An integrated mechanism and data model for adaptive wear state diagnosis via moving wear particles

Moving wear debris analysis (M-WDA) serves as a pivotal means for online wear state diagnosis of friction pairs. However, the diagnosis accuracy has been hampered by two major challenges: redundancy of particle indicators and high randomness in particle generation. To address this issue, an adaptive wear state diagnosis model (AWSD) is developed that integrates wear rate and wear mechanism via the structured modeling of particle indicators. Considering the redundancy in particle information, a random forest based selection strategy is constructed to refine the particle indicators by evaluating their significance. On this basis, a three-layer structure encompassing indicator-attribute-state is proposed for wear state characterization, and then applied to guide the neural network modeling for adaptive wear state diagnosis. With this methodology, wear rate and wear mechanism are integrated to mitigate the uncertainty that stems from the randomness of particle generation. For verification, the constructed model is tested using aero-engine particle samples under various operating stages, and the average diagnosis accuracy of wear states has been improved from 72.5 % to 95 % when compared to the existing methods. Additionally, the proposed AWSD model is employed to analyze the particles in accelerated rolling-sliding friction tests and identifies fatigue wear as the primary wear mode of bearing rollers.