| Popis: |
In recent years, machine learning and deep learning based fault diagnosis methods have been studied, however, most of them remain at the experimental stage mainly because of two obstacles, briefly, a) inadequate faulty examples and b) various working conditions of industrial data. In this literature, a practical algorithm named Data Simulation by Resampling (DSR) is proposed for data augmentation to alleviate the two problems in fault diagnosis. In essence, as a form of Vicinal Risk Minimization (VRM), DSR utilizes a two-stage resampling operation to simulate vicinal examples in both time domain and frequency domain. By doing so, DSR can both increase the sample diversity and the quantity of training set, which regularizes machine learning and deep learning based methods to achieve a higher generalization performance. Our experiments verify the effectiveness of DSR and show the possibility of combining it with other augmentation algorithms. |
