Finding the root cause of a control issues, does an artificial intelligence optimise a building more effectively than automated fault detection?

Buildings and their occupants are responsible for a large portion of global energy consumption, roughly half of which is associated with the HVAC systems. HVAC systems in large buildings are connected and controlled by a Building Management System (BMS) which are prone to errors, faults and mismanagement and HVAC systems can suffer wastage of up to 30% of total energy consumption through poor controls (Zhang 2021, Granderson 2017). This discrepancy between design and reality is often referred to as the ‘performance gap’. Consequently, there is scope to save a significant amount of energy by addressing these issues. And with the global need to urgently reduce energy demands and carbon emissions it has been a popular topic of research.

It follows that a method is needed to identify faults, miscalibration and energy wastage in buildings. Such methods are commonly referred to as fault detection and diagnosis (FDD) methods. Building Management Systems in commercial buildings often connect thousands of sensors and actuators together. The constant communication of these devices produces large amounts of data which has historically been discarded or used in a limited capacity for troubleshooting or commissioning. The infrastructure for this data already exists in most buildings, presenting opportunity for deeper analysis and optimisation. Using this data the FDD methods can be automated, taking engineering knowledge and supplying a data set with rule-based algorithms to find any root causes of control inefficiencies within the building. A secondary approach is to use artificial intelligence models over these data sets to find ways to operate the building more efficiently.

This paper shows the effectiveness of an automated FDD approach via 6 case studies in commercial office buildings, highlighting the methods by which faults were identified and resolved, as well as the impacts of the corrective action. It then compares these examples with the findings of research into the effectiveness of the AI-based approach and discusses whether the AI-based approach is more effective than the automated FDD approach or whether the AI-based approach is using significantly more computer power and time to, in fact, find rules to correct control patterns that are already widely known within general engineering practise. The results of this research should indicate whether AI or automated FDD are currently more suitable to close the performance gap in commercial buildings.