Design and Development of a Neoteric Predictive Maintenance Technique Implemented through Comparative Analysis of ML Algorithms to reduce machine failure

Abstract

Predictive maintenance uses software and historical data collected using sensors to prevent machine failure by analyzing the production data, to identify the operating pattern and thus predict issues before they commence. Predictive maintenance is the process of continually gathering and transmitting the behavior data of the product , storing the data in a central storage hub, and applying analytics methodologies available in advanced big data to sort  the massive amount of data so as  to identify important data pattern.  The data is fitted into machine learning models and trained with the past data to successfully predict the probable failure of the machine. Five machine learning algorithms such as Support Vector Machine, K-Nearest Neighbor, Random forest, Naive Bayes and the Stochastic Gradient Descent are implemented on the real time data obtained from the machines using sensors. Three models with highest efficiency of prediction are taken to predict machine status collectively. The machine status which is predicted by the majority of the machine learning models is taken as the final machine status.  The data of whether machine failure will occur or not is constantly being updated in a real-time dashboard and is made accessible to the workers. The dashboard for the generated insights is created to allow maintenance engineers to perform corrective action.