A novel model-independent data augmentation method for fault diagnosis in smart manufacturing

With the rapid development of information technology, data-driven fault diagnosis has gained more and more attention because it provides a new way for enterprises to save costs. Considering that there are few abnormalities in equipment operation in actual industrial applications, it is still a challenge to implement data-driven fault diagnosis that requires a large amount of fault data. To tackle the challenge, this paper proposes a model-independent data augmentation method, which is a weighted combination of the two time series data augmentation methods, i.e. Gaussian noise and signal stretching. The experimental dataset is collected from an intelligent motor test platform. The fault diagnosis model based on support vector machine and feedforward neural network are applied to study the ability of the proposed data augmentation method in terms of model independence. Experimental results show that the proposed data augmentation methods can significantly improve the accuracy of fault diagnosis.