Uncertainty-involved evaluation of real-time lubrication states in tilting pad thrust bearings

Lubrication failures in tilting pad thrust bearings arise from the lubrication state degradation. Therefore, real-time evaluation of the lubrication state is critical for monitoring and prediction of lubrication failures. Traditional methods, such as those based on the Stribeck curve, have slow response times and industrial implementation challenges. In contrast, the lubricant film thickness relative to the composite roughness of two contacting surfaces, commonly referred to as the lambda ratio, provides a faster assessment of lubrication states, but it struggles to accurately identify the transition zones between different lubrication states. To address this issue, this paper presents an uncertainty-involved evaluation method of real-time lubrication states in tilting pad thrust bearings. Extensive lubrication state degradation tests are performed on a thrust pad bearing test rig. By leveraging the synchronous variation between the coefficient of friction and the lambda ratio, the maximum likelihood estimation technique is employed to establish threshold sets for the lambda ratio. Additionally, kernel density estimation is used to derive the probability distribution functions for these threshold sets. Through the integration of cumulative probability and probability assignment, a probability characterization method of lubrication states is developed and validated with experimental data, achieving 88% identification accuracy for three-stage classification and 83% overall accuracy for four-stage classification across diverse operating conditions.