Hybrid Deep Neural Network with CNN and RNN alongside 1st order B-spline Differential-Based Methodology for Real Time Fatigue Crack Growth Rate Monitoring Using Only AE Sensors

Prediction of crack growth rate (CGR) is crucial for the reliability and health management of complex structures (e.g., ships, bridges, aircraft). An autonomous prognostic and health management system is required in the shipping industry for safe and cost-effective maintenance of critical components. This study utilizes a hybrid deep neural network (DNN) comprising a one-dimensional convolutional neural network (CNN) and long short-term memory (LSTM) to predict CGR in DMR 249A high-strength low-alloy ship steel. Three fatigue tests with load ratios (R) of 0.1, 0.25, and 0.5 corresponding to sea-state IV conditions were conducted. The raw AE data was pre-processed with B-spline basis functions and first-order derivatives to enhance CGR prediction accuracy. The predicted crack growth rate achieved a correlation of 79.59%, 86.00%, and 95.97% for load ratio R 0.1, 0.25, and 0.5, respectively with the actual CGR. This work demonstrates the feasibility of using raw AE signals for online CGR determination with the proposed hybrid DNN model.