Output-relevant fault reconstruction based on Total PLS

Process monitoring is critical for efficient operations of industrial processes. When a fault occurs, relevant measured data are affected by the fault, which leads to poor quality of products consequently. This paper proposes a new output-relevant index for detecting faults that affect the output or quality, and studies the fault detectability based on total projection to latent structures (T-PLS). Given a fault subspace, fault-free data are reconstructed and the fault magnitude is estimated based on the new index. A simulation example is used to show the effectiveness of the proposed methods.