Digital Twin-Driven Physically Constrained Generative Adversarial Network for Industrial Boiler Fault Diagnosis

Abstract

As important energy equipment, industrial boilers require more reliable fault diagnosis technology to ensure their continuous safe operation under high-temperature and high-pressure conditions. Due to the difficulty in collecting fault data, traditional fault diagnosis methods based on a large amount of labeled fault data cannot be effectively applied to actual industrial scenarios. This article proposes a physically constrained generative adversarial network (PCGAN) driven by digital twins (DTs) for industrial boiler fault diagnosis, which can accurately diagnose boiler faults when training data is insufficient. First, this article uses DT technology to establish a virtual model of the boiler and uses the constructed DT model to emulate the failure conditions of the boiler and generate simulated data. Then, a novel PCGAN fault diagnosis method is designed, and data obtained by the DT model is used to train and validate its effectiveness. Finally, the feasibility of the proposed method in industrial boiler fault diagnosis is verified by experimental results. The results show that the fault diagnosis accuracy of this method reaches 96.28%, which is significantly higher than other data-driven fault-diagnosis methods.