Forecasting faults of industrial equipment using machine learning classifiers

This work presents a predictive maintenance methodology so as to forecast possible equipment stoppages (or faults) of an industrial equipment for anode production along with the fault type in real time, utilizing process sensor data from operation periods. The warning timeframe so as equipment stoppage to be predicted has been set by the process experts as far as possible before the incident occurs. For the forecasting, some widely used machine learning architectures are tested. The visualization of the features patterns and the simulation results show that a warning timeframe around 5-10 minutes before the incident occurs is a feasible goal.