DEVELOPMENT OF A PREDICTIVE TOOL FOR MACHINE FAILURE TIMES USING ADVANCED DATA ANALYTICS METHODS

Abstract

Any company in the industrial sector requires constant and uninterrupted operation of its systems as it directly affects an organization's success. Both planned and unplanned line stoppages result in excessive production losses, thus imposing adverse effect on the firm’s profitability. The incorporation of the predictive maintenance practices enables firms to find a “golden mean” between conducting unnecessary scheduled repairs and preventing from unexpected downtimes due machine breakdowns. This study encompasses development of two predictive maintenance models using Machine Learning (ML) algorithms to increase the current system reliability of JTI’s company. One of the model is based on the Long Short-Term Memory (LSTM) to predict the emergence of next alarm, whilst the second model estimates the Remaining Useful Life (RUL) of the machines. The preliminary results revealed that both models showed an acceptable performance with an accuracy rate of 60% for the first model, and Mean Squared Error (MSE) of 134 minutes for the second model. Although the performance values of the models are not excellent, they are still meaningful and the methods are robust and can be implemented to the firm’s operations in order to facilitate maintenance decisions on key machines.