In this paper, we propose a predictive maintenance scheme for centralized HVAC systems by autonomous monitoring and analyzing their acoustic emissions. Our proposed solution allows a building to be retrofitted to monitor its HVAC without having to modify the existing infrastructure. Our approach is to employ an energy-efficient, low-cost, and distributed acoustic sensing platform to capture and process audio signals from HVAC systems. As part of this project, we develop audio models of a running HVAC system using a combination of unsupervised and supervised machine learning techniques with a human-in-the-loop for fault identification and prediction.