Prediction of Hydrodynamic bearing performance based on effective parameters by Neural Network

Condition monitoring of rotary device is one of the major concerns which all industries have followed the new method for
preventing unpredicted failure. In this way predictive maintenance (PM) is the most significant part of our industries. In this
work, the main objective is to analyze the variation of three parameters including, Shear stress, Power losses and RMS (as an
acoustic emission feature) on hydrodynamic journal bearing under different lubricants for various loading conditions and various
rotational speeds. The results obtained experimentally from variation of parameters at different levels lead to find the relationship
between output parameters and input factors. Artificial neural network (ANN) by using multilayer perceptron algorithm is applied
to process the set of large number data from the test, with 80% used for training and 20% used for testing the predicted model.
In addition to this, 20% of real data have been applied for test of the mentioned network. The accuracy of predicted model is
about 0.001. The results show that the presented model from neural networks, constituting methodical basis for the control and
diagnostics the bearing without prior knowledge of the relative rotational speeds or load conditions can be predicted with
reasonable accuracy which hitherto has not been explored. Also, this method can utilize for the Interpolation of parameters which
cannot be tested in real condition for the assessment of behavior of output parameters.