The DMF: Fault Diagnosis of Diaphragm Pumps Based on Deep Learning and Multi-Source Information Fusion

Abstract

Effective fault diagnosis for diaphragm pumps is crucial. This paper proposes a diaphragm pump fault diagnosis method based on deep learning and multi-source information fusion (DMF). The time-domain features, frequency-domain features, and modulation features are extracted from the vibration signals from eight different positions. After feature enhancement and data preprocessing, the features are input into auto encoders (AE), convolutional neural networks (CNN), and support vector machines (SVM) to obtain the diagnostic results. The results indicate that the DMF method achieves a fault diagnosis accuracy of 99.98%, which is on average 9.09% higher than using a single diagnostic model. The demodulation method is more suitable for vibration signal feature extraction of the diaphragm pump, while the CNN is more suitable for identification of diaphragm pump faults. Specifically, it outperformed the sampling point 1-DPCA-AE model by 13.98% and the sampling point 4-DPCA-SVM model by 8.98%.