Printed PZT Sensors and Machine Learning for Intelligent Fault Diagnosis in EMS Trays

Fault diagnosis in highly automated and sealed EMS trays presents significant challenges. This study proposes a novel method utilizing printed lead zirconate titanate (PZT) strain sensors combined with machine learning. Micro/nanoscale PZT structures were fabricated via printing, and the impact of annealing on their crystallization, microstructure, and resultant piezoelectric/dielectric properties and impedance characteristics (for both thick films and 3D structures) was investigated. A neural network model was developed, with its hyperparameters (weights, thresholds, and topology) optimized using a Bayesian approach. Comparative analysis of model performance demonstrated the method's effectiveness in achieving accurate fault diagnosis for EMS electric vehicle systems, providing valuable theoretical and technical support for their detection, operation, and maintenance.