Abstract
The flow restrictor, one of the critical elements of hydrostatic bearings, provides the self-adjusting capability of pocket pressure and stiffness. With the increasing usage of hydrostatic bearings, reliability requirements for flow restrictors have become more crucial. The purpose of this thesis is to provide a useful method to analyze and optimize the dimension, preload, stiffness and various design parameters of flow restrictors. The method to carry out this study is using artificial neural network (ANN) models. By using feedforward neural network, the ANNs are capable of estimating any function by massively connecting appropriate number of neurons and different kinds of layers. After data preprocessing and cross-validation, the network is more stable and workable for our system with non-linear characteristic. In this thesis the function of fluid restrictors affecting the performance of hydrostatic bearings is investigated both theoretically and experimentally. An apparatus is constructed, which integrates a sliding hydrostatic table with several flow restrictor connectors. With these connectors, various kinds of restrictors such as capillary or membrane type restrictor are assembled and installed due to individual requirements. Furthermore, the pressure, flow and temperature of each pad are measured on-line simultaneously. Since limited experimental measurements are available to illustrate the properties and interaction between various parameters, and additionally, the theoretical derivation is complicated, non-linear and hard to be verified, the proposed neural network structure is demonstrated very suitable to model the flow restrictor design problem. By means of a well-trained network, the design parameters and the data with the uncertainty of the derivation had been successfully correlated.