An IoT-based Failure Prediction Solution Using Machine Sound Data

Industry is seeking to benefit from the advantages of integrating IoT solutions in daily operation to achieve smart manufacturing as well as smart maintenance. As a matter of fact, maintenance cost is already a big part of the budget. Despite the integration of IoT in industry, cost remains high due to the complexity of the developed solutions. In this paper we are proposing an IoT-based failure prediction solution architected to reduce the cost of maintenance. Our proposal permits a highly accurate failure prediction that leads to an effective action when there is an actual need for it. Furthermore, we could lower the cost of the solution itself by optimizing the data transmission between the sensor-nodes and the server. We conducted a demonstration of the proposed analysis scheme and the system design consisting on recording the sound data of a DC-motor for about 23 minutes with the variation of speed to mimic some failure scenarios. We obtained results that confirmed the effectiveness of our solution in differentiating between the failures signs with no prior learning of the failures and in tracking the slight drift in the machine behavior. We were able to start predicting failures since day 1. The proposed system design permitted us to limit the payload of data packets which would reduce the cost of the sensor-node data transmission, the power consumption in the sensor node as well as the network traffic.