150 / 2021-03-31 15:13:04

Vibration and sound data fusion for bearing using convolution neural network

Vibration; Sound; Data fusion; CNN

Machine condition monitoring and fault diagnosis

The health states of machine can be expressed by a variety of physical quantities. Vibration and sound are two homology signals which indicates the health states of machine and there are a number of similarities between them. On the other hand, they contain some different fault features, so that signals fusion helps achieve higher accuracy. Therefore, this paper proposes a CNN model, which add a two-dimensional convolution layer before the traditional 1-D CNN to fuse vibration and sound data. Vibration and sound data are stacked row by row to form a 2D input matrix. As the convolutional filters move across the input matrix, data is fused together. After training with the whole model, the fusion layer can better fuse two groups of data. The advantage of the this approach is that it avoids the problems of manual feature extraction and information loss. The effectiveness of this method is validated through the data collected from a rolling bearing test rig at 10, 20, 30, 40 Hz. A total efficiency of recognition is over 97%. Compared with the same model using vibration data, sound data and the data simply add them together, the proposed model performs better. To evaluate the amount of information contained in the fuse data, information entropy was used. In order to verify the feature extraction ability of the proposed model, t-SNE was used. The proposed model gets higher accuracy than classical machine learning algorithms using the above data or artificially extracted features, which demonstrates the effectiveness of method.