Fault Detection and Diagnosis Based on the Machine Learning Method of Lifting Scheme Wavelet and PCA

Abstract

In order to improve the wavelet threshold denosing effect and overcome the low efficiency and accuracy problem of conventional fault detection and diagnosis (FDD) methods, an novel approach based on threshold denosing function with double variable parameters and lifting scheme wavelet is proposed. Firstly, the proposed method is applied to denose the data of TE process. Then, the preprocessed data is classified by Principle Component Analysis (PCA) to detection and diagnose the faults. To certify the characteristic of the method, the proposed method is applied to detect and diagnose the faults in TE process, and compare with the soft and hard threshold methods which are used with lifting wavelet and PCA. Simulation results show that, the ensemble denosing method based on threshold denosing function with double variable parameters and lifting scheme wavelet is better than conventional denosing methods, meanwhile, the accuracy of fault detection and diagnosis with PCA is improved. These steps, which should require generation of the final output from the styled paper, are mentioned here in this paragraph. First, users have to run “Reference Numbering” from the “Reference Elements” menu; this is the first step to start the bibliography marking (it should be clicked while keeping the cursor at the beginning of the reference list). After the marking is complete, the reference element runs all the options under the “Cross Linking” menu.