], [4], [5]. As opposed to this, average industrial robots have a positioning accuracy after calibration of about ±0.1mm [6], i.e. by factor 100 worse than with machine tools. The inaccuracy is even higher in the case of robot machining. Since robot structures, due to low stiffness, are not well-suited for machining applications, the currently achievable machining accuracy is ±1mm, by factor 1000 worse than machine tools. Furthermore, this ±1mm accuracy is not even constant throughout the workspace, contrary to machine tools [ 5 ] , [ 6 ] , [ 7 ] , [ 8 ] . All these differences can be attributed to the serially linked rotational axes, which drive the tool attached to the flange of the robot. Many research papers address this issue, and try to come up with either mechanical, or software solutions to improve the accuracy. Model-based calibration and compensation are already state of the art [9], [10], and commercial gauges are used to increase repeatability and accuracy. In this case, especially errors caused by thermal drift are compensated [9]. Mechanical changes usually mean parallelizing the kinematic chain (i.e. using parallel structures) to enhance stiffness and thus reduce the positioning errors. These hybrid designs [ 5 ] , however, reduce the workspace of the robot considerably [11]. Another approach to increase the positioning accuracy is to use sophisticated path planning methods as applied in CNC systems. The architecture of a robot controller (RC) is very similar to a CNC, with one major difference: the kinematic transformations are nonlinear and need far more computational power. Furthermore, in the case of a robot, the 257

III.

kinematic transformations include more than one solution. The mathematical operations that describe the transformation from Cartesian space to joint space are very different in the case of a 6 DOF (degrees of freedom) robot with rotational axes than in the case of a 3 DOF machine tool with translation axes, or in the case of a 5 DOF machine tool with rotational and translational axes. Currently developed CNC systems (and the theoretical background they are based on) can only function in Cartesian space. For machining with robots, the same general architecture (including the CAM software) is used as for high accuracy machine tools, with the final Cartesian to joint space transformation added at the end. This approach is a direct adaptation of the classical CNC to a robot structure, hereafter referred as robot CNC. Regarding the toolchain which is responsible for a CAD model becoming an actual part [12], there is no major difference in case of using robot CNC. Here, the CAD model is imported into a CAM tool, where all the machining operations are defined, exactly as they would be for a machine tool. However, a robot machining module of the CAM software simulates the machining by solving the inverse kinematics problem for the discrete milling paths, and monitors any breaches in axes limits or any possible singular poses of the robot. After a successful simulation, a G-code is generated, which defines the path of the tool, not the path of the robot axis. The G-code is then transferred to the RC, where the CNC module executes the code (please note that, depending on the manufacturer, robotprogramming languages can also be generated and executed, but best results have been obtained using a dedicated CNC module, an optional software extension to the RC). The CNC module operates exactly the same way as it would operate on the machine tool, with the difference, that it is constantly polling the RC to solve the inverse kinematic problem for each point. This task has to be performed not just for the current TCP coordinates, but also for the new points in the look-ahead horizon, which characterizes the ability for planning the trajectory in advance. This requires computation time and is one of the reasons for which robot CNCs have a narrower look-ahead horizon than CNCs for machine tools. In addition, the trajectories, which the interpolator plans, are smooth in the Cartesian space. On the other hand, due to the nonlinear transformation between Cartesian space and the joint space, they are not always smooth in the joint space. Path planning in joint space would allow a reduction of the computational burden on the RC by transferring all computation required for the kinematic transformation to an offline phase. The computational power now available can be used for achieving reduced cycle times and/or increasing the look-ahead horizon. In most cases, increasing look-ahead horizon leads to increased machining velocity [13], [14]. In the following chapter, this paper addresses a conceptual solution for a continuously joint space-based CAM-NC tool chain, in order to improve the exploitation of the available performance of robots in machining operations. The advantages and challenges of such an approach are presented in detail followed by simple experiments that serve as an early validation of the concept.

CONCEPT OF A JOINT SPACE-BASED CNC FOR ROBOT MACHINING

As outlined in section II, both industry and academia are trying to improve the accuracy of industrial robots in terms of using them for milling tasks. Here, one approach concentrates on path planning issues by using CNC systems for controlling the TCP of the robot. In ideal poses, this approach allows a higher positioning accuracy up to 0.08mm as well as an enhancement of surface quality [15]. Moreover, it can process G-codes according to ISO 6983 and provides standard CNC functionalities, such as tool radius compensation and smoothening. However, CNC systems fulfill their task in Cartesian space. Considering the kinematic structure of an industrial robot, control systems based on CNC techniques have to expand their path planning strategy by using an additional transformation stage including all additional costs and efforts that are connected to this approach. Hence, as an industrial robot is a joint space system by nature, path planning continuously executed in joint space seems to be preferable. The following section describes a holistic approach of a CAM- CNC tool chain for joint space path planning. A. Overall Concept The approach for a joint space path planning CNC is based on the idea that the CAD-CAM tool chain generates a G-code holding machining operations that define already joint space waypoints for each axis of an industrial robot (see Figure 1. For this purpose, the CAM tool has to solve the robot-specific inverse kinematic problem. As CAM tools do not necessarily support this kind of G-code generation, it is suggested to use a separate post-processing application that analyzes the given G-code generated by a CAM tool. and transforms Cartesian waypoints into joint space. In this way, independently of how the G-code has been generated (by a CAM tool or by an additional post processing application), the output clearly defines the motion of each robot joint. These motion references are then executed by the CNC system. All further adaptions conducted by the CNC system, e.g. tool radius compensation and smoothing, have to be done in joint space. Thus, the drives, which execute the appropriate axis movements, receive their reference values without further transformations. The advantage is that computing time required for inverse kinematic calculation is no longer necessary. B. Generation of Waypoints by CAM Tools The generation of waypoints by a CAM system for joint space path planning has to meet the following requirements: Conducting path planning in Cartesian space allows subsequent adaptations of axis positions, since only the TCP position and two orientations do not fully define the inverse kinematics problem (for machining the orientation around the tool axis has no importance). In joint space path planning, however, the motion of each axis is already unambiguously defined during CAM processing (or postprocessing). A sub-sequent change is not possible. Hence, the generation of axes reference values has to be performed considering the workspace limits of the robot and the

258

implementation of these functions, in order to plan milling paths in joint space and to fully benefit from a joint spacebased CNC for robots.

dynamic capabilities of each axis. This implies the consideration of the inverse kinematics of the robot and all dynamic constraints related to the axes already in the CAM processing phase.

D. Generation of Constant TCP Feedrate The Cartesian velocity of the milling tool has to be kept constant in order to assure a good surface quality of the workpiece. In joint space planning, assuring a constant Cartesian velocity is not at all a trivial task. Nevertheless, there are many examples for manufacturing, where the quality of the surface is not of utmost importance and these usually coincide with applications where manufacturing tolerances are relatively high. These are the best use-cases for machining with robots. The problem of a constant Cartesian velocity is not disregarded, but a perfectly constant milling velocity is not the most important aspect of this research. Because of the spindle, the 6 DOF industrial robots in milling applications have an infinite number of solutions for their inverse kinematics problem. The solutions for the inverse kinematics problem have an effect on how continuous the Cartesian velocity will be. Using an optimization method, the inverse kinematics problem is to be solved in a way, which minimizes the variations of the Cartesian velocity. In other words, the solution of the inverse kinematics problem leading to the smallest change in the Cartesian velocity is selected.

Figure 1. Conceptual Design of a Joint Space-based CNC Toolchain.

IV.

Furthermore, the accuracy of the robot regarding its position within its workspace has to be considered when generating the waypoints. CAM tools generate waypoints based on an assumed constant accuracy throughout the workspace of the machine tool. Robots do not have a constant accuracy throughout their workspace. The correlation of waypoints with machine accuracy is called “compression” in the case of some CNC manufacturers. In the case of robots, the compression functionality has to be closely correlated with the pose of the robot [16], since the accuracy of the robot is pose-dependent. Currently, compression is only done in Cartesian space and it is not correlated with the pose of the robot. For example, paths in the part of the workspace where accuracy of ±1mm is achievable are compressed the same way as paths in workspace areas where a ±0.1mm accuracy can be achieved.

P RELIMINARY EXPERIMENTAL RESULTS

A. Setup To measure the advantages of path planning in joint space, a preliminary experiment with a KUKA KR 60-3 High Accuracy (HA) robot has been conducted. This particular type of robot was chosen for the experiment because it has an integrated CNC-Extension developed by the company ISG GmbH. This allows offline programming in G-code. Furthermore, it allows the execution of G-code programmed in Cartesian space as well as in joint space, which provides an ideal platform for comparisons between path planning in Cartesian space and joint space. In this experiment, the robot TCP is programmed in both Cartesian space and joint space to follow a labyrinth-like path (see Figure 2). Furthermore, the set-up of the path planning for both methods contains trapezoidal acceleration profiles, linear interpolation and polynomial smoothening of corners. The focus of this study concerns solely the trajectory, e.g. velocity and accuracy, resulting from the two different planning methods. Therefore, the effects of the deflection force are of secondary importance for the comparisons. Thus, the experiment has been carried out without an actual workpiece and the pre-defined paths are executed in air.

C. Generation of Axes Reference Values by CNC Systems The waypoints generated by a CAM tool or by the suggested post-processing application serve as reference values for the CNC system. As described in the state of the art, a CNC system provides functions for tool radius and length compensation, path smoothening, workspace surveillance, velocity profile calculation, slopes, interpolation and transformation. However, these functions are designed to work for Cartesian systems like a classical machine tool. Even though some of these functions can be used on both Cartesian and joint space-based systems (e.g. the look-ahead function), most of the functions cannot. The proposed approach of a joint space-based CNC system, therefore, requires a joint space solution for the

259

Figure 2. Path for the Robot TCP.

The G-code for path planning in Cartesian space is generated automatically by means of a CAM tool where four waypoints are used for each line segment of the labyrinth. As opposed to this, the G-code for path planning in joint space is obtained manually rather than relying on the CAM tool. This is because the RC has a manufacturer-specific positioning error compensation during the calculation of the inverse kinematic for path planning in Cartesian space which is not available in any CAM tool. In order to eliminate this perturbation factor from the presented results, the joint angles for the path planning in joint space are obtained by executing the G-code of the Cartesian space and recording the joint angles on each waypoint directly in the robot engineering interface. In this manner, the built-in error compensation in the RC is also reflected in the obtained joint angles.

Figure 3. Velocity comparison on TCP.

However, the enhancement of the machining speed has been achieved at the expense of the path accuracy. Figure 4 illustrates the TCP position comparison on x-y-level (note that the diagonal line depicts the TCP movement from the starting position to the machining entry point, which is not relevant for the comparison). As shown in Figure 4, the path of the joint space planning is less accurate - especially in ydirection - and shows deviations compared to the Cartesian space planning. In the y-direction, the maximum deviation measures approx. 1.08mm, whereas in x-direction the maximum deviation amounts to approx. 0.18mm. As expected, the deviation varies on each line segment (see number 1-9 in Figure 4) and depends on the travel distance and direction (please note that, as described above, a constant number of waypoints per line segment have been used). For example, in the y-direction (number 1, 3, 5, 7, 9), as the line segment becomes shorter, the accuracy deviation also becomes smaller. The biggest deviation in the ydirection occurred at number 1 (x1  1.08mm) and the smallest at number 9 (x9  0.15mm).

B. Results The result of Cartesian space and joint space planning differs significantly in the machining speed. Figure 3 illustrates the comparison of the TCP path velocity (please note, that for clarity only the first half of the path is plotted). The actual machining process for the Cartesian space planning starts at approx. the 4th second and for the joint space planning at approx. the 3rd second. The movement before the mentioned times depicts the velocity during the TCP movement to the machining entry point. As shown in Figure 3, the machining process planned in Cartesian space operates with almost constant velocity of ±0.034 m/s. As opposed to this, the TCP of the joint space method is accelerated to velocities between ±0.235 m/s and ±0.12 m/s. As a result, the Cartesian space planning took more than 48 seconds to manufacture, whereas the joint space planning took less than 17 seconds to execute the depicted path, meaning an overall reduction of factor 3 in manufacturing time.

Figure 4. Position comparison on TCP (x-y-level).

260

challenge for path planning in joint space is thus to select the inverse kinematic solution where the pose inherently exhibits less inaccuracy.

On the other hand, in the x-direction (number 2,4,6,8), the accuracy deviation depends on both the travel distance and direction. In the positive x- direction, i.e. number 2 and 6, the accuracy deviations amount to y2  0.04mm and y6  0.03mm whereas the negative x-direction exhibits bigger deviations (y4  0.18mm, y8  0.08mm). Similar to ydirection, the accuracy deviation of the z-axis depends solely on the travel distance and amounts to z2  0.8mm at the longest segment line at number 2. TABLE I summarizes the measured deviations.

V.

TABLE I. TCP DEVIATION OF JOINT SPACE PLANNING ON X-Y-LEVEL. Segment number 1 3 5 7 9

Segment length (mm) 200 180 150 120 80

Deviation (mm) 1.08 0.7 0.5 0.18 0.15

yy+ yy+ y-

2 4 6 8

180 150 120 90

0.04 0.18 0.03 0.08

x+ xx+ x-

DISCUSSION

This paper evaluates the approach of a consistent joint space-based path planning for machining with industrial robots using CNC techniques. The evaluation of this approach shows advantages and drawbacks that are discussed in this chapter. First of all, it can be outlined that milling a given contour in joint space results in a much higher TCP velocity and consequently offers a high economical potential by reducing production times. It may be assumed that a joint space-based path planning leads to the increase of the look-ahead horizon, as computation time is reduced for transformation tasks and, therefore, is available for the look-ahead task. A higher lookahead horizon enables, in most cases, higher axis velocities. Furthermore, the dynamic abilities of the axes of the robot are fully exploited, apparently this is not the case in a Cartesian space-based planning. However, in regard to the TCP velocity, a much higher velocity fluctuation has to be stated using joint space-based path planning, whereas path planning in Cartesian space leads to a nearly constant TCP velocity. There may be different reasons for this, but one may be the fact that most of the CNC functions of current CNC systems only work in Cartesian space, but not in joint space. Velocity planning is executed on an axis level; on an overall level, planning of a constant TCP velocity in joint space is not performed. Decreased path accuracy is one of the disadvantages that is not pursued to be improved during this research. This limits the application of the joint-based planning to machining operations where accuracy is not a priority. Similarly, the discontinuous TCP velocity is a disadvantage that will not be improved during this research. Defining the waypoints based on an optimal criterion will have positive influence on these two unwanted effects, but the targeted applications are only the ones not requiring high accuracy and low surface roughness. Regarding the waypoint definition in the CAM phase, it has to be considered that the kinematic model of a robot is dependent on the manufacturer and the robot type. Furthermore, a precise kinematic model (calibrated) is custom for each and every robot. This means, that a particularization of the machine code for a specific robot will be required. This can be done in an offline phase, but then the generated code will not be fully interchangeable between different robot types. In order to overcome this, the postprocessing of the G-code (the customization for a specific robot), can be done on the robot prior to execution. Another important issue is related to the additional functions of the CNC, which currently work only in Cartesian space. Tool correction and interpolation are some examples. In order to plan milling paths in joint space, a joint space solution for these additional functions has to be developed. Only this way a joint space-based CNC for robots can be fully exploited.

Moving direction

C. Evaluation With the joint space path planning method, each joint features a higher axis velocity. As a result, an enhancement of machining speed up to factor 3 has been achieved. However, the resulted velocity on the TCP is not constant, which can be disadvantageous in some machining processes. Another drawback of the path planning method in joint space is the path inaccuracy. With four supporting points for each line segment, the result features deviations up to 1.08mm compared to the planning method in Cartesian space. Nevertheless, as shown in the experiment, the deviation varies depending on the travel distance and direction. Generally, the shorter the travel distance, the more accurate the path and the lower the slope of the TCP velocity. Therefore, waypoints programmed in the G-code, i.e. the segment length per program block, play an important role for the TCP velocity and path accuracy. In some machining processes, such as rough machining or deburring, a faster machining process is more desirable than constant velocity, and path inaccuracy up to ±1mm is still tolerable. On the other hand, a fine machining process requires a more constant velocity and high path accuracy, which influences the surface roughness respectively contour accuracy. A challenge for path planning in joint space is thus to identify the ideal number of waypoints, which achieves an optimal trade-off (pareto-optimum) between machining speed, path accuracy and constant path velocity. Additionally, the experiment reveals that in some cases the path accuracy deviation also depends on the travel direction. As shown in the experiment, in the positive xdirection the path appears to be more accurate compared to the negative x-direction. For instance, in spite of longer travel distance the deviation at segment line number 2 is smaller than at segment line number 4 and 8. A possible reason for this is that the accuracy is normally not distributed on every point in the workspace but depends on the pose. A

261

VI.

[5]

SUMMARY AND OUTLOOK

Machining with industrial robots constitutes in some areas (e.g. for large workpiece dimensions) a viable alternative to machine tools. However, machining with industrial robots still exhibits some drawbacks regarding accuracy. Research scientists and industry are working on many aspects of this topic. One approach is to target the improvement of issues related to path planning by using CNC techniques in combination with the RC. However, CNC systems generally work in Cartesian space whereas RC systems in joint space. Thus, consistence in path planning is not given at this moment. For this reason, the presented research, which is still in an early stage, outlines advantages and challenges of a continuous joint space-based path planning. The approach seems to be highly interesting as preliminary experiments showed significant improvements regarding the TCP velocity that can be increased by factor 3. However, the need of further research activities is exposed: using a sophisticated approach for waypoint generation, the decreased accuracy caused by joint space planning can be somewhat counterbalanced. Furthermore, additional functions of CNC systems have to be adapted for execution in joint space.

[6] [7]

[8]

[9] [10]

[11]

[12] [13]

REFERENCES [14] [1]

[2]

[3]

[4]

A. Karim and A. Verl. "Challenges and obstacles in robotmachining." Robotics (ISR), 2013 44th International Symposium on. IEEE, pp. 1-4, 2013. B. Denkena, J. Immel, and M. Schönherr, “ Industrial robots for machining purposes” (Industrieroboter für spanende Bearbeitungen) Werkstattstechnik online: wt, vol. 9, pp. Düsseldorf, Springer VDI Verlag, 2011. Y. Chen and F. Dong. "Robot machining: recent development and future research issues." The International Journal of Advanced Manufacturing Technology 66.9-12, pp. 1489-1497, 2013. C. Brecher and M. Weck. "Werkzeugmaschinen. Band 5. Messtechnische Untersuchung und Beurteilung, dynamische Stabilität.", (2006).

[15] [16]

262

A. Puzik. Increasing the machining accuracy of industrial robots by applying an active error compensation using a 3D-Piezo compensation machanism (Genauigkeitssteigerung bei der spanenden Bearbeitung mit Industrierobotern durch Fehlerkompensation mit 3DPiezo-Ausgleichsaktorik), Dissertation. 2011. L. Beyer. Genauigkeitssteigerung von Industrierobotern: insbesondere mit Parallelkinematik. Shaker, Dissertation, 2005. Weigold, Matthias. Kompensation der Werkzeugabdrängung bei der spanenden Bearbeitung mit Industrierobotern. Shaker, Dissertation 2008. T. Bonghardt. Methode zur Kompensation betriebsabhängige Einflüsse auf die Absolute Genauigkeit von Industrierobotern. Herbert Utz Verlag, Dissertation 2003. U. Wiest. Kinematische Kalibrierung von Industrierobotern. Dissertation, 2001. T. Alban and H. Janocha. Dynamic Measurement of Robots with Inertial Measurement Systems (Dynamisches Vermessen von Robotern mit Inertialmesssystemen. tm–Technisches Messen 67, 2000. C. Brisan and A. Csiszar, “Computation and analysis of the workspace of a reconfigurable parallel robotic system,” Mechanism and Machine Theory, vol. 46, no. 11. pp. 1647 – 1668, 2011. G. Pritschow. Einführung in die Steuerungstechnik. München: Hanser, 2006. M. Keinert and A. Verl. "System platform requirements for highperformance CNCs." Proceedings of FAIM 2012 22nd international conference on flexible automation and intelligent manufacturing, Helsinki, Finland. 2012. M. Keinert, B. Kaiser, A. Lechler, and A. Verl, “Analysis of CNC software modules regarding parallelization capability,” in Proceedings of the 24th International Conference on Flexible Automation and Intelligent Manufacturing. DEStech Publications, 2014, pp. 79–86. D. Scheifele. „KUKA.CNC – Machining with robot“ (KUKA.CNC – Fräsen mit dem Roboter). Vienna-tec, 2012. A. Karim, A. Verl, and R. Höhne, “ Vibration analysis on machining robots” (Schwingungsanalyse an Bearbeitungsrobotern), wt-online, vol. 1/2-2014, pp. 049 – 054, 2014.