Machine learning enabled preventive maintenance strategy for improving boiler reliability of a thermal power station
1 Department of Industrial and Manufacturing Engineering, National University of Science and Technology, Bulawayo, Zimbabwe.
2 Department of Agricultural Engineering, National University of Science and Technology, Bulawayo, Zimbabwe.
Research Article
Open Access Research Journal of Science and Technology, 2026, 17(02), 018–027.
Article DOI: 10.53022/oarjst.2026.17.2.0070
Publication history:
Received on 13 June 2026; revised on 21 July 2026; accepted on 23 July 2026
Abstract:
Coal-fired power stations remain critical to electricity security in Zimbabwe, yet their contribution is weakened when boiler failures increase forced outages, maintenance costs and generation instability. This paper improves and consolidates a case study of Local Thermal Power Station Stage 2 in Zimbabwe, by developing a model to optimise a preventive maintenance strategy for boiler reliability. The study analysed a 13-year monthly operating dataset from 2010 to 2023, stakeholder evidence on existing maintenance practices, reliability trends, plant availability, generation output, boiler efficiency, failures, maintenance cost and mean time between maintenance. Several machine-learning classifiers were evaluated, and the Random Forest classifier was selected for the preventive maintenance decision model due to its superior predictive performance and suitability for non-linear operational data. The results show that the existing maintenance strategy is highly reactive, with corrective maintenance accounting for about 72% of the strategy mix. The long-term reliability profile was depressed relative to world-class thermal power plant expectations and declined gradually across the review period. The developed model identified an optimal reliability point of about 66%, associated with improved availability, an average generation of 154.26 MW, electricity sent out of 43.7 GWh, a mean time between maintenance of 300 hours, reduced predicted maintenance costs, and fewer boiler failures. Validation against the first 16 weeks of 2024 indicated a strong positive relationship between predicted and actual outcomes, while also showing a performance gap between current practice and model-optimised values. The paper argues that a data-driven preventive maintenance model can support more disciplined boiler maintenance planning, reduce unplanned outages, and strengthen electricity supply reliability in coal-fired power stations.
Keywords:
Boiler Preventive Maintenance; Coal-Fired Power Station; Machine Learning; Random Forest; Reliability Optimisation
Full text article in PDF:
Copyright information:
Copyright © 2026 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution Liscense 4.0