Predictive maintenance (also known as PHM or equipment health monitoring') refers to the intelligent monitoring of equipment to avoid future equipment failures. In contrast to conventional preventive maintenance, the maintenance schedule is not determined by a prescribed timeline; instead, it is determined by analytic algorithms, such as machine learning, using data collected from equipment sensors and/or via simulation.Predictive maintenance offers the following benefits for operators and equipment manufacturers:reduction of equipment downtime by identifying issues before failure, thereby enabling convenient scheduling of equipment service and extending equipment lifetimeautomatic determination of the root cause of the failure, enabling appropriate service to be performed without utilising resources to determine a diagnosis avoidance of costs of unnecessary maintenance and unexpected failures improved safety and product quality.Algorithms and data are critical to predictive maintenance success. Sensor data preprocessing is performed using advanced statistical and signal processing techniques. In the case that data isn't available for the full range of operating modes, operating scenarios can be simulated. Machine learning techniques are then used to estimate equipment health. Once tested, predictive maintenance algorithms may be operationalised in an IT environment such as a server or cloud. Alternatively, algorithms may be implemented in an embedded system directly on the equipment, allowing for faster response times and significantly reducing the amount of data sent over the network.In this paper, we use both historical and simulated data of the plant under various modes of operation including run to failure, and we then leverage machine learning techniques in MATLAB to estimate degradation and remaining useful life. This technique, of combining both simulation models and machine learning, in the areas of predictive maintenance and equipment health monitoring has been adopted by Safran, makers of aircraft jet engines, since 2007. They now have over 160 modules developed (Lacaille and Aurelie, 2015)._x000D_
Citation: Willingham, D and Marchant, R, 2016. Predictive Maintenance Using Simulation and Machine Learning, in Proceedings 13th AusIMM Mill Operators' Conference 2016, pp279-284 (The Australasian Institute of Mining and Metallurgy: Melbourne).