Independent component analysis( ICA) has been widely applied to the monitoring of non-Gaussian processes. Despite lots of applications,there is no universally accepted criterion to select the dominant independent comp...Independent component analysis( ICA) has been widely applied to the monitoring of non-Gaussian processes. Despite lots of applications,there is no universally accepted criterion to select the dominant independent components( ICs). Moreover, how to determine the number of dominant ICs is still an open question. To further address this issue,a novel process monitoring based on IC contribution( ICC) is proposed from the perspective of information storage. Based on the ICC with each variable,the dominant ICs can be obtained and the number of dominant ICs is determined objectively. To further preserve the process information, the remaining ICs are not useless. As a result,all the ICs are regarded to be divided into dominant and residual subspaces. The monitoring models are established respectively in each subspace, and then Bayesian inference is applied to integrating monitoring results of the two subspaces. Finally, the feasibility and effectiveness of the proposed method are illustrated through a numerical example and the Tennessee Eastman process.展开更多
基金National Natural Science Foundations of China(Nos.61374140,61403072,61673173)Fundamental Research Funds for the Central Universities,China(Nos.222201717006,222201714031)
文摘Independent component analysis( ICA) has been widely applied to the monitoring of non-Gaussian processes. Despite lots of applications,there is no universally accepted criterion to select the dominant independent components( ICs). Moreover, how to determine the number of dominant ICs is still an open question. To further address this issue,a novel process monitoring based on IC contribution( ICC) is proposed from the perspective of information storage. Based on the ICC with each variable,the dominant ICs can be obtained and the number of dominant ICs is determined objectively. To further preserve the process information, the remaining ICs are not useless. As a result,all the ICs are regarded to be divided into dominant and residual subspaces. The monitoring models are established respectively in each subspace, and then Bayesian inference is applied to integrating monitoring results of the two subspaces. Finally, the feasibility and effectiveness of the proposed method are illustrated through a numerical example and the Tennessee Eastman process.
