Feature fusion based fault diagnosis of hoist inverter

Abstract

The difficulty in fault diagnosis of mine hoist inverters lies in extracting the features that characterize faults. At present, signal processing methods are mainly used to obtain fault statistical features, or the fault depth features are extracted by neural networks. In the actual working environment, the hoist inverter is affected by factors such as background noise and load changes. Therefore, it is difficult to obtain features that can characterize the faults effectively by using a single feature extraction method, resulting in low fault diagnosis accuracy of the hoist inverter. In order to solve the above problems, a fault diagnosis method of hoist inverter based on the fusion of statistical features and depth features is proposed. Firstly, the Hilbert-Huang transform(HHT) is used to conduct modified ensemble empirical mode decomposition(MEEMD) of the inverter output current signal so as to obtain the fault statistical features. At the same time, the squeeze and excitation with densely connected convolutional network(SE-DenseNet) is used to extract the depth features of the output current signal. Secondly, the local fisher discriminant analysis(LFDA) is used to perform fusion and dimensionality reduction processing on the combination of the two features to obtain low-dimensional fusion features of statistical features and depth features. Finally, the low-dimensional fusion features are input to the extreme learning machine to obtain inverter fault classification. Experiments are conducted for a single IGBT open-circuit fault in the hoist inverter. The results show that the low-dimensional fusion features obtained by this method are more capable of fault characterization than single features, which improves the fault recognition accuracy effectively.

Keywords

• mine hoist; inverter open circuit; fault diagnosis; feature fusion; feature dimensionality reduction; hilbert-huang transform; modified ensemble empirical mode decomposition; squeeze and excitation with densely connected convolutional network