Prediction of dynamic characteristics of cracked plates repaired with composite patches using deep learning

This study investigates the dynamic properties of aluminum plates repaired with composite patches, focusing on how natural frequencies change after repairs on plates with various crack configurations. To predict these changes, an Artificial Intelligence (AI) model was developed using data from 2000 Finite Element (FE) analyses conducted in ANSYS. A Deep Neural Network (DNN) was trained using this dataset, which included 2000 damage scenarios based on four parameters: defect length ( L ), radius ( r ), depth ( h ), and the defect region. The optimal network configuration (11-[15-15-5]-4) was trained to predict the first four natural frequencies of these single-sidedly patch-repaired, cracked plates. The model’s accuracy was rigorously validated using an unseen test sample (Region 6, L = 35 mm, r = 2.5 mm, h = 5 mm) not included in the training data. The AI’s predictions were compared against both numerical results and an Experimental Modal Analysis (EMA). The results confirmed the high success rate of the model, showing minimal differences from the numerical simulations (max 0.39% error) and experimental results (max 3.15% error).