Intelligent fault diagnosis of rolling bearing using variational mode extraction and improved one-dimensional convolutional neural network

When the rolling bearing fails, the fault features contained in bearing vibration signal are easily submerged by fortissimo noise interference signals, and have obvious non-stationary and nonlinear properties. This means that it is extremely challenging to acquire useful bearing fault features and identify bearing fault patterns effectively by traditional diagnosis methods. To more efficiently learn bearing fault information and improve bearing fault diagnosis accuracy, this research proposes a new intelligent fault diagnosis method for rolling bearing based on variational mode extraction (VME) and an improved one-dimensional convolutional neural network (I-1DCNN). Firstly, a new adaptive signal processing method named VME is employed to handle the collected bearing vibration signals with the aim of obtaining the desired mode component and removing the noise interference information. Meanwhile, the extracted mode components are randomly divided into the training set, validation set and test set. Then, the training set and validation set are input into the proposed I-1DCNN model for training, where the proposed I-1DCNN model may not only learn the discriminant features intelligently, but also boost the computational efficiency and alleviate the problem of over-fitting by incorporating the early stopping method and self-attention mechanism into the traditional one-dimensional convolutional neural network (1DCNN). Finally, the test set is input into the well-trained I-1DCNN to realize the automatic identification of different fault types of rolling bearing. The effectiveness of the suggested method is illustrated by analyzing two experimental data sets. In addition, by comparing with other representative methods, the superiority of the proposed method is testified in bearing health condition identification.