Structural & Multidisciplinary Optimization Group Deciding How Conservative A Designer Should Be: Simulating Future Tests and Redesign Nathaniel Price 1 Taiki Matsumura 1 Raphael Haftka 1 Nam-Ho Kim 1 University of Florida, Gainesville, FL 1
Structural & Multidisciplinary Optimization Group Presentation Outline Introduction Design, testing, redesign and calibration Methods Simulating Future Test & Future Redesign Analytical Techniques To Remove Sampling Noise Results Two Redesign Strategies Redesign to increase performance (reduce mass) Redesign to improve safety Discussion & Conclusion IntroductionMethodsResultsConclusion 2 / 19
Structural & Multidisciplinary Optimization Group Tests Prescribe Redesign and Calibration High levels of epistemic uncertainties at the design stage would exact large weight penalties to compensate for Tests can be used to reduce epistemic uncertainties These tests are used to: Calibrate analysis models and improve their accuracy Prescribe redesign if tests indicate that design is unsafe or too conservative There is a tradeoff between weight and redesign costs, in that conservative design is heavier but less likely to require redesign However, this tradeoff is currently implicit and based on experience at a company level IntroductionMethodsResultsConclusion 3 / 19
Structural & Multidisciplinary Optimization Group Two Redesign Strategies Aircraft designers face two challenges How to maximize safety while minimizing mass How to avoid high probability of redesign Two redesign strategies Redesign for Performance Start with a conservative design (higher safety factor) Redesign if test reveals that design is too conservative (too heavy) Redesign for Safety Start with a less conservative design (lower safety factor) Redesign if test reveals that design is unsafe (safety factor is too low) Future test & redesign Time DesignTestRedesign IntroductionMethodsResultsConclusion 4 / 19
Structural & Multidisciplinary Optimization Group Simplest Demonstration Example A solid bar with circular cross section under axial loading Uncertain Parameters DescriptionClassificationDistributionSymbolC.O.V / Bounds (%) Applied LoadAleatoryNormalP0.20 Material StrengthAleatoryNormalσ allow 0.12 Calculation ErrorEpistemicUniforme calc ±30 Measurement ErrorEpistemicUniforme meas ±10 P A IntroductionMethodsResultsConclusion 5 / 19
Structural & Multidisciplinary Optimization Group Design & Redesign Procedure Deterministic design optimization (DDO) is performed to minimize mass subject to a stress constraint A single test is performed and the test will be passed if the measured factor of safety is within the lower and upper safety factor limits, S L and S U The test result is more accurate than the calculation and therefore we can calibrate our model using the test result Using the updated calculation we may wish to redesign for a new safety factor, S re S ini S U S L S re IntroductionMethodsResultsConclusion 6 / 19
Structural & Multidisciplinary Optimization Group Simulating a Future Test & Possible Redesign There are errors in calculated and measured stresses We can combine the above equations to simulate a test result (assuming we know the calculation and measurement errors) We can use Monte Carlo Simulation (MCS) of errors to generate a distribution of possible future test results For n pairs of error samples we obtain n possible futures For each possible future we can use first order reliability method (FORM) to calculate true probability of failure For our simple problem an analytical solution of FORM provides exact probability of failure Future test: treat e calc & e meas as a random variable IntroductionMethodsResultsConclusion 7 / 19
Structural & Multidisciplinary Optimization Group Analytical Method to Eliminate MCS Sampling Errors Joint PDF of errors (independent) Heaviside functions to model the redesign event (indicator function) Expectations for these functions are easily calculated After redesign using (S ini, S L, S U, S re ) IntroductionMethodsResultsConclusion 8 / 19
Structural & Multidisciplinary Optimization Group Optimization of Safety Factors & Redesign Rules For an individual designer, the design problem is deterministic Safety factors are determined by regulations and additional company safety margins However, a design group seeks the optimum set of rules to balance performance (mass) against probability of redesign We formulate the following multi-objective objective optimization problem to minimize mass (area) and probability of redesign Constraint on probability of redesign is varied to capture Pareto Front of optimal designs Redesign for PerformanceRedesign for Safety Formulation: IntroductionMethodsResultsConclusion 9 / 19
Structural & Multidisciplinary Optimization Group Optimization of Discrete Cases 4 possible future scenarios e calc = +30%(conservative) or -30%(unconservative) e meas = +10% or -10% P re = 50% Error [e calc, e meas ] [-30,-10] [ 30,-10] [-30, 10] [ 30, 10] IntroductionMethodsResultsConclusion 10 / 19
Structural & Multidisciplinary Optimization Group Graphical optimization Redesign for safety: Start with low S ini and redesign with high S re Redesign for performance: Below S ini = 1.45, P F constraint cannot be satisfied IntroductionMethodsResultsConclusion 11 / 19
Structural & Multidisciplinary Optimization Group Distribution of mass after redesign Redesign for performance reduces average mass by 25.2 Redesign for safety increase average mass by 32.6 For performanceFor safety IntroductionMethodsResultsConclusion 12 / 19
Structural & Multidisciplinary Optimization Group Distribution of P F after redesign Redesign for performance can reduce mass with almost no penalty in reliability Redesign for safety has larger impact on reliability For performanceFor safety IntroductionMethodsResultsConclusion 13 / 19
Structural & Multidisciplinary Optimization Group Discrete Error Simulation (4 cases) Redesign for performance yields about 3% lower average mass Redesign for safety significantly improves P F Error [e calc, e meas ] Area Before Redesign (mm 2 ) Area After Redesign (mm 2 ) P F Before RedesignP F After Redesign Perform SafetyPerformSafetyPerformSafetyPerformSafety [-30,-10] e e-52.65e-5 [ 30,-10] e e-61.49e-56.65e-6 [-30, 10] e e-51.69e-7 [ 30, 10] e e-69.16e-86.65e-6 Mean e e-5 IntroductionMethodsResultsConclusion 14 / 19
Structural & Multidisciplinary Optimization Group Tradeoff Curve: ±30% Calculation Error / ±10% Measurement Error e calc ~ U[-0.3, 0.3], e meas ~ U[-0.1, 0.1] Redesign for performance yields about 3% lower mass P re = 20% Probability of Redesign, P re (%) Mean Area, E(A) (mm 2 ) Performance Safety IntroductionMethodsResultsConclusion 15 / 19
Structural & Multidisciplinary Optimization Group Mass distribution after redesign Too conservative designs are redesigned to save mass (a large reduction in mass) Unconservative designs are redesigned to satisfy required P F For performance For safety IntroductionMethodsResultsConclusion 16 / 19
Structural & Multidisciplinary Optimization Group PF distribution after redesign Redesign for safety (right) significantly changes low PF, while the change is relatively small for performance For performance For safety IntroductionMethodsResultsConclusion 17 / 19
Structural & Multidisciplinary Optimization Group Conclusions Starting with a conservative initial design pays off on average on the long run Redesign to reduce mass is like winning the lottery There is a small chance of very large mass reduction when redesigning for weight, but most designs are heavier Poor impression on initial design: a large mass change in redesign Redesign to improve safety will often result in lower mass designs and when redesign is needed the increase in mass will be small Good impression on initial design: design may fail but requires small change But for company’s point of view, conservative design is better Redesign for performance yields less average weight IntroductionMethodsResultsConclusion 18 / 19
Structural & Multidisciplinary Optimization Group THANK YOU Questions? IntroductionMethodsResultsConclusion 19 / 19