Data-Driven Model of the Distribution Lubrication on Water-Lubricated Bearing under Severe Operating Conditions

To resolve the contradiction between the method used to design bearings based on traditional lubrication theory and the actual state of service of water-lubricated bearings (WLBs), this paper proposes a data-driven method for the model of the distribution of lubrication on WLBs. A full-sized WLB test bench featuring multi-sectional pressure due to the film of water and a system to measure the axis of the orbit was built to perform tests under severe operating conditions (75 kN, 25–220 rpm). A dataset of the operating parameters of the bearings was obtained based on the results of tests under varying operating conditions. An artificial neural network algorithm was applied to train the proposed model, and its capabilities of prediction and extrapolation were systematically analyzed by using samples with different ranges of values. The proposed model was then used to examine the distributed characteristics of lubrication of the WLB to investigate the effects of variations in speed and elevation on bearing performance. The results showed that it has satisfactory capabilities of prediction and extrapolation under the same elevation and variation in speed. Under severe operating conditions, two significant peaks of pressure of the film of water appeared at both ends of the WLB, and variations in the speed of the shaft and the elevation of the bearings had prominent effects on the state of distributed lubrication of the bearings. The results reported here provide a new approach to designing and optimizing the structure of WLB.