Fault features recognition of circular saw-cutting processes for aluminum alloy profiles based on the ReliefF-PCA algorithm

For the fault diagnosis process, the Principal Component Analysis (PCA) algorithm is widely used to optimize and reduce the dimension of the feature set. It can not only reduce the dimension of the feature vector to reduce the amount of computation, but also eliminate the features with interference effect to improve the recognition accuracy. However, the variance contribution rates used to determine the trade-off of principal components in the PCA algorithm do not fully reflect the ability of the features to distinguish categories. Principal components with a small variance contribution rate may often contain important information to identify sample differences. An improved ReliefF-PCA algorithm was proposed to solve this problem. In the improved ReliefF-PCA algorithm, the judgment criterion of determining the trade-off of principal components according to the size of the variance contribution rate in the traditional PCA algorithm is abandoned. A ReliefF algorithm is introduced to give different weight coefficient to the principal components according to the correlation between the principal components and the categories. Meanwhile, the Support Vector Machine (SVM) model is used to verify the effectiveness of this method. The results show that compared with the traditional PCA algorithm, the SVM model using the feature set optimized by the improved ReliefF-PCA algorithm can achieve 100% recognition accuracy with fewer features. Meanwhile, the running time of the SVM model is reduced by 43.58%. The results show the effectiveness and advantages of ReliefF-PCA algorithm in the fault diagnosis of circular saw-cutting processes for aluminum alloy profiles.