Composite fault diagnosis of analog circuit system using chaotic game optimization-assisted deep ELM-AE

When multiple electronic components in an analog circuit fail simultaneously, since its high complexity, it is arduous to establish an accurate fault model and diagnose the composite faults precisely. Traditional approach for composite fault diagnosis is severely constrained since it excessively relies on domain expertise and signal processing technology. The rapid development of deep learning provides a new idea for it. In this study, an improved strategy based on deep extreme learning machine with autoencoder (DELM-AE) is proposed, which possesses more prominent capability on feature extraction and representation. The proposed method first utilizes multiple extreme learning machine with autoencoder (ELM-AE) to adaptively extract the hidden features from original collected signals layer by layer, which does not need to design representative features manually. Subsequently, efficient classification will be conducted based on extreme learning machine algorithm. All parameter values of the proposed network are determined at one time, eliminating the iterative updating process. To further enhance the performance of DELM-AE on feature extraction and classification, a novel chaos game optimization algorithm (CGO) is introduced to optimize the initial network parameters. The experimental results on two typical test circuits demonstrate that the proposed method, CGO-DELM-AE, can effectively diagnose the composite faults of analog circuit.