Improving the Stiffness of Hydrostatic Bearings Using Multilayer Perceptron

摘要

The stiffness of hydrostatic bearings is mainly affected by the flow resistance of the restrictor, however, accurate estimation of which is often unattainable because of variation of environment conditions, resistance from oil tube and incompatible assumptions in the theory of hydrostatic bearings. This paper proposed a design method to improve the stiffness of hydrostatic bearings by use of multilayer perceptron (MLP). The MLP model constructed a multi-input and multi-output (MIMO) system with supply pressure, load, and the depth of groove as the inputs and the oil-film thickness as the output. The MLP model employed gradient decent algorithm as the optimizer with an input layer, three hidden layers, and an output layer. According to this malleable nonlinear model and various functions, the MLP model could find the hidden patterns from the training data and predict the output. Simulation of bearing characteristics was performed on the basis of the hydrostatic bearing theory. An experimental setup was constructed to verify the film thickness obtained from both simulation and predictive results of the MLP model. A number of flow restrictors with distinct groove depths together with parameters such as supply pressure and load were used in experiments. Meanwhile, the pressure, flow rate, load, temperature and oil-film thickness were measured by the corresponding sensors directly. The MLP model for the stiffness of hydrostatic bearings was effectively trained with the collected data. Compared to the simulation, the proposed method demonstrates more applicable for the design of hydrostatic bearing systems.