Machine Learning Applied to Machinery Fault Prediction
Despite the potential to reduce operational costs and the increase in the availability of the computational power demanded by this task, machinery health prognostics still unavailable to many companies either due to the lack of human resources or due to the costs involved. Here we propose an approach and perform some validation experiments of predicting an imminent failure without needing a priori information about the monitored system. The prediction is achieved by continually clustering sensor data using unsupervised ML techniques and analyzing the evolution in time of those clusters in real-time. Initial validation tests using public available datasets showed encouraging results, giving in some experiments alerts about future failures within the desired time frame in the majority of runs while avoiding false-positives, pointing that this approach can be useful especially when it is not possible to use a more precise physical model or expert knowledge.