Automated Fault Diagnostics for AHU-VAV Systems: A Bayesian Network Approach

Although it is widely accepted that 20%-30% of total HVAC energy in commercial buildings is wasted due to faulty or inefficient operation, there presently exist no widely adopted solutions to identify and remedy this waste. This lack of adoption is due primarily to the high upfront costs associated with the manual process of customizing commissioning solutions for each individual building. In order to reduce these costs, diagnostic technologies must automate the process of installing and customizing solutions for each implementation. A novel approach to addressing this problem for air handling units (AHUs) and variable air volume (VAV) boxes is presented here. This strategy utilizes a Bayesian network to identify and understand the system operation, identify faults that are wasting energy or impacting occupant comfort, and then generate performance baselines against which future operation will be compared. When this algorithm is first connected to a new building, an adaptive diagnostic Bayesian network identifies the components and configuration of the AHUs and VAVs, and then generates probabilistic outputs indicating the root causes of potential faults and inefficiencies. Once these issues have been rectified to the satisfaction of the building operator, the algorithm then begins to accumulate training data with detailed information about the system operation (while simultaneously continuing to monitor for additional faults). As this training data is accumulated, the diagnostic confidence of the Bayesian network is continuously improved. Additionally, this use of an operational baseline allows for accurate detection and diagnosis of faults causing gradual performance degradation in addition to faults that abruptly occur.