A Novel Time-Driven Multiple Modes Division Based NCA Method for the AHU System Fault Detection

Fault detection is crucial for air handling unit (AHU) system's safe operation. The AHU operation represents a unique process with time-wise multiple mode characteristics and highly nonlinear characteristics. To address these challenges, a novel time-driven multiple modes division based neural component analysis (MNCA) method for the AHU system fault detection is proposed in our work. To figure out the time-wise multiple mode characteristics of the AHU system, the time-driven based multiple modes division method is adopted to precisely divide the entire process dataset into multiple operating modes along the time axis. To address the highly nonlinear and boost the efficiency of the fault detection, neural component analysis (NCA) fault detection models are established for each operating mode to nonlinearly extract the features with orthogonal properties. The effectiveness of the presented method is confirmed by results from the ASHRAE Research Project RP-1312 experimental datasets.