Constructing high-quality health indicators from multi-source sensor data for predictive maintenance applications

Prognostics and health management are critical for ensuring the reliability, safety, and economic efficiency of modern industrial equipment. However, with the growing volume and diversity of multi-source sensor data, effectively processing these data and extracting valuable information to accurately assess equipment health remains an urgent challenge. In response, this paper proposes a novel prognostics and health management approach based on health indicator construction. By integrating the nonlinear feature extraction capability of kernel principal component analysis and the deep representation learning strength of deep autoencoders, significantly enhancing the expressiveness of the constructed health indicators. Furthermore, a stochastic degradation model based on the Wiener process is incorporated with the health indicators to provide dynamic, uncertainty-aware estimation of the remaining useful life. Based on the predicted remaining useful life distribution, a cost-driven maintenance decision-making strategy is proposed to optimize maintenance timing. Experimental results obtained on the C-MAPSS dataset demonstrate significant improvements in prediction accuracy and provide a robust decision-making framework for predictive maintenance. These findings highlight the potential of the proposed method to enhance industrial reliability while reducing maintenance costs.