Force Controlled Robotic Assembly David P. Gravel Senior Technical Specialist Ford Motor Company Advanced Manufacturing Technology Development Center Robot Force Control Partners Kawasaki Heavy Industries FANUC Robotics Robotics Research Corp. Case Western Reserve University, Cleveland,OH National Center for Manufacturing Sciences Sandia National Lab ComauPICO MicroDexterity Systems Marquette University in Milwaukee, WI National Center for Standards and Technology (NIST) 1
Current Robotic Applications Dispensing 3% Arc Welding 13% Assembly 3% Painting 18% Inspection 1% Material Handling 27% Material Removing 4% Spot Welding 31% Source: Robotic Industries Association Current Robot Paradigm Assembly with position-servo robot control cannot occur where the assembly tolerance is less than the positional uncertainty. Position controlled robots do not have the compliance required for accommodating alignment errors in the inherent in the assembly process. Even devices such as a RCC device are not sufficient to allow conventional robots to tackle complex assembly tasks. 2
Business Drivers Thousands of components in powertrain and vehicle assembly can now only be manually assembled Many heavy parts can cause Repetitive Stress Injuries, and require relief workers Quality Issues can arise from tired workers assembling heavy parts for long periods of time Cycle time improvements can boost productivity and capacity Worker safety in improved if the forces robots can impart to their environment is limited Example 3
Two Human Attributes Needed Humans have natural attributes used to successfully manipulate delicate objects with purposeful intent and excellently integrated tactile and visual senses: 1. Force-control: Provides the robot with a soft touch to react to contact forces for assembly, insertion, or for meshing of gears 2. Learning: Humans have an innate learning capacity not found yet in industrial machines Project Goals Develop a new robot control paradigm that allows robots to control the forces they impart to an environment Develop human-like assembly strategies to direct robot forces and moments in an intelligent fashion Implement force controlled robots on ergonomic problem applications that can have a quality impact 4
History of Force Control Passive Compliant Device research from Marquette University Trellis / Kawasaki force control with CWRU Robotic Research Corp. application development NIST ATP for Flexible Robotic Assembly for Powertrain Applications ($3.5M) Kawasaki Force Control Joint Development Project FANUC Robotics Force Control Development Project How Force Control Works New method for robotic control, not a position servo but a motor torque control scheme that limits and directs contact forces during assembly. 5
Force Control Test Plan Setup a graduated difficulty applications where each phase builds on skills acquired in previous phases. 4R70W Clutch Hub AX4N Intermediate Clutch Hub AX4N Triple Clutch Initial Force Control Application 4R70W forward clutch hub Locate the top clutch ring surface and control impact force and vertical insertion force Impose a search pattern that threads through the clutch plates Repeat until done 6
Second Application AX4N Intermediate Clutch Hub - More difficult since plates have less internal friction, making relative motion between the hub and the clutch plate difficult Need high speed accel/decel with force control More agility required for this task, a high performance version of the forward clutch algorithm. Hardest Application AX4N Triple Clutch Delicate sun gear to insert into a planetary gear set Internal spline gear mates approx. ¼ after initial sun gear insertion Heavy part, ~10kg 7
RRC Test Results Robotics Research robot was able to do all three tasks but lacked a good robot controller system and has not established industrial reliability KHI Test Results Excellent FC hub insertion cycle times Superb reliability for AX4N triple clutch, total cycles times close to the 24 sec. line rate 8
FANUC Test Results Excellent results with all force control efforts FANUC is installing a intermediate clutch application at the Van Dyke Transmission Plant Excellent AX4N triple clutch cycle times. FRAPA NIST ATP Flexible Robotic Assembly for Powertrain Applications (FRAPA) $3.5M NIST ATP Paradex I results from Case Western Reserve University Dramaticlly lower and more repeatable forces 9
Machine Learning Humans are able to learn complex assembly tasks relatively quickly. Machines, however, are notoriously difficult to teach. Work at CWRU on machine learning has focused on four areas. How do people perceive edges and boundaries when mating two parts together. Strategies people invoke under various force feedback conditions. Create a model based system that uses signal feedback to direct robot motion. Implement the knowledge gained in steps 1-3 into a neural net robot controller. CWRU Experiments Moments exist near a hole Experiments using a 6 DOF joystick people deprived of moment info have dramatically more trouble performing a peg in hole assembly 10
Edge and Moment Data Presentation Even heavily filtered, or impoverished data, presented to humans was sufficient for quick convergence of the peg into the hole. Importance of Training Paths taken by trained and untrained operators to accomplish the peg in hole task with filtered moment and edge data. 11
Neural Net Results Near the location of the hole the neural network was able to pick up signals out of the noise of dragging the peg on the surface. The results are plotted to the right. 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0.2 0.4 0.6 0.8 1 Actual Sensor Moment Data Moment information as a slightly tilted 9.8 mm peg is dragged across 10mm hole in various locations 12
Genetic Algorithms A self learning system that self optimizes system parameters Removes burden of training from plant floor personnel Adapts to changing conditions (stack up) Less plant floor expertise needed to maintain system 13
Date: PUBLICATIONS ORDER FORM ID#: internal use only ITEM QTY. COST TOTAL: SHIPPING INFORMATION Name Title Company Address City State Zip Telephone Fax Email Address PAYMENT INFORMATION Check Enclosed Invoice me (publications will not be shipped until payment is received) Credit Card American Express Visa MasterCard Account #: Name on Card: Expiration Date: Please complete this form and mail to: Automation Technologies Council 900 Victors Way, P.O. Box 3724 Ann Arbor, MI 48106 For faster service, fax this form to: (734) 994-3338 Phone: (734) 994-6088.