A BEARING DEEP LEARNING TRANSFER DIAGNOSIS METHOD BASED ON OPTIMIZATION OF SYMMETRIC POLAR COORDINATES

Abstract

Aiming at the problem of graphical feature representation of one-dimensional mechanical vibration signals, a bearing fault diagnosis method based on symmetric polar coordinates and residual network migration learning is proposed, which combines the powerful image classification and recognition ability of convolution neural network. Therefore, a bearing fault diagnosis method based on symmetric polar coordinates and residual network transfer learning is proposed in this paper. In order to highlight the fault characteristics of bearing vibration signals and take into account the calculation efficiency, the proposed method uses the symmetric polar coordinate method to convert the one-dimensional mechanical vibration signal into a mirror-symmetric snowflake map quickly and the transformation parameters and data sampling length are optimized synchronously by NSGA-II to obtain the image features with better distinguishability. Then, the transfer learning of residual network is used to train and classify. The bearing dataset of Case Western Reserve University which includes different rotational speeds and load is used to verify this method and a good recognition effect has been achieved.

Keywords

• Fault diagnosis;Symmetric polar coordinates;Feature extraction;Transfer learning;Residual network