Model-based fault detection and diagnosis for HVAC systems using convolutional neural network

Abstract

In building heating, ventilation, and air conditioning (HVAC) systems, fault detection and diagnosis (FDD) is crucial for achieving high energy efficiency. In this study, a novel method for FDD is proposed, which includes fault database generation by detailed simulation, convolutional neural network (CNN) training using a database, and FDD of real data using the trained CNN. The CNN is a classifier with sufficiently high accuracy for diagnosing the subtle fault features emerging in the fault behavior data of HVAC systems. It was confirmed that FDD of real data was possible by the trained CNN, in addition to learning the generated database with high accuracy. Thus, this methodology can assist in analyzing real data because it is possible to locate the fault and assume its relative severity approximately.