Improved Wavelet Threshold Denoising and Multi-scale Principal Component Analysis Method with Application to TE Process Monitoring

There is a lot of noise in the actual industrial process data, which will lead to the decline of data quality, and then affect the effect of fault diagnosis. In practical applications, the wavelet threshold method is generally used to remove noise. To overcome the shortcomings of traditional soft and hard threshold methods, an adaptive threshold denoising method based on normal distribution is proposed. In this method, normal distribution test is introduced to adaptively select soft threshold and hard threshold to reduce the noise of wavelet coefficients in each layer. Then the improved wavelet threshold denoising method and multi-scale principal component analysis method are combined to propose a fault diagnosis strategy based on improved multi-scale principal component analysis (IMSPCA). Firstly, the normalized distribution test is used to improve the wavelet threshold denoising. Then the improved wavelet threshold denoising method and multi-scale principal component analysis are combined to establish a comprehensive PCA model for fault diagnosis and improve the reliability of fault diagnosis. Finally, the feasibility and effectiveness of the IMSPCA method are verified by the simulation of TE process.