VIBRATION-BASED ROBUST DAMAGE DETECTION UNDER ASSEMBLY-INDUCED UNCERTAINTY: THE OUTPUT-ONLY CASE

摘要

The problem of unsupervised robust vibration based damage detection under assembly- induced uncertainty is considered with a focus on the non-measurable excitation (output-only) case. The study employs a composite aerostructure and damage scenarios based on bolt loosening and characterized by 'minor' effects on the dynamics which are appreciably 'masked' by those of assembly-induced uncertainty. Detection is based on two Statistical Time Series type methods: a Multiple Model (MM) based and a Principal Component Analysis (PCA) based. The results, based on hundreds of test cases, indicate that both methods achieve very good detection performance, with the PCA-based one exhibiting an edge over its MM-based counterpart (99.5% versus 96% correct detection at a 5% false alarm rate). Although the non-measurable nature of the excitation unquestionably increases the problem difficulty, the degradation in performance is quite small, especially for the PCA-based method (from 100% to 99% at 4% false alarm).