An LSTM approach for fault prediction

Hornyák, Olivér (2025) An LSTM approach for fault prediction. In: Proceedings of the International Conference on Formal Methods and Foundations of Artificial Intelligence. Eszterházy Károly Katolikus Egyetem Líceum Kiadó, Eger, pp. 90-101. ISBN 9789634963035

Preview

Text fmfai2025_pp090-101.pdf - Published Version Download (824kB) | Preview

Abstract

Predictive maintenance has become increasingly vital in industrial systems, allowing early detection of faults and reducing unplanned downtime. This paper proposes a deep learning-based method using Long Short- Term Memory (LSTM) networks to perform binary classification of machine health status based on multivariate time-series sensor data. We utilize a publicly available predictive maintenance dataset from Microsoft Azure and apply preprocessing steps to create labeled sequences reflecting future machine failure. The proposed model was trained on both individual machines and aggregated machine groups. Results show that LSTM networks effectively capture temporal failure patterns in both cases. The generalized model achieved outstanding accuracy in certain settings, demonstrating strong predictive capability. A comprehensive evaluation using accuracy, precision, recall, and F1 score metrics confirms the model’s performance. Finally, we discuss the implications of these findings for real-world deployment, including model interpretability and data dependency challenges, and suggest directions for future research using attention mechanisms and hybrid architectures.

Actions (login required)

Edit Item