Deep learning-guided optimization of piezoelectric sensor placement for high-fidelity mechanical damage localization

Structural health monitoring (SHM) plays an important role in the safety, reliability, long life and early localization of mechanical damage in engineering structures. This research proposes a deep learning-based sensor placement optimization method with high damage localization accuracy and fewer sensors. The proposed framework combines finite element-based guided-wave simulation, piezoelectric sensing, signal processing, convolutional neural network (CNN) based feature extraction and sensor placement optimization into a single workflow. Damage-sensitive features are autonomously extracted by the CNN from guided-wave signals, without manual feature engineering, which improves localization performance. The positions of sensors are optimized for maximum observability and prediction accuracy of the structure. The framework was extensively simulated with cracks, corrosion, and delamination under various operating conditions to validate the framework. The localization accuracy, localization error, precision, recall, F1-score, computational time, sensor reduction ratio were used to evaluate the performance. Simulation-based validation demonstrated a localization accuracy of 98.91%, a mean localization error of 2.14 mm, and a 40% reduction in the required number of sensors compared with conventional uniform sensor placement. The results presented suggest that the proposed framework is efficient, scalable and scalable for structural health monitoring, but experimental validation on physical structures is pointed as another relevant direction for future research.