KC-19-A018 -- Automatic Fault Detection and Diagnosis of Air Handling Unit Using an Online Machine Learning Algorithm

This paper proposes an evolving learning approach which enables adaptation of the existing AFDD algorithm to incorporate the new observed state such as new operational mode or a new fault in the air handling unit. The proposed AFDD algorithm is constructed based on an evolving Gaussian mixture model. A dynamic building energy simulation model generated data for validation of the new AFDD algorithm. The results demonstrate the success of the proposed algorithm in AFDD of common faults and unseen dynamics in the air handling unit.