A novel multiblock method using latent variable partial least squares

It has been demonstrated that, when applying multivariate statistical techniques to industrial plants with large numbers of process variables distributed into several process units, it can often be beneficial to apply the algorithms to the sub-sets of the process variables. This approach has been termed multiblock and procedures exist for applying the technique to both the principal component analysis and partial least squares (PLS). In this paper a modified form of multi-block PLS is presented. An application of this algorithm to a simulated de-isobutaniser suggests that for this system, the proposed algorithm provides greater fault detection and isolation capabilities than traditional MBPLS approaches.