Independent component analysis model utilizing de-mixing information for improved non-Gaussian process monitoring

摘要: We focus on the de-mixing matrix, which is rarely studied in ICA model, to extract data information for fault detection.Multi-block strategy employed deal with big a novel way.The numerous divided through similarity index Generalized Dice's coefficient.Bayesian inference also combine results noise weakened.The way of diagnosis modified selected variables checked. The matrix generated from independent component analysis (ICA) can reveal about relations between and components, but traditional model does not preserve whole purpose feature extraction dimensionality reduction, so that some important may be abandoned. Multi-block has been improved an efficient method data. However, manner dividing original still subject discussion priori knowledge necessary process division. This paper proposes totally data-driven divides based coefficient combines sub-blocks using Bayesian inference. All fully utilized ability monitoring non-Gaussian improved. Meanwhile, corresponding contribution plot developed find root causes. performance proposed illustrated numerical example Tennessee Eastman benchmark case study.