Benchmark of Unsupervised Machine Learning Algorithms for Condition Monitoring

Abstract

Predictive maintenance and condition-based monitoring technique used to monitor the health of bearings, pumps, turbine rotors, gearboxes, etc. It uses the idea of data mining, statistical analysis, and machine learning technique to accurately predict early fault of mechanical components and calculate the remaining useful life. The paper is about condition-based health monitoring of heavy engineering equipment and their predictive maintenance. Data is gathered from the bearing of our experimental setup using unsupervised learning on type of failure and remaining useful life should be determined to predict the maintenance of a machine. In this paper, we consider a data collected from the bearing and fit different unsupervised learning algorithm, gaussian mixture model and clustering technique to check its performance, accuracy, and sturdiness. In conclusion, we have proposed a methodology to benchmark different algorithm techniques and select the best one.