Document Type: Research Paper
Department of Industrial Engineering, Tarbiat Modares University, Tehran, Iran
Failure of machines, due to stopping the production line, results in financial losses. Preventive maintenance, significantly extends the machineries life, and reduces the costs. On the other hand, predicting the remaining useful life (URL) of the equipment and machineries, provides adequate time for maintenance engineers to repair or replace the parts before failure occurs, and avoid the overhaul costs (conditional-based maintenance). These actions are more important for rotary machines such as turbines, pumps and compressors, than the others. Hence, in this paper, we predict the URL of the Olefin unit of Pars Petrochemical Company turbine pumps based on the bearings health by artificial neural networks (ANN) and support vector machine. First, we provided the prediction model by the RMS, mean, peak and crest factor of one bearing, which was used to estimate the RUL of the four bearings using the above methods. Results showed that the accuracy of prediction by SVM method was more than single-layer ANN.