Analysis of the Variation in Temperature and Pressure for Journal Bearing by Applying the Back-Propagation ANN Learning Approach

Abstract

Journal bearing are of great importance for supporting the loads in rotating shafts. There are three most common types of journal bearing viz., dry, hydrostatic and hydrodynamic. The developed pressure and temperature conditions plays a significant role for the proper functioning of the bearing. In this paper ANN model approach has been employed for analyzing the developed temperature and pressure at different load conditions. In view of this, an investigative analysis of temperature and pressure (maximum) has been carried out for journal bearing. A neural network framework has been designed in order to carry out the investigative analysis (examination) for a number of different temperature and pressure conditions. The concept of ANN (artificial neural network) system has been derived from natural NN system and its working is also quite similar to that of natural phenomenon. As these artificial networks can predict the pressure of the system to be examined quite easily. The learning capacity is faster due to their novel (unique) parallel design in its structure. Thus, the structure of neural network can be employed for such types of analysis, as it is evident from the results that for modeling and simulating the real time application NN approach can be used. The investigative work has been carried out in two stages as discussed in the subsequent parts of this paper. The first stage is to accumulate the suitable data while second stage is applying the data for simulation purpose. This design data that is collected from DDB i.e., design data book like the variation in pressure in the first stage of the experiment is fed as input and is also used for the training in learning and for testing of data as per the corresponding or desired condition. The proposed NN is a feed forward having a network of three layers. The output that is obtained from the last layer of the ANN design during its training if having any deviations from the estimated or proposed output then the weights are adjusted or updated with the help of the back propagation so that the deviation could be minimized. Thus, while designing or developing any system of ANN, training is a prominent part which is only possible by the algorithm of back propagation as the weights are updated while doing the back propagation. In order to model the disturbances in loads for journal bearing Neural network predictor have been used.