Development of an enhanced predictive maintenance model for industrial systems using deep learning techniques
Communication in Physical Sciences · 15.05.2026 · 13 (1) · 124-138
DOI: 10.4314/cps.v13i1.10 ↗Kurz gefasst
Fachartikel in der Zeitschrift Communication in Physical Sciences, online erschienen am 15.05.2026. Verfasst von Confidence Ifeoma Odoh und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Predictive maintenance has become essential in modern industrial systems for reducing unplanned downtime, lowering maintenance costs, and improving equipment reliability. This study presents a hybrid deep learning framework that combines Long Short-Term Memory (LSTM) and Multilayer Perceptron (MLP) networks for accurate machine failure prediction. The model was trained using multivariate sensor data, including air temperature, process temperature, rotational speed, torque, and tool wear, enabling comprehensive monitoring of machine health. The hybrid architecture integrates LSTM’s strength in temporal sequence learning with MLP’s capability for nonlinear feature-based classification. Training results showed a steady reduction in loss and convergence in accuracy over 30 epochs, with the model achieving a training accuracy of 98.10%. During testing, the hybrid model achieved an overall prediction accuracy of 99.20%, outperforming standalone LSTM and MLP models. The system effectively detected multiple failure modes, including power failure, overstrain failure, and heat dissipation failure, while maintaining strong performance in distinguishing normal operating conditions. To demonstrate real-world applicability, the model was deployed via a Streamlit-based web interface for real-time monitoring and prediction. An integrated automated email alert system provided immediate notifications when potential failures were detected, supporting proactive maintenance decisions. Although minor performance variation was observed for less frequent failure categories due to class imbalance, the overall results confirm the robustness and scalability of the proposed framework. The findings highlight the significant potential of hybrid deep learning models in transforming maintenance strategies from preventive to data-driven predictive approaches, ultimately enhancing operational efficiency and system longevity in industrial environments.
Zitieren
APA
Odoh, C. I., Nweze, R. C., & Maduahonwu, U. V. (2026). Development of an enhanced predictive maintenance model for industrial systems using deep learning techniques. Communication in Physical Sciences, 13(1), 124-138. https://doi.org/10.4314/cps.v13i1.10
Vancouver
Odoh CI, Nweze RC, Maduahonwu UV. Development of an enhanced predictive maintenance model for industrial systems using deep learning techniques. Communication in Physical Sciences. 2026;13(1):124-138. doi:10.4314/cps.v13i1.10
BibTeX
@article{Odoh2026,
author = {Odoh, Confidence Ifeoma and Nweze, Rosemary Chika and Maduahonwu, Ukamaka Victoria},
title = {Development of an enhanced predictive maintenance model for industrial systems using deep learning techniques},
journal = {Communication in Physical Sciences},
year = {2026},
volume = {13},
number = {1},
pages = {124--138},
doi = {10.4314/cps.v13i1.10}
}
EUVIMED