Nonlinear dynamic process monitoring based on DLLE-SVDD

A novel process monitoring method for dynamic nonlinear industrial processes is proposed by combining dynamic Locally Linear Embedding(DLLE) with Support Vector Data Description(SVDD). Firstly, the data matrix is augmented taking correlation of the samples into consideration. Then, LLE manifold learning algorithm is performed for nonlinear dimensionality reduction and feature extraction. The mapping matrix from data space to feature space was calculated by using local linear regression which guarantees the real-time property. Next, in order to avoid the influence of noise and disturbance on the traditional statistics, the fault detection model is obtained based on SVDD in the feature space, in which a corresponding monitoring index and its control limit are determined. Finally, the feasibility and efficiency of the proposed method are shown through the TE process.