Use of truncated Fisher sensitivity analysis for design verification of mechanical systems under uncertainty

Sensitivity analysis is becoming an essential part of simulation based engineering design, but few sensitivity methods can directly consider measurement data for design verification. Here we present a new data-driven method, based on the truncated Fisher Information Matrix (tFIM), to identify the key drivers of the discrepancy between simulated and measured frequency response function and to guide the design verification process. We found tFIM is as effective as the more commonly used Monte Carlo Filtering (MCF) for design verification, but offers additional insights to sensitivity information from parameter interactions and from within the truncated region. To overcome the non-desirable issue of a fixed truncation, we discuss the augmentation of the input parameters by random truncation thresholds. Application of tFIM to the dynamics verification of a model floating wind turbine successfully identifies the most important parameter out of 15 input random variables, and the sensitivity guided design update is parsimonious and interpretable. Thanks to its efficiency as a one-sample approach, we expect the new tFIM method to complement global uncertainty and sensitivity analysis and become an integral part of design verification for the dynamic performance of mechanical systems under uncertainties.