Predictive Maintenance for Smart Buildings: Balancing QoS and Cost Efficiency

The rapid advancement of sensing technologies and connectivity has revolutionized predictive maintenance (PdM), particularly for smart buildings (SB). Despite these advancements, implementing data-driven approaches faces challenges, mainly due to scarce failure data and the SB systems complexity. In this paper, we propose a cooperative multi-agent reinforcement learning (RL) based approach to address these challenges in which each agent is responsible for maintaining one subsystem. As a case study, we consider three subsystems within the SB: lighting, network, and IT. The network subsystem ensures power supply, while the IT subsystem controls and shares data. The proposed approach enables the design of comprehensive maintenance decision processes that minimize maintenance costs for these subsystems while satisfying their quality of service (QoS) standards. Our approach’s advantage lies in its ability to ensure the effective operation of each SB subsystem and the SB itself, even with limited data. Moreover, the proposed approach is adaptative according to the operating conditions of each SB subsystem.