A hybrid RNN-fuzzy-PSO model for forecasting multiple transverse cracks in laminated composite beam-like structures

This study investigates the use of a hybrid artificial intelligence (AI) model combining Recurrent Neural Networks (RNN), Fuzzy Inference (FI), and Particle Swarm Optimization (PSO) for predicting the position and severity of multiple transverse cracks in glass-fiber-reinforced-laminated-composite (GFLCB) beams. To assess the model's accuracy, a verification was conducted using a GFLCB intact and a double cracked beam. Finite Element Analysis (FEA) was employed to determine the first three relative natural frequencies (RNFs) under double crack conditions. Then the obtained RNFs are used to train the programs to locate the crack location and depth. The supremacy of the model over mPSO, RNN, and RNN-mPSO is verified. The maximum error percentage in calculation of first crack location by RNN-FUZZY-PSO, mPSO, RNN, and RNN-mPSO is found to be 3.08%, 5%, 7.2%, and 8.11% respectively. While in detection of crack locations, the error percentage are 1.1%, 5%, 7.2%, and 8.11%, respectively. Further RNN-FUZZY-PSO is used to diagnose the crack severity and locations of multiple cracks (nine crack) in hybrid GFLCB. Results indicated that the RNFs are significantly influenced by the number and severity of the cracks. The predicted crack positions and severities by the method are with a marginal error of 1.53% and 1.3%, respectively. The model shows improved accuracy as the number of cracks increased, especially for the ninth crack, where the mean square error is 0.01154, with maximum error percentage of 0.8%. The findings demonstrate the proposed AI model's effectiveness for precise identification of crack positions and severities in GFLCB structures with multiple cracks.