Data Analytics to support Predictive Maintenance

Predictive maintenance is a technique that tries to predict imminent problems, forecast future failures and discovers criticalities when a piece of equipment might stop working so that proactive strategies can be applied just before that happens. These predictions can be done on the basis of equipment's condition which is estimated based on data collected with the help of condition monitoring sensors and strategies. To this aim, the predictive analytics has been measured to predict the belt tensioning level and to further support robot cycle labelling. A machine learning algorithm has been applied to smart data so as to forecast a tensioning level as per a new cycle of data. This helped in identifying clusters of production cycles through similar time independent features. The best configuration has been selected after comparing various supervised and unsupervised metrics on different clustering algorithms. And lastly, we talked about data distribution of patterns i.e. cluster characterization. The most relevant features of our clusters has been extracted. Cluster characterization helps domain expert in labelling of the data, in our case to robot cycles. Keywords: Predictive maintenance, machine learning, clustering algorithms, Cluster characterization