A Multi-Sensor Health Indicator Construction Method Based on Principal Component Analysis and Variational Autoencoder

Prognostics and health management play a crucial role in ensuring both the reliability and economic benefits of modern industrial equipment. However, the rapid increase and complexity of multi-source heterogeneous sensor data pose significant challenges to equipment health assessment. Effectively processing these data and extracting valuable information for accurate health assessment have therefore become urgent problems. In light of this, this paper proposes a prognostics and health management method based on constructing health indicators. Specifically, a feature-level fusion approach is introduced, combining principal component analysis and variational autoencoder with a weighted fusion strategy, thereby generating composite health indicators that accurately capture equipment degradation trends. Experimental results show that the constructed health indicators possess stronger correlation and clearer trending characteristics, providing a comprehensive and precise representation of equipment health and offering vital support for health monitoring and predictive maintenance in complex systems.