Machine Learning Fault Detection for Piezoelectric Actuators in a Microassembly System

Piezoelectric actuators are essential for high-precision microassembly. However, monitoring their health state presents significant challenges due to their compact size and the inherent complexity of their modeling. This study presents a machine learning-based approach to predict failures in a micromanipulation system with four piezoelectric actuators. Log data was used to extract features, and autoencoders were trained on healthy-state data to detect deviations signaling faults. The method was validated using real-world training data and simulated failure scenarios, successfully distinguishing between healthy and faulty states. This approach offers a promising solution for monitoring the health of piezoelectric actuators in precision systems.