Fault prediction of drag system using artificial neural network for prevention of dragline failure

Undetected faults in heavy earth moving machineries (HEMMs) are linked with high maintenance cost and downtime losses. Draglines are capital intensive HEMMs used in surface coal mines for stripping overburden. It is revealed from the annual maintenance worksheet of three draglines that drag system failures alone contributed to 49% of the total downtime of draglines. However, inclusion of an effective fault prediction methodology in the preventive maintenance policy of the drag system is expected to reduce most of these failures and downtimes of the dragline. This paper demonstrates a data-driven approach for predicting faults in drag system using multilayer perceptron (MLP) in artificial neural network (ANN) using past two years cause, symptom, and fault data recorded through the sensor, logbook, and visual inspection. A total of 452 data when symptoms exceeded the threshold limit was observed; out of which there were 199 faults that led to 16 failures in the drag system, and these data have been transformed to categorical data. Two ANN models have been developed to understand the fault occurrence process using seven causes, seven symptoms and five fault parameters of drag system. The prediction accuracy of symptoms using the cause was 94.2% and that of fault using symptom was 97.1%. The sensitivity of causes and symptoms were ranked for individual faults that can help the maintenance engineer to predict the faults and to make the sequence of preventive maintenance action plan in order to minimize the unwanted downtime and maintenance cost of the dragline.