Guided wave (GW) ultrasonic testing is an effective tool for detecting damage on pipe structures because guided waves can have 100% volume coverage. In conventional GW-SHM, damage can be detected by subtracting a measurement from a baseline record, after properly compensating for any temperature difference between the tests. However, compensation methods often cannot perfectly remove the benign variations produced by complex environmental and operational conditions (EOCs), leaving residual noise that can mask the damage signal. With the recent advances in computing, it has become feasible to process a batch of historical records, and data-driven approaches have been developed to detect small damage signals in the presence of EOC variations more robustly. In the presentation associated with this paper, we evaluate two recently-developed, data-driven damage detection methods: singular value decomposition (SVD) and independent component analysis (ICA). We implement the two methods to process synthetic dataset that contains superposition of experimental GW records collected at varying environmental conditions, and artificial damage signals at various locations. Such a synthesis process enables us to investigate the performance of damage detection at different EOC conditions without damaging the pipe, which is prohibitively expensive if a large number of scenarios are to be investigated. We then validate the results using experimental GW records taken from an industrial scale pipe system, with a flat-bottom hole drilled gradually to a maximum of 0.5% of the cross section area. We compare the performance of the SVD, ICA, and the conventional baseline-subtraction method in two aspects. First, we evaluate how well the extracted damage feature resembles the true damage signal in terms of location and amplitude, by using receiver operating characteristics, which plots the probability of detection against the probability of false alarm. Second, we evaluate how well the damage features extracted from sequential ultrasonic measurements track the true progression of the damage, by using statistical trend analysis. This paper sets out the methodology used and ROC curves using the residual method; the final results including comparisons with ICA and SVD will be presented in the talk and published later.
展开▼