Abstract

Predictive maintenance (PdM) is a data-driven approach to maintenance that uses machine learning (ML) to predict machine failures before they occur. This can help to reduce downtime, improve equipment availability, and extend the useful life of components. In this paper, we present a PdM system using LSTM, XGBoost, and fbprophet. We evaluate the performance of these models on a dataset of temperature and humidity data collected from sensors connected to a Raspberry Pi. The data was streamed to AWS Greengrass and then used to train the ML models. The results show that all three models achieved high accuracy and precision, with LSTM performing the best. The LSTM model was able to predict anomalies with an accuracy of 98% and a precision of 95%.

References

1. P. Poór, J. Bašl, and Z. Zenisek, “Predictive maintenance 4.0”, in 2019 IEEE 1st International Conference on Industrial Cybernetics Systems (ICICS), pp. 22-27, IEEE, 2019.
2. M. Liu, J. Gao, and Z. Liu. 'A novel deep learning framework for predicting machine failure time.' IEEE Transactions on Industrial Electronics, 66(12):9869-9879, 2019.
3. L. Li, D. Xu, H. Wei, S. Li, and Y. Zhou. 'A hybrid deep learning approach for machine fault diagnosis.' IEEE Transactions on Industrial Electronics, 66(4):2979-2988, 2019.
4. B. Li, S. Zhang, Y. Ding, and J. Zhang. 'A hybrid deep learning model for remaining useful life prediction of bearings.' IEEE Transactions on Industrial Electronics, 67(12):10235-10245, 2020.

Download PDF
Back