Author(s)

Yong Chen, Rong Bian and Wenzheng Ding

Corresponding Author

Yong Chen

Abstract

Data-driven fault diagnosis methods such as deep learning is a remarkable new way to find the problem. In this paper, considering the small volume of labeled samples in fault diagnosis of CNC machine tool spindle, we propose a fault diagnosis method based on deep transfer learning to achieve our goals more quickly and efficiently. First, the original vibration signals acquired by multiple 3-axis acceleration sensor mounted on spindle were collected and a transformation method converting signals to image is developed to get inputs. Second, several pre-trained networks are compared and the best one is used to extract lower level features. Third, the output layer with target classes is replaced and the higher levels of the neural network are fine-tuned. Finally, the learning method is employed to learn the features of the network and has been tested on the datasets. The experimental results in this paper have proved that our method was correct identify machine tool spindle conditions and showed considerable potential for fault detection in manufacturing.

Keywords

CNC machine tool spindle; Fault diagnosis method; Deep transfer learning