ABSTRACT As modern equipment becomes increasingly complex and long‐lasting, statistical data on mechanical products, especially mechanical systems, remains relatively scarce. This scarcity poses significant challenges in accurately determining the distribution of fundamental random variables and design parameters of these devices. Therefore, the urgent scientific and engineering need is to develop methods for analyzing the health status of complex mechanical systems under conditions of limited data. In this paper, a health status assessment method for complex mechanical systems based on multi‐heterogeneous information is proposed. To address the issue of limited experimental data, the relationship between contact force and motion clearance under wear failure mechanisms is considered to develop a more realistic simulation data generation method. Additionally, the integration of simulation data and experimental data is studied to enrich training data. As for the issues of poor timeliness and lack of labels, this study investigates a health status assessment method based on a pair of models combining long short‐term memory (LSTM), attention mechanism, and autoencoders, while also offering a simplified version that enables health assessment using only simulation data Finally, a dual‐axis driving mechanism case demonstrates the effectiveness and robustness of the proposed method.
Fan et al. (2026) studied this question.