ICA based on LTSA and its application in chemical process monitoring

The independent component analysis(ICA)has been successfully applied in the linear non-Gaussian processes monitoring. However,ICA can not deal with the process when the data are strongly nonlinear. On the other hand,the local tangent space alignment(LTSA)is able to extract the nonlinear structure of the process dataset. Hence,a LTSA-ICA monitoring method is proposed in combining the advantages of ICA and LTSA. LTSA is applied to extract the underlying manifold structure,and ICA is applied in the sub-manifold space. The proposed method showed a satisfactory performance when used to monitor the Tennessee Eastman process.