This is the authors' abstract. We don't add key points for this paper.
To address the problem that the traditional convolutional neural network (CNN) cannot fully extract the fault feature information in a single channel and the accuracy of hydraulic system fault diagnosis based on a single signal is not high, a fault diagnosis method combining dual‐channel CNN and improved D‐S evidence theory is proposed. First, the pressure signal, temperature signal, and vibration signal are denoised using the wavelet threshold denoising method, and the denoised pressure signal, temperature signal, and vibration signal are put into two channels for feature extraction, respectively. One channel uses 1D‐CNN to extract the features, and the other channel uses an improved HHT to process the input signal to obtain the time‐frequency map and then uses 2D‐CNN to extract the features, and finally, the features of the two channels are concatenated for fault diagnosis. The fault diagnosis results based on the denoised pressure signal, temperature signal, and vibration signal, respectively, are used as three pieces of evidence, the similarity coefficients of each evidence are obtained by calculating the value similarity and direction similarity between the evidences, the weight of each evidence is calculated based on the similarity, the new evidence is gained by the weighted average method to replace the most conflicting evidence, and information fusion based on the improved D‐S evidence theory is performed to determine the cause of the fault. Finally, the experiments are conducted based on the hydraulic system fault dataset, and the experimental results show that the accuracy of the hydraulic system fault diagnosis using this method reaches 99.75%, which is 7.0%, 7.5%, and 6.5% higher than the results of the fault diagnosis based on a single pressure signal, temperature signal, and vibration signal, respectively. The proposed method improves the accuracy of fault diagnosis and has some application value.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Qi et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: