CNN based structural health monitoring using EMI-PZT in a deep reinforcement learning

A growing number of industrial sectors are showing an interest in structural health monitoring (SHM) as a result of component material shifts from metals to plastic or composite. Damage in the materials can’t be seen with naked eye, making detection challenging. Exciting advances in piezoelectric materials, sensors for SHM have been made in the recent several decades. The monitoring systems begun to use CNN for primary emphasis on vibration analysis, but aren’t many cases when CNN is conjunction with electromechanical resistance method on SHM systems. Here, we provide a novel SHM strategy that encourages this combination by combining the EMI-PZT and CNN in DRL environment. A red, green, and blue frame is formed by dividing the EMI signature into its component elements and then determining the geometric distances between them. Consequently, we provide a dataset of 720 frames’ worth of EMI-PZT signals as four different structural circumstances. An experimental evaluation of the CNN-based approach was conducted in a case study using three PZTs adhered to an alloy plate. The findings demonstrate successful pattern categorization with 96.4% ± 2.1% accuracy validated through 5-fold cross-validation, with an independent test set achieving 94.8% accuracy. Comparative analysis shows the proposed method outperforms traditional EMI indices (RMSD, MAPD) and classical machine learning approaches.