Fault identification with modified reconstruction-based contribution based on kernel principal component analysis

In industrial manufacturing processes, it is crucial to correctly identify the root cause of a fault. Reconstruction-based contribution with kernel principal component analysis (KPCA-RBC) was proposed to tackle this problem. However, this conventional RBC method focuses only on the fault magnitude, which is how much a faulty sample is moved along each axis by the reconstruction procedure with a fault detection index, and this might have limited its identification performance. In this paper, a new fault identification method, modified KPCA-RBC, is proposed. The proposed method takes into account how much a fault detection index is reduced by reconstruction along each axis at the same time as the fault magnitude. A numerical example showed that the proposed method outperformed the conventional RBC method in the identification performance. In addition, the practicability of the proposed method was confirmed through a case study of the vinyl acetate monomer (VAM) plant model.