Data-driven remote fault detection and diagnosis of HVAC terminal units using machine learning techniques

The modernising and retrofitting of older buildings has created a drive to install building management systems (BMS) aimed to assist building managers pave the way towards smarter energy use, improve maintenance and increase occupants comfort inside a building. BMS is a computerised control system that controls and monitors a building’s equipment, services such as lighting, ventilation, power systems, fire and security systems, etc. Buildings are becoming more and more complex environments and energy consumption has globally increased to 40% in the past decades. Still, there is no generalised solution or standardisation method available to maintain and handle a building’s energy consumption. Thus this research aims to discover an intelligent solution for the building’s electrical and mechanical units that consume the most power. Indeed, remote control and monitoring of Heating, Ventilation and Air-Conditioning (HVAC) units based on the received information through the thousands of sensors and actuators, is a crucial task in BMS. Thus, it is a foremost task to identify faulty units automatically to optimise running and energy usage. Therefore, a comprehensive analysis on HVAC data and the development of computational intelligent methods for automatic fault detection and diagnosis is been presented here for a period of July 2015 to October 2015 on a real commercial building in London. This study mainly investigated one of the HVAC sub-units namely Fan-coil unit’s terminal unit (TU). It comprises of the three stages: data collection, pre-processing, and machine learning. Further to the aspects of machine learning algorithms for TU behaviour identification by employing unsupervised, supervised, and semi-supervised learning algorithms and their combination was employed to make an automatic intelligent solution for building services. The accuracy of these employed algorithms have been measured in both training and testing phases, results compared with different suitable algorithms, and validated through statistical measures. This research provides an intelligent solution for the real time prediction through the development of an effective automatic fault detection and diagnosis system creating a smarter way to handle the BMS data for energy optimisation.