Multiblock regression model for fault diagnosis

For multiple output faults arose from external interference and/or noise, the traditional multiblock regression approaches may not effectively diagnose them as the input- unrelated information corresponding to each output variable is computed at super block level or even not computed. In this paper, a new multiblock regression-based fault diagnosis approach is proposed to diagnose multiple unpredictable output faults, that is, multiple input-unrelated faults. On the basis of multiblock global orthogonal projections to latent structures, the full block/subblock output-related and output-unrelated information are computed. At the same time, the input- unrelated and block/subblock output-unrelated information are further analyzed to diagnose detailed fault information. Once an input-unrelated fault is confirmed, variables contributions to detection statistics are calculated at block level to locate the fault variables. A numerical example is applied to declare the fault diagnosis ability of the proposed method.