Designing for Reliability and Robustness with MATLAB Parameter Estimation and Tuning Sensitivity Analysis and Reliability Design of Experiments (DoE) and Calibration U. M. Sundar Senior Application Engineer Amit Doshi Application Engineer 2016 The MathWorks, Inc. 1
Requirement Requirement Performance Measure, f(x) What is a Reliable and Robust Design? Reliability probability of satisfying a requirement Probability of success, p Probability of failure, 1-p Robustness ability to handle variation without loss of performance Robust Design Df(x) Optimal But Might Not be Reliable Not robust Not optimal Reliable Robust Df(x) Design Requirement x o x r Design Variable, x Initial (mean target) 2
Engineering Design Workflow Iterate till you find the best parameter set for a model satisfying design requirements Design Requirement Parameter Estimation Optimization MODEL Boundary Linear and Nonlinear Estimate Equalities Gradient Based Global Model or Prototype Calibration and validation: NEW Scenarios Simulation 3
Integral Powertrain Helps Bentley Motors Increase Horsepower, Reduce Emissions, and Improve Driveability Challenge Test multiple engine improvement strategies with reduced use of costly physical prototypes Solution Use MathWorks tools for Model-Based Design to redesign, model, and simulate the engine s valve train and to calibrate the powertrain Results Horsepower increased, strict emissions standards met Design iterations accelerated Test times reduced by 80% The Bentley Arnage. Simulating the dynamic valve train in Simulink enabled rapid development of a more powerful and responsive version of the engine that meets strict emission regulations." John McLean Integral Powertrain Link to user story 4
Demo: Suspension System Design Optimization Task: To design the suspension system for a new luxury car model Solution Approach: Model Parameter Tuning for Optimal Performance Is system reliable & robust? Yes Reliable and Robust Design No Monte Carlo Simulations 5
Parameter Tuning for Optimal Performance Modify Design Parameters NO Initial Design Parameters X Model or Prototype F(X) Objectives Achieved? YES Final Design OPTIMIZATION PROCESS Design process can be performed: Manually (trial-and-error or iteratively) Automatically using optimization techniques F(X) = Goal or min/max F(x) By changing X Subject to Design Requirements Optimization benefits include: Finding better (optimal) designs Speeding up design evaluation Reducing labor/effort Finding non-intuitive designs 6
Demo: Suspension System Design Optimization - Setup Minimize passenger vertical and rotational acceleration Z, Design Variables: front/rear springs kf, kr front/rear shock absorbers cf, cr Constraints: Level car Available parts Low natural frequency Required damping ratio Modify [kf, kr, cf, cr] NO Initial Design Parameters [kf,kr,cf,cr] [kf kr cf cr] Z, Minimum Found? YES Optimal Design Car mass = constant 7
Is My System Reliable? Minimize passenger vertical and rotational acceleration Z, Design Variables: front/rear springs kf, kr front/rear shock absorbers cf, cr Constraints: Level car Available parts Low natural frequency Required damping ratio range Will my system perform 100000 Kms? Uncertainty in vehicle mass Modify [kf, kr, cf, cr] NO Initial Design Parameters [kf,kr,cf,cr] [kf kr cf cr] Z, Minimum Found? YES Optimal Design Distribution Car mass = 8
Demo: Suspension System Design Optimization Task: To design the suspension system for a new luxury car model Solution Approach: Model Parameter Tuning for Optimal Performance Is system reliable & robust? Yes Reliable and Robust Design No Monte Carlo Simulations 9
Recap: Reliable and Robust Design using MATLAB and Simulink 10
Can our design accommodate changes? Challenge: Initial fan does not circulate enough air through the radiator to keep the engine cool during difficult conditions (such as stop-and-go traffic or hot weather) Two Solutions: Modify chassis which can accommodate a bigger fan with more capacity Modify the fan, rather than the chassis 11
Demo: Optimization of an Engine Cooling Fan Design distance Design requirements clearance Airflow > 875 cubic feet per minute pitch Design parameters Range Parameter Min Max Units Distance from radiator 1 1.5 in Blade pitch angle 15 35 deg Blade tip clearance 1 2 in 12
Case study: Improving the Design of an Engine Cooling Fan Task: Improve the design of an Engine Cooling Fan so that it can meet the newer airflow requirement Approach: Import Test Data Visualize and Examine Historical Data Apply statistical methods to design, conduct and analyze experiments Find optimized design values using parallel computing Automate documentation and share work with an APP 200 fans/day were tested over a 50 days 13
Platform for Calibration Process 14 14
Test Configuration I/O of Turbocharged 2.2L DOHC Engine with Dual-Independent Continuously Variable Cam Phasing (DIVCP) and Continuously Variable Intake Valve Lift (CVIVL) Intake VCP Exhaust VCP Spark Advance RPM AFR Intake Valve Lift VGT Vane Fraction Brake Torque Turbocharger Speed Intake Manifold Pressure Exhaust Temperature Load Load Command AFR Command Closed-Loop Load Controller Temperature-Limited AFR and Spark Control 15
Is Our Design Ready Already? Have we considered all the practical scenarios? Is it possible to consider test / simulate all the scenarios? Solution: Take the statistical representation of the physical system 16
Key Takeaways: Parameter Estimation and Tuning Sensitivity Analysis and Reliability Design of Experiments (DoE) and Calibration MATLAB: Single Platform Parallel Computing Statistics and Machine Learning Model Based Calibration Simulink Optimization Simulink Design Optimization 17