We have presented an overview of our most recent efforts towards the development of a comprehensive validation framework for design of vehicle systems. We begun our research work with evaluating existing measures for comparing multivariate functional data (i.e., time histories) since: (1) vehicle systems are dynamic with multiple time-dependent responses, and (2) comparison between computational and test data is the basis of validation. After demonstrating several limitations of traditional measures, we introduced a deterministic, three-component validation measure that can be used in conjunction with subject matter expert data to build regression-based validation metrics. Recognising that the issue of uncertainty in both tests and computational models must be addressed inevitably, we proceeded to investigate the applicability of Bayesian methods for design validation. After processing the available multivariate data through PPCA (probabilistic principal components analysis) for efficient uncertainty treatment, dimensionality reduction and significant feature (or information) extraction, we quantify confidence using Bayesian interval-based hypothesis testing. We developed a robust implementation of this method by introducing a coherent variance calculation method and a technique for determining the appropriate interval width based on information content. Lastly, we employed the implemented confidence quantification method in a novel sequential design optimisation approach with simultaneous calibration-based validation. Our motivation was that typical a-priori, global model validation for design optimisation is only adequate when unlimited resources are expended to sample the high-dimensional design variable and parameter space, which is obviously not the case due to limited resources and/or time restrictions. We have demonstrated that the use of such models can lead to both invalid and suboptimal results and have shown that sequential optimisation with calibration-based (when necessary) validation can use resources and associated information effectively. Our current efforts focus on completing the comprehensive framework for simulation-based design validation by developing the missing techniques for determining the minimum amount of tests necessary at each stage of the sequential method the size and shape of the associated local trust regions. Future work could focus on the development of a methodology that utilises the data and model calibration parameter values generated during the sequential process to build confidence metamodels that can used for extrapolation in domains where data cannot be obtained.
Towards a comprehensive framework for simulation-based design validation of vehicle systems
International Journal of Vehicle Design ; 61 , 1-4 ; 233-248
2013
16 Seiten, 12 Bilder, 25 Quellen
Article (Journal)
English
Towards a comprehensive framework for simulation–based design validation of vehicle systems
Online Contents | 2013
|Towards a comprehensive framework for simulation-based design validation of vehicle systems
Automotive engineering | 2013
|Towards a Comprehensive Framework for Display Compatibility
British Library Conference Proceedings | 2012
|Comprehensive Framework for Integration of Calibration, Verification and Validation
British Library Conference Proceedings | 2012
|