[3C2] Guided wave detection of metallic defects using thin-film sparse arrays and convolutional neural networks
C Dick, M McInnes, D Irving and D Hughes
Novosound Ltd, UK
Guided wave inspection is a quickly and ever-expanding area in condition monitoring, providing both long-range detection and wide-area coverage. However, for inspections requiring high resolution of defects, wide-area high resolution remains a persistent challenge. With the recent rise in powerful signal processing methods utilising artificial intelligence, complex inspection techniques, such as guided wave, are becoming far more accessible. This work demonstrates the capabilities of a permanently installed sparse array on the surface of a steel structure to perform wide-area, high-resolution defect detection. This was achieved via a combination of guided wave inspection and by use of convolutional neural networks. Datasets were acquired for a discrete number of conditions, of progressively worsening damage, and passed as labelled training data to a four-layer convolutional neural network. The output of the network is condensed to a single neuron allowing for binary classification. Defects were also identified analytically, with distinct phase changes noticed on receive sensors local to defects. Overall, the model achieved a classification accuracy of upwards of 95% in multiple out-of-sample datasets. Moreover, with increasing dataset complexity, a recurrent neural network approach is also presented to allow continuous monitoring and prediction of asset integrity degradation over time.
