AI Driven Predictive Maintenance System Using Time Series Analytics for Smart Manufacturing Environments

Unplanned downtimes in smart manufacturing systems result in significant economic losses, making it imperative to move away from reactive and planned maintenance practices toward intelligent and condition-based maintenance practices. In this paper, we propose a predictive maintenance approach based on multivariate time-series analysis and deep sequential learning for predicting machine degradation and remaining useful life. Our proposed approach makes use of a temporal convolution-recurrent skip attention network (TCRSANet) model in combination with a spatio-temporal graph model to account for inter-sensor relationships. We provide a framework that comprises of a complete pipeline including data collection, preprocessing, feature extraction, degradation modeling, and maintenance decision-making with pseudocode for each step of the process. The performance of our approach is evaluated using a benchmark turbofan degradation dataset, where it outperforms the baselines of recurrent and convolutional models in terms of root mean square error of remaining useful life prediction while ensuring computational efficiency suitable for edge devices.