Abstract
Industrial data is usually nonlinear and corrupted by noise and outliers, which brings great challenges to fault detection and modeling in industrial data. To this end, L21-norm-based kernel principal component analysis is incorporated into the self-paced learning framework (L21-KPCA-SPL) in this study. It is innovative in sense that: (1) L21-KPCA is proposed, which can solve the nonlinear problem of data and increase the robustness of the algorithm; (2) self-paced learning (SPL) framework can avoid the local optimal solution problem caused by non-convex optimization; (3) based on the process monitoring of L21-KPCA-SPL, the pixel cumulative contribution of monitoring statistics is proposed. Compared with traditional PCA-like methods, the proposed algorithm is more robust. Compared with other robust methods, the proposed algorithm is more suitable for dealing with nonlinear data. Extensive experiments have been conducted on image classification datasets to demonstrate that the proposed method is more effective than other state-of-the-art methods. Furthermore, the proposed algorithm is used to detect ore blockage fault at the turn of conveyor belt. The experimental results further verify the effectiveness of the proposed method, which can replace the traditional manual detection and meet the requirements of real-time detection of ore blockage.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call