Machine Learning Techniques for Building Predictive Maintenance: A Review

Background and aim – Proper maintenance is crucial for ensuring the sustainable use of building systems and equipment throughout their life cycles. Predictive maintenance strategies aim to minimise unplanned downtime and improve equipment lifespan, but their implementation is complex. Machine learning (ML), on the other hand, offers a novel solution for making systematic predictions across various disciplines. This review analyses the interrelationships between predictive maintenance and ML techniques to identify current research trends and potential areas for further study.
Methods / Methodology – A bibliographic analysis was conducted on a sample of 102 journal articles with VOSViewer. Key topics generated by co-occurrence analysis were then discussed semi-systematically, focusing on the most popular predictive maintenance applications and ML techniques.
Results – The results show a distinct relationship between the two terms, yet co-author analysis reveals a lack of global collaboration among authors. Additionally, Support Vector Machines, Artificial Neural Networks, Deep Neural Networks, Decision Trees, Random Forests, Bayesian Networks, and K-nearest neighbours are found to be the most frequently used ML techniques.
Originality –The study recognises the current research trends and provides future research implications. This study highlights the importance of adopting ML for predictive maintenance to achieve sustainability and NetZero carbon policy goals, which have not been explicitly addressed before.
Practical or social implications – The recommendations of this research broaden the scope of predictive maintenance studies. Emphasising collaborations between authors, institutions, and countries could significantly enhance research output in Facilities Management and Building Life Cycle.