Investigation of Artificial Neural Networks Capability to Predict Viscosity of Lubricants Used in Journal Bearings

Lubrication is one of the essential parts of various processes and instruments namely rotational devices such as bearings. Artificial intelligence (AI) systems such as artificial neural networks (ANNs) have been applied to solve, predict and optimize in engineering problems in the last decade. Present study assesses the capability of artificial neural networks (ANNs) in estimating viscosity of lubricants used in the lubrication of journal bearings. Using neural networks instead of running various tests to predict lubricant viscosity reduces costs and eliminates the necessity of using various devices and instruments by the researcher. The data of SAE 10, 20, 30, and 40 lubricants were used to train the neural networks. Back propagation (BP) error algorithm employed to networks training. Also, lubricant temperature was used as the input and its viscosity as the output of the simulation. In order to increase the accuracy of network training, a separate network was learned for each grade of lubricant, and the structure of each neural network and the effective parameters were optimized for each lubricant type through trial-and-error so that they would make more accurate predictions. The results of this simulation show that the error of neural networks in estimating lubricant viscosity in SAE 10, 20, 30, and 40 lubricants are in very small range, which indicate the capability of neural networks in the estimation of the desired parameters.