An industrial signal generation method based on improved Fast Kurtogram filtering and Diffusion-TS

As a mechanical fault detection method, fault diagnosis can reduce the waste of resources and eliminate safety hazards. However, fault signals often contain interference components and are accompanied by the problem of sample imbalance, which seriously affects diagnosis accuracy. Therefore, this paper proposes an industrial signal generation method based on improved Fast Kurtogram filtering and Diffusion-TS(FLSESGASG-DTS). First, the improved Fast Kurtogram filtering (FLSESGASG) is used to filter the fault signal to remove the interference component. Then, Diffusion-TS (DTS) is used to generate fault samples to solve the problem of sample imbalance. In addition, this paper also proposes a multiple greedy Gaussian segmentation standardization method (MGGSN), which aims to standardize the fault signal without destroying the fault characteristics. The dataset of XJTU-SY is used for verification in this paper, and the results show that the FLSESGASG proposed in this paper effectively removes the interference component of the fault, and the Correlational Score of the generated samples reaches 0.0013, which is extremely close to 0. These experimental results validate the effectiveness of the proposed FLSESGASG-DTS method in signal filtering and generation and further provide theoretical support for industrial fault diagnosis.