Fault detection and diagnosis for heat source system using convolutional neural network with imaged faulty behavior data

Abstract

Faults that impair performance can occur in a heat source system because it comprises various devices and has complex controls. This article presents a novel method for fault detection and diagnosis (FDD). This study focused on a real system with a water thermal storage tank. First, system behaviors in response to faults were determined using a detailed system simulation. Then, a fault database was generated using the simulation results with fault labels. We preprocessed the database and converted the data into images. Then, convolutional neural networks (CNNs) were trained using the database, and the trained CNNs were used for diagnosing real data. The accuracy of the CNNs was 98.7% in training, and real data were diagnosed with probabilities. We analyzed the real data, where the probability indicated the likely presence of a fault and reviewed how the real data were similar to the fault assumed in the simulation. We concluded that the proposed FDD method will help in analyzing real data, as it indicates faults emerging in the real data with probability, whereas conventional data analysis requires checking the data using expert knowledge.