Rapid Concept Realization for Conceptual Design of Modular Industrial Robots

Transcription

1 NordDesign 2010 August 25 27, 2010 Göteborg, Sweden Rapid Concept Realization for Conceptual Design of Modular Industrial Robots Edris Safavi, Mehdi Tarkian, Johan Ölvander IEI/Machine Design Linköping University Linköping Sweden Abstract When conducting design on novel mechatronic products, it can be valuable to test and evaluate the performance and properties of the concepts throughout the design process by producing them as downscaled prototypes, see Jouannet et al. [1]. This is especially true when the product is of unconventional design and the designer can get increased confidence of the proposed concept by testing it as a sub scaled version. Nonetheless, the process of realization of new concept should be done in a rapid fashion in order not to halt the design process and simultaneously increasing explicit knowledge about the concept. A case study will be illustrated which demonstrates how fully automated design and construction of downscaled prototypes is performed. Keywords: Automated design, multidisciplinary design, industrial robots 1 Introduction A novel design framework is currently being developed at Linköping University to more effectively understand and manage the complexity of multidisciplinary and automated design, e.g. for industrial robots. In this framework a variety of optimization algorithms are utilized to search through a vast design space. The design framework has in previous work been used to design modular industrial robots. Since the modular industrial robots are designed with the prospect of sharing links as well as having hollow-shaft actuators, they are classified as an unconventional design. Due to the unconventional design, the number of design uncertainties is initially high. Therefore, more knowledge concerning the new concept can effectively increase the designer s confidence. The downscaled prototype can help the designer to actually use and test the product. The concept of modular redundant industrial modular robots has been discussed intensely since late 80, e.g. Krenn et al. [2] and Paredis et al. [3]. Recently these robots are becoming increasingly attractive for robot manufacturers, such as Motoman [4] Nachi [5] and Robotics Research [6], with features such as maximizing the operational work space. Also the increased flexibility followed by increasing number of DOF makes redundant modular robots more suited for applications where increased mobility is requested see Bluethmann et al. [7]. To more effectively understand and manage the complexity of this technology and find the optimal solution for a family of modular robots, a novel design framework is being developed at Linköping University, see Lundén et al. [8] and Petterson et al. [9]. The mechanical structure of the modular redundant robot consists of a base followed by a series of modular structure links. Each module consists of drive-train components (servo actuator, combining precision Harmonic Drive gearing with highly dynamic AC servo

2 motors). Major components of the robot controller are power units, rectifier, transformer, axis computers and a high level computer for motion planning and control. The design choices made on the links before applying modularization and reuse are not trivial. Considering the fact that limited knowledge is at hand concerning the properties of the finished product, some design aspect should be easy modifiable during the entire design phase. One example is the type of drive train used. The choice of type of motors and gears dictates the shape of the robot links. Changing the gear type should not be encountered with re-modeling of the entire geometry model. An important factor here is therefore a possibility for the design team to remedy shortcomings of certain design choices made early on. In order to do so, various properties of the robot should be analyzed throughout the geometric design process. Important properties to check are the kinematics and dynamics of the robot which in turn are utilized to compute life time estimations on certain components. To faster evaluate the design alternatives, the dynamic and geometric models are integrated through a user interface where the design data for both models are stored, see Figure 1. Figure 1. A multidisciplinary design approach for modular robot design. Following input in the user interface, fast data from both models are gathered simultaneously, supporting the design of the robot following a holistic approach, where various engineering aspects of the robot are analyzed concurrently. Also in the component library both the geometrical and dynamical aspects of the components are stored Research outline In order to create a multidisciplinary design framework, formally articulated and documented knowledge, in other words explicit knowledge, are required [10]. Another issue is design uncertainties in view of unconventional design. This paper will address a methodology which can be adopted to increase the level of know-how on unconventional designs, i.e. modular industrial robots. The methodology can also be utilized to increase the level of explicit knowledge in order to achieve automated design. To be capable of increasing know-how and level of explicit knowledge, following steps are required (Figure 2): 1. Models depicting various disciplines of the product. 2. Seamless dataflow between the models in a multidisciplinary design approach. The cluster of connected models is here referred to as virtual model. 3. Automating the optimal design search with various optimization algorithms. 4. Creating the means to rapidly generate a physical prototype. 5. Test how the physical prototype performs compared to the virtual model and thereby formulate new knowledge in order to improve the virtual model.

3 User Interface Dynamic Model Component Database Genetic Complex RF Geometric Model Non Gradient MD Modeling SM Simulation DC Evaluation and optimization OM Physical Prototype VM DV MD = Model Definition SM = System Model DV = Design Variables DC = Design Characteristics OM = Optimized Concept VM = Verified Model Figure 2. Proposed Design Framework Although high fidelity tools are employed to create the virtual model, the conducted research is still classified under the conceptual phase since the actual choice of concept is still not finalized. In this paper the outlined steps will be depicted for an industrial modular robot. In previous work the virtual robot has been employed in multidisciplinary and multi objective optimization frameworks [11] Paper outline Throughout the paper, the different components of the framework will be described in detail. One of the main challenges is to increase confidence and knowledge about the concept under evaluation by producing down scaled prototypes. The various components of the framework such as the geometric and the dynamic models will first be depicted. The following section, Rapid Concept Realization, gives a description of the methodology approaches taken to produce the prototypes. In Framework Validation, various test results from a manufactured prototype are presented and finally concluded in the last section. 2 Geometry model To simulate and evaluate the properties of any given product, the geometry of the product is needed. CAD tools can be used to calculate this geometry, but because simplifications are introduced in the geometry, the estimations are usually initially inaccurate and re-modeling has to occur in a frequent rate in order to define a sufficiently accurate model. Nevertheless considering the time demanding process to facilitate new CAD models, this tool has traditionally been introduced in the later stages of design when the geometry of the product is better defined. Commercial CAD tools available in the market are becoming increasingly suitable to generate automated geometries for multi-disciplinary optimization (MDO) and design. CAD tools such as CATIA, Solid Works, Pro Engineer and NX6 all offer parametric design with varying functionalities. The advantages of creating flexible and robust geometries for automated design have been demonstrated from various research groups and disciplines. The aircraft research domain has made big strides for describing methods for creation of automatic generated geometries as framework enablers. This has been effectively demonstrated by Lundström et al [12], for micro-uav design, Tarkian et al.[13] for Civil Aircraft design and La Rocca et al. [14] in the analysis for specific aircraft feature. The advantages of automated geometric modeling for holist design have been illustrated for various other applications such as kinematic and dynamic optimization for industrial robots, Tarkian et al [11], as well as modeling airfoil shapes for use in wind turbine design, as presented by Cooper et al. [15]

4 Geometric modifications made on a CAD model will either alter the shape of the elements (morphology) or alter the number of elements (topology). Topological parameterization is accomplished by defining templates and context manuals, see Figure 3. Figure 3. Topological instantiation by defining a template and a context. The manuals contain complete construction procedures of the template objects and to which geometric features they are constrained to. These definitions enable the template to be instantiated into different contexts, increasing reusability of created geometries. The instances are both context dependent and able to vary parametrically. The geometry models created in CAD tools can be very flexible and robust in the sense that both the shape and the number of geometric objects of the model can be parametrically defined Geometry model of the modular robot The geometry hierarchy in this application consists of three main assemblies and each assembly is defined by importing the geometries of high level templates, stored in template libraries (Figure 3). The geometric templates are stored outside the geometric model and initiated parametrically using VB script. Figure 4. The relations between the assemblies and template libraries visualized 3 Dynamic model The dynamic properties of the robot have to be simulated in order to reject poor geometric design choices. Simultaneously, to be able to compute the dynamic behavior, the weight properties have to be known. Therefore there is a need to have integration between these disciplines as seen in Figure Dynamic model of the modular robot The dynamic model created in the tool Dymola (Elmqvist [16]) consist of three subcomponents, as seen in Figure 5. The first component, a trajectory planner, computes the trajectory for the robot in joint space. The joint space trajectory is then sent to the drive train consisting of electrical and mechanical models of the motors and gears. The output from the drive train component is then used to generate motion on the rigid body model, containing mechanical structure of the links.

5 Figure 5. Dynamic model created by utilizing the in house modeled trajectory planner and the Modelica standard library The trajectory planner Figure 6 consists of four sub-components, namely a Trajectory Generator, a DH Parameter component, a Dynamics component and finally an Inverse Kinematics component Trajectory planner The components visualized in Figure 6 have the following features: DH Parameters: The actual kinematic structure of the robot is defined by choosing the Denavit-Hartenberg parameters of the sought after concept. This component doesn t contain any mathematical equations to execute upon simulation. The purpose of the component is to forward the DH parameters to the Dynamics and Inverse Kinematics components where the kinematic structure of the robot is needed. Trajectory Generator: The Cartesian space trajectory is user defined, i.e. number of trajectory segments. Inverse Kinematics: This component is divvied into Inverse Position and Inverse Velocity components. In Inverse Position, the joint angles are computed and sent to the Inverse Velocity component. Here the transformation matrix for each axis is computed by utilizing on the pre-described DH-parameters and joint angles. Once the transformation matrix for each axis is defined, the Jacobian of the robot can be determined and thereby joint velocities and accelerations computed as well. Dynamics: Here the Newton Euler formulation [17] is implemented where the link velocities and acceleration are computed, forward recursively, starting with the first link. When the kinematic properties of the links are calculated then the force and torque vectors between the links are backward recursively computed, starting with the last link. Figure 6. The trajectory planner is modeled using the Modelica language When performing a simulation, following event sequence will take place: 1. DH parameters are sent to both the Dynamics and Inverse Kinematics components to attain the correct kinematic structure of the robot.

6 2. The tool center position, velocity and acceleration vectors in Cartesian space for each time step are generated by the Trajectory Generator and sent to the Inverse Kinematics component. 3. The Inverse Kinematics component determines the joint angles for each time step and by using the Jacobian, computes the joint velocities and accelerations. The joint angle, velocity and accelerations are sent to the Dynamics Component for each time step. 4. The Dynamic Component calculates the force and torque vectors at each axis. 5. The joint values generated in the Trajectory Planner component are sent to the drive train component as input signals to ultimately put the Rigid Body Model into motion Drive Train for full scale model The drive train model of the full scale model consists of a Harmonic Drive gearing and a dynamic AC motors. However being out of scope of this paper, it will not be further depicted Drive Train for downscaled prototype The drive train consists of a power supply, motor driver, electric motor and gearbox for the downscaled prototype, See Figure 7. Properties such as friction and life time estimation of each component are computed in the Drive-Train component. The models used are briefly explained below. Figure 7. Drive train consists of a power supply, motor driver, electric motor and a gear. The stepper motor is created, see Figure 8, by modifying the DC-motor in the Modelica Standard Library [16]. The motor is described by the motor L (motor inductance), R (motor resistance) and K (Transformation coefficient). The planetary gear utilized is also in the standard library. The losses in gear are also modeled by defining the gear efficiency and backlash coefficient. The pulse generator is created in order to provide the step pulse to the motor. The velocity and position of the motor is controlled by adjusting the duration and width of the pulses. A schematic of the drive train components is illustrated in Figure 8. Motor Driver Motor Gearbox Figure 8. The Modelica model of the drive train components 4 Rapid concept realization As described earlier the virtual model can simulate the properties of both the full scale model and the down scaled prototype. This is since the geometry model is constructed in a dynamic top down fashion and the topology of the robot assembly is modifiable, only new link and drive train templates are needed to be added in the template library when the downscaled version is to be modeled, see Figure 9.

7 Figure 9. The type of drive train was modified during the design phase of the concept realization robot which resulted in new templates in both the MDS and MDC libraries By adding new geometry templates in the template library, it is possible to define and assemble a new type of modular robot from the user interface, see Figure 10. Figure 10. Comparison between the CAD and the physical prototype link 1 & 2 The concept realization process portrayed in Figure 11 begins by defining the geometry parametrically. The mass properties are then automatically exported to the dynamic model where the dynamic properties are estimated. In the Dynamics component, see Figure 6, the torques and forces required at each link are computed. By specifying the motor parameters in the drive train component, see Figure 5, the dynamic behavior of the motors are further calculated. Thereby suitable geometries for the links and selection of drive train units are faster approximated. When the final design is set, a physical prototype can be assembled. The control computer of the prototype receives the reference joint values from the trajectory planner of the dynamic model. The joint values are converted to signals for the motor driver by using the Enhanced Machine Control (EMC2) software [18]. The signals are transferred to the motor driver via the serial port (communications ports). Since the implemented motors are of the type hybrid stepper motors, there is no need of feedback control and the angular position of the rotor is equal to the number of steps inputted. Figure 11. The process of concept realization initiates with geometry definition, evaluated in a dynamic model, then a controller is automatically defined and finally a prototype manufactured The physical prototype is manufactures with a 3D prototyping machine. The 3D printers used are suitable for this purpose, since complex geometries are produced time efficiently. Following various test cycles performed on the prototype, the dynamic model can be further modified in order to calibrate the reference variables of the robot controller. If the physical prototype does not fulfill the defined requirements, the process of a new concept can be initiated. This iterative process continues until the prototype s performance is satisfactorily evaluated or the concept scraped.

8 4.1. Framework validation In order to evaluate the predictability of the framework, cycles performed by the physical prototype is compared to ones simulated by the virtual model. The functionalities which can be measured and compared are i.e. the trajectory and the degree of freedom of each axis of the prototype, compared to the virtual model. The motion is generated by firstly defining the Denavit Hartenberg (DH) parameters [17] of the sought after concept and then specifying a number of trajectory segments in the trajectory generator, see Figure 12. Figure 12. The DH parameters (above figure) and Trajectory segments (below figure) Two separate test motions were prepared. The first test consists of 15 line segments of which 6 segments are on a vertical board surface (to write LIU), see Figure 13. The second test consists of 6 line segment of which 4 segments are on a vertical board surface (to draw a rectangle), see Figure 13. When performing the cycles it was noted that the physical prototype did not manage to execute the cycles in the manner which the virtual model indicated. The arisen situation clearly demonstrated the lack of explicit knowledge utilized to formulate the virtual model. It was concluded that the formulation of backlash was missing from the gear manufacturer, resulting in this parameter not to be taken into consideration when designing the robot. Hence the virtual model indicated an outcome which turned out very different in reality. Figure 13. The lack of backlash data in the virtual robot (robots to right) gave a false impression of how the physical robot would actually perform (robots to left) After taking the backlash effect into consideration, the physical robot could perform the cycles as defined in the dynamic model, see Figure 14.

9 Figure 14. Comparison between the trajectories performed by the dynamic model and the physical prototype Taking a closer look at the specific axes of the robot during the performed cycles also demonstrates the similarities between the physical prototype and virtual model as seen for axis 2 in Figure 15. Figure 15 Comparison of the performed angular movement for axis 2 with the dynamic model (top) and the physical prototype (bottom) 5 Discussion and conclusion In this paper an approach to conduct design on industrial modular robots is presented. The product is analyzed on a system level by effectively connecting the multidisciplinary models to a common user interface. The framework which is used to optimize the actual full scaled prototype is utilized to create the downscaled prototypes as well. Being able to model and simulate widely different scales of products, shows the flexibility and usefulness of the framework. It is demonstrated that by producing downscaled prototypes of the concept under evaluation, explicit knowledge about the product can increase, thereby boosting the confidence about the concept. Furthermore a variety of methods for how to decrease the time needed to generate the prototypes, thereby achieving rapid concept realization, is also presented. 6 References [1] Jouannet, C., Lundström, D., Amadori, K. and Berry, P., Design of a Very Light Jet and a Dynamically Scaled Demonstrator, Jan. 2008, 46th AIAA Aerospace Sciences Meeting and Exhibit, Reno, NV, USA [2] Krenn, R.; Schäfer, B.; Hirzinger, Dynamics Simulation and Assembly Environment for Rapid Manipulator Design. 7th ESA Workshop on Advanced Space Technologies for Robotics and Automation, ESTEC, Noordwijk, Netherlands, November 19-21, 2002 [3] Paredis, C.J.J., Brown, H.B., Khosla, P.K., A rapidly deployable manipulator system, Proceedings of the IEEE International Conference on Robotics and Automation, MN, 1996, pp. 1434]1439, Minneapolis, USA. [4] Motoman SIA20

10 [5] Nachi MR20 [6] Robotics Research [7] W. Bluethmann, R. Ambrose, M. Diftler, S. Askew, E. Huber, M. Goza, F. Rehnmark, C. Lovchik, D. Magruder, Robonaut a robot designed to work with humans in space, Autonomous Robots 14 (2003) [8] Johanson B., Ölvander J., Pettersson M., Component Based Modeling and Optimization for Modular Robot Design, ASME DAC 07, Las Vegas, USA, September 4-7, [9] Petterson M., Andersson J., Krus P., Methods for Discrete Design Optimization, in proceedings of ASME DETC'05, Design Automation Conference,, Long Beach, California, USA, September 24-28, [10] McInerney, Claire. (2002). Knowledge Management and the Dynamic Nature of Knowledge. Journal of the American Society for Information Science and Technology, 53 (12): [11] Tarkian, M. Ölvander, J., Feng X., Pettersson M., Design Automation of Modular Industrial Robots, ASME CIE09, San Diego, USA, Sep [12] Lundström, D., Amadori, K., "Automation of Design and Prototyping of Micro Aerial Vehicle" - 47th AIAA Aerospace Sciences Meeting, Jan. 2009, Orlando, FL, USA [13] Tarkian, M.,, Zaldivar, F., Aircraft Parametric 3D Modeling and Panel Code Analysis for Conceptual Design, 26th ICAS, Anchorage, USA, Sep [14] La Rocca, G., van Tooren, M.J.L., Enabling distributed multi-disciplinary design of complex products: a knowledge based engineering approach, J. Design Research, Vol. 5, No. 3, pp [15] Cooper D., La Rocca G., Knowledge-based Techniques for Developing Engineering. Applications in the 21st Century. 7th AIAA ATIO Conference [16] Elmqvist E., Brück D., and Otter M., Dymola - User's Manual, Dynasim AB, [17] Spong W. Mark and Vidyasagar M Robot Dynamics and Control, John Willey & Sons Inc, pp 65-71, [18] Linux CNC