Machine learning approach for predictive maintenance in an advanced building management system

Predictive maintenance is a concept linked to Industry 4.0, the fourth industrial revolution that monitors equipment's performance and condition during regular operation to reduce failure rates. The present paper deals with a predictive maintenance strategy to reduce mechanical and electrical plant's malfunctioning for residential technical plant systems. The developed strategy can guarantee a tailored maintenance service based on machine learning systems, drastically reducing breakdowns after a maximum period of 3 years. The developed strategy evaluates an acceptable components failure rate based on statistical data and combines the average labour costs with the duration of each maintenance operation. The predictive strategies are elaborated on the minimum cost increase necessary to achieve the abovementioned objectives. A case study based on a 3-year-period has been conducted on a modern residential district in Rome composed of 16 buildings and 911 apartments. In particular, the analysis has been performed considering mechanical, electrical and lighting systems supplying the external and common areas, excluding the apartments, to avoid data perturbation due to differential user's behaviours. The overall benefits of predictive maintenance management through Big Data analysis have proven to be the substantial improvements in the overall operation of different plants as mechanical and electrical plants of residential systems.