Efficient and Robust Process Monitoring Using Structured Low-Rank Representation

In this brief, a new multivariate statistical method based on structured low-rank representation (SLR) is proposed to detect minor faults for industrial process monitoring (PM). The core idea of the proposed SLR is to enhance the representation of minor faults by using the $\ell _{2,0}$ -norm, and improve the robustness to noise by introducing a regularization term. Further, a learnable manifold constraint is incorporated to preserve the cause-effect relationship between monitoring variables. More importantly, a distributed optimization algorithm is developed with convergence analysis. Simulation examples are conducted to demonstrate the effectiveness and robustness of the proposed PM method.