Multimode Process Monitoring Based on Density Space Clustering Locally Linear Embedding Technique

To meet the demand for modern multimode processes, effective process monitoring techniques are needed to increase the quality of products and reduce safety hazards. In this article, a novel scheme based on density space clustering locally linear embedding (DBSLLE) strategy is proposed for multimode process monitoring. DBSLLE improves the monitoring performance of the locally linear embedding (LLE) method in multimode processes and overcomes the difficulty of new samples unable to be projected by the LLE method. The basic idea of the proposed method is first adopting the density based space clustering of applications with noise (DBSCAN) rule to eliminate the multimode structure; next, the conventional LLE is used to extract nonlinear features from each model data and achieve dimensionality reduction. Finally, the weighted average distance sum of square is constructed as a statistic to monitor the process. The new sample is determined to belong to which model based on the sample density of the space in which it resides; A cost function is constructed between the new sample and the samples in its model, and the weight matrix is obtained by minimizing it; The weight matrix is multiplied with the low-dimensional projection of the training data to obtain the projection of the new sample; The statistical value of the projection sample is calculated to obtain the current state of the process. The efficiency of the proposed monitoring scheme is implemented in a multimode numerical simulation case, Tennessee-Eastman processes, and in the Continuous Stirred-Tank Reactor (CSTR). The experimental results indicate that the proposed method outperforms the traditional LLE monitoring schemes.