ScilabTEC 2015 - Noesis Solutions
2 likes4,883 views
The document presents an evolutionary optimization approach for tuning a PID-controlled robotic arm using the Optimus platform by Noesis Solutions. It highlights the integration of Scilab and various optimization strategies to automate the PID tuning process, leading to significant time and resource savings. The results demonstrate improved efficiency in managing the robotic arm's performance through automated workflows and intelligent optimization methods.
ScilabTEC 2015 - Noesis Solutions
- 1. Using an Evolutionary Optimization approach to tune a PID controlled robotic arm Taylor Newill – Noesis Solutions, NA Technical Services Silvia Poles – Noesis Solutions, Engineering Manager
- 2. Agenda Optimus Overview Integrating Scilab Genetic optimization of a PID 1 2 3 Future Development 4
- 3. Agenda Optimus Overview Integrating Scilab Genetic optimization of a PID 1 2 3 Future Development 4
- 4. Optimus: A Technology Connection Platform Creating a repeatable, automated process: • Multi code, • Multi CPU, • Synced Rapid identification of key design variables: • Histograms • Sobol indices, ANOVA • Taguchi Process Integration Design of Experiments • Create surrogate models from any dataset • Multiple approaches are tried and the best one can be used Response surface models Optimization • Gradient • Evolutionary • Hybrid Robust Design • Design for Six Sigma • Taguchi • FORM • FOSM Optimization & Robust Design
- 5. Optimus: An Open Platform • Can drive any CLI • Embedded as a UCI/UCA • Externally through generic Interfaces • External algorithms can be used with ‘User Optimization’ • Driven from simple XML script • Any code can be used, python, C, scilab, etc. Connect any Software Apply any Algorithm • Java based GUI • ‘One click’ connection with queuing systems Drive from any Environment All functionality available through python API Full API
- 6. Noesis Solutions - Solutions for Engineering Optimization … more than 100+ person-years experience in PIDO … sales offices across Europe, US and Asia realizing double digit profitable growth for 18+ years. A leading software & services provider A strong worldwide presence … Optimus is our only product and focus, and we partner with all major CAE and mathematical modeling vendors An independent innovation partner Noesis Solutions
- 7. Noesis Solutions Most extensively deployed software of its kind…
- 8. A few default interfaces
- 9. Automated Multi code Multi CPU Repeatable CFD 20 Design parameters FEA Post-Processing
- 11. Should I Automate? (XKCD 1319) Mouse over Text: ‘Automating’ comes from the roots ‘auto-’ meaning ‘self-’ and ‘mating’, meaning ‘screwing’
- 12. Is it worth the time? (XKCD 1205)
- 13. Integrating Scilab scripts in Otimus • User Customizable Interface • The UCI is easily configured with XML-files, respecting a very simple syntax • Drag and drop functionality for easy multiple disciplines
- 14. Scilab UCI The SCE file is updated with every experiment, the xcos file is imported, then the results are extracted and then calculations are done on the resultant curves
- 16. Evolutionary Optimization of a PID Controlled Robotic Arm 2
- 17. Robotic Arm • Programmable mechanical arm • The links of of the robot are connected by joints allowing – rotational (angular) displacement – translational (linear) displacement • The links of the robot form a kinematic chain • The terminus of the kinematic chain is the end effector
- 18. PID Control (Wikipedia) • When the robotic arm is given a target, a PID loop is used to control the movement of the arm relative to the target • PID is a proportional-integral-derivative controller • It is a control loop feedback mechanisms • A PID controller calculates an error value as the difference between a measured process variable and a desired target • The controller attempts to minimize the error by adjusting the process through use of a manipulated variable.
- 19. The Challenge • Typically, robotic arms are tuned by tuning one PID loop at a time and cycling through the loops until the overall behavior is satisfactory. • This process can be time consuming and is not guaranteed to converge to the best overall tuning. • In this example we will automate the tuning of 1 PID, multiple PID’s have since been added
- 20. Robotic Tools for Scilab/Xcos • From the ‘Scilab Ninja’ – Dr. Varodom Toochinda • Beta Version – Kinematics – Dynamics – Path generation – Control
- 21. Evolutionary Optimization • Start with a population covering the design space • Make slight changes to input variables of the best performing experiments • Make a new population based on the best performing experiments
- 23. Parametric Usage of the Toolbox • Optimus will detect any controllable inputs • Outputs can be extracted – From memory – From output files – From response variables
- 25. Automation
- 27. Design of Experiments • Individual runs took about 30 seconds • Wanted to adequately cover the design space – 3 inputs, 3 responses – 81 experiments tested • Latin Hypercube DOE was used
- 28. Response Surface Model • 18 different RSM’s were tested, RBF was selected • Lowest error, created quickly, easily exported
- 30. Evolutionary Optimization • Multiple strategies will be compared • All run on RSM • Results validated with simulation • Objective is to reduce PID error and reduce experiment count • Single Objective Optimization Algorithms Tested – Differential – Self adaptive – Simulated Annealing – CMA-ES – Particle Swarm
- 31. Comparison of Optimization Strategies
- 33. Performance Impact Strategy Experiments Normalized PID Error Validated Difference from Nominal Nominal 0 892553 Differential 256 660291 659332 26.1% Self Adaptive 333 660575 660027 26.05% Annealing 153 660100 659301 26.13% Particle Swarm 664 660085 658948 26.17% CMA-ES 233 661889 661149 25.92%
- 34. Resource Impact Task Time w/o Optimus Time with Optimus Total Time Saved Create Robotic arm model 120 min 120 min 0 Create workflow 0 5 min -5 min Test each PID setting (3) 2 min 0 6 min Tune entire PID (3) 60 min 0 180 min Simulate arm movement 2 min 2 min 0 Run Optimization (253 simulations) 506 CPU min 200 CPU min 306 CPU min
- 35. Is it worth the time? • Workflow development took 10 minutes • For one tuning routine Optimus saved – 3 hours of human time – 5 CPU hours
- 36. With Scilab and Optimus you can Conclusion Save Time Consolidate Knowledge • Drive Scilab and combine with other tools • Automate Repetitive Tasks • Maximize efficient use of your simulation resources • Simplify your design work by focusing on key parameters • Automate parametric studies • Intelligent optimization methods • Create fast and accurate meta models • Share model data through Excel, etc… Improve Performance