Fault class prediction in unsupervised learning using model-based clustering approach

Manufacturing industries have been on a steady path considering for new methods to achieve near-zero downtime to have flexibility in the manufacturing process and being economical. In the last decade with the availability of industrial internet of things (IIoT) devices, this has made it possible to monitor the machine continuously using wireless sensors, assess the degradation and predict the failures of time. Condition-based predictive maintenance has made a significant influence in monitoring the asset and predicting the failure of time. This has minimized the impact on production, quality, and maintenance cost. Numerous approaches have been in proposed over the years and implemented in supervised learning. In this paper, challenges of supervised learning such as need for historical data and incapable of classifying new faults accurately will be overcome with a new methodology using unsupervised learning for rapid implementation of predictive maintenance activity which includes fault prediction and fault class detection for known and unknown faults using density estimation via Gaussian Mixture Model Clustering and K-means algorithm and compare their results with a real case vibration data.