Deep Learning based Crack Growth Analysis for Structural Health Monitoring

This paper presents an ensemble deep learning (DL) based structural health monitoring approach for complex systems. The proposed methodology consists of crack detection and crack growth prediction. An ensemble DL-based image segmentation technique, which is ResNet-UNet, has been developed to detect the existence of a crack pattern. The ensemble technique has very good performance in image classification, object detection and image segmentation problems. Once a crack has been detected from the image, the same image is put forward into crack length extraction phase. The pixel-wise crack length extraction technique tries to extract crack length via counting the binary pixel values corresponding to the crack region. The ARMA time series forecasting model is then trained on crack length feature to estimate remaining-useful-life (RUL) of crack surface. The proposed approach has been validated on airplane fuselage data set. The proposed approach is very promising in structural health monitoring of complex systems. Copyright (C) 2022 The Authors.