LAUSR

Abstract The pervasive monitoring of HVAC systems through Building Energy Management Systems (BEMSs) is enabling the full exploitation of data-driven based methodologies for performing advanced energy management strategies. In this context, the implementation of Automated Fault Detection and Diagnosis (AFDD) based on collected operational data of Air Handling Units (AHUs) proved to be particularly effective to prevent anomalous running modes which can lead to significant energy waste over time and discomfort conditions in the built environment. The present work proposes a novel methodology for performing AFDD, based on both unsupervised and supervised data-driven methods tailored according to the operation of an AHU during transient and non-transient periods. The whole process is developed and tested on a sample of real data gathered from monitoring campaigns on two identical AHUs in the framework of the Research Project ASHRAE RP-1312. During the start-up period of operation, the methodology exploits Temporal Association Rules Mining (TARM) algorithm for an early detection of faults, while during non-transient period a number of classification models are developed for the identification of the deviation from the normal operation. The proposed methodology, conceived for quasi real-time implementation, proved to be capable of robustly and promptly identifying the presence of typical faults in AHUs.