Intelligent Machine-Failure Prediction System (IMPS)

Abstract

In mission critical systems, system failure is a major hazard and may cause huge losses including loss or threat to lives. Organisations, industries, hospitals and companies can benefit hugely if an accurate prediction of the impending failure can be made, with enough time to initiate appropriate maintenance routines. Here, we propose and demonstrate that machine failure prediction can be done using suitable machine learning models with high accuracy. We apply the principles of Logistic Regression, Bootstrap Aggregation and Multinomial Logistic Regression to a predictive maintenance dataset of 10,000 data points to predict machine failure under five independent failure modes. Applying ensemble methods like bootstrap aggregation push the accuracy to greater than 99% The machine fails even if one failure mode is true. We are able to predict the possible cause of failure too, with a high accuracy of up to 99%.