Improved Projection to Latent Structures for quality-relevant process monitoring

A process monitoring method called Improved Projection to Latent Structures (IPLS) is proposed in this paper. It aims at changing the oblique projection to orthogonal projection on the input space in basic PLS, thus a decomposition on input space to two orthogonal subspaces is achieved. Moreover, the output space is divided into three parts-a predictable subspace, an unpredictable principal subspace and the unpredictable residual space. The detailed decomposition on both the inputs and outputs enables producers to tell whether an output-relevant fault can be predicted by X. The simulated numerical examples and Tennessee Eastman Process Cases are used to confirm effectiveness of IPLS.