Optimal Sparse Principal Component Analysis With a Varying Regularization Coefficient for Industrial Fault Diagnosis

With the increasing complexity of industrial systems, modeling and intelligent diagnosis of high-dimensional industrial data have become increasingly challenging. To address this issue, this study proposes an optimal sparse principal component analysis (OSPCA) method with a varying regularization coefficient. The lower bound of the sparse regularization coefficient for achieving optimal feature selection is given, and the theoretical proof is provided. Subsequently, an iterative optimization algorithm with a varying regularization coefficient is proposed for model optimization. The proposed OSPCA is applied to unsupervised fault clustering and dynamic industrial process monitoring, and the experimental results show the excellent performance of OSPCA in the fault diagnosis tasks.