High Efficiency Analysis Of A Elastohydrodynamic Tilting Pad Thrust Bearing With Boundary Slippage Using A Deep Residual Network Without Batch Normalization

Thrust bearings as mechanical components are designed to carry axial loads in axial heavyduty rotating machinery.Due to the excellent characteristics of the tilting tile bearing,it has been widely used in machinery,water conservancy,electrical,chemical,shipbuilding and other industries.In recent years,with the emergence of high speed,high load and high power machinery,tilting tile thrust bearing working environment is more and more severe,which led to the increase in the rate of bearing accidents.Boundary slip thrust bearings have higher load and lower power loss compared to bearings without boundary slip,so further research on tilting tile thrust bearings with boundary slip is very important,and can provide theoretical basis for the design,manufacture and optimization of tilting tile thrust bearings with boundary slip phenomenon.In order to study the lubrication performance of the tilting tile thrust bearing considering boundary slip,while avoiding the multiple iterations and high time cost of the traditional finite element analysis method,a depth residual network model without batch normalization is proposed in this paper.The model is used to predict the tilting angle of the tilting tile thrust bearing with different surface materials,the thickness of the fluid film at the support point,and then analyze the influence of different parameters on the lubrication performance of the tilting tile thrust bearing by calculating the parameters obtained from the depth residual network.In this paper,we first construct a deep residual network without batch normalization and compare and analyze the effect of different numbers of hidden layers on the performance of the normal fully connected network and the deep residual network.After proving that the deep residual network outperforms the normal fully connected network,we continue to verify and analyze the effect of the number of samples and the number of batches on the performance of the deep residual network.It is proved that the deep residual network with more samples and smaller batches has better prediction accuracy and better fitting effect.Then,the depth residual network with the validated network structure and parameters was used to predict the tilting tile thrust bearing of polymer surface material and the tilting tile thrust bearing of rigid material surface in terms of tilting angle and liquid film thickness at the support point,and the results showed that the depth residual network can accurately predict the parameters of the tilting tile thrust bearing of both materials.Based on the high-precision prediction of the deep residual network,the lubrication performance analysis of the tilting tile thrust bearing was carried out,separately for bearings with different material surface materials,namely,bearings with polymer material surface and bearings with rigid material surface.The lubrication performance analysis then includes the effect of ultimate shear stress on the lubrication performance of the bearing and the effect of load on the lubrication performance of the bearing,in which the effect of ultimate shear stress and load on the minimum fluid film thickness,power loss,fluid film thickness distribution and fluid film pressure distribution are mainly analyzed.The results are also compared and analyzed with those obtained by applying the traditional finite element method to prove the accuracy and reliability of the depth residual network proposed in this paper.