A New Modelling and Feature Extraction Method Based on Complex Network and Its Application in Machine Fault Diagnosis

The application of the existing complex network in fault diagnosis is usually modelled based on the time domain, resulting in the loss of sign frequency-domain features, and the extracted topology features of network are too macroscopic and insensitive to local changes within the network. This paper proposes a new method of local feature extraction based on frequency complex network (FCN) decomposition and builds a new complex network structure feature on this basis, namely, subnetwork average degree. The variation law of signals in frequency domain is obtained with the aid of the structural features of complex network. The local features that are sensitive to local changes of the network are applied to characterize the whole network, with flexible application and without limitation in mechanism. The average degree of subnetwork could be regarded as feature parameters for rolling bearing fault diagnosis and degradation state recognition. Analysis on the experimental data and bearing life cycle data shows that the method proposed in this paper is effective, revealing that the extracted features have effective separability and high accuracy in fault recognition and the degradation detection of the life cycle of rolling bearings combined with neural networks. Moreover, the proposed method has reference value for the processing and recognition of other nonstationary signals.