State-space independent component analysis for nonlinear dynamic process monitoring

The cost effective benefits of process monitoring will never be over emphasised. Amongst monitoring techniques, the Independent Component Analysis (ICA) is an efficient tool to reveal hidden factors from process measurements, which follow non-Gaussian distributions. Conventionally, most ICA algorithms adopt the Principal Component Analysis (PCA) as a pre-processing tool for dimension reduction and de-correlation before extracting the independent components (ICs). However, due to the static nature of the PCA, such algorithms are not suitable for dynamic process monitoring. The dynamic extension of the ICA (DICA), similar to the dynamic PCA, is able to deal with dynamic processes, however unsatisfactorily. On the other hand, the Canonical Variate Analysis(CVA) is an ideal tool for dynamic process monitoring, however is not sufficient for nonlinear systems where most measurements follow non-Gaussian distributions. To improve the performance of nonlinear dynamic process monitoring, a state space based ICA (SSICA) approach is proposed in this work. Unlike the conventional ICA, the proposed algorithm employs the CVA as a dimension reduction tool to construct a state space, from where statistically independent components are extracted for process monitoring. The proposed SSICA is applied to the Tennessee Eastman Process Plant as a case study. It shows that the new SSICA provides better monitoring performance and detect some faults earlier than other approaches, such as the DICA and the CVA.     

Nonlinear Dynamic Process Monitoring Using Canonical Variate Analysis and Kernel Density Estimations

The Principal Component Analysis (PCA) and the Partial Least Squares (PLS) are two commonly used techniques for process monitoring. Both PCA and PLS assume that the data to be analysed are not self-correlated i.e. time-independent. However, most industrial processes are dynamic so that the assumption of time-independence made by the PCA and the PLS is invalid in nature. Dynamic extensions to PCA and PLS, so called DPCA and DPLS, have been developed to address this problem, however, unsatisfactorily. Nevertheless, the Canonical Variate Analysis (CVA) is a state-space-based monitoring tool, hence is more suitable for dynamic monitoring than DPCA and DPLS. The CVA is a linear tool and traditionally for simplicity, the upper control limit (UCL) of monitoring metrics associated with the CVA is derived based on a Gaussian assumption. However, most industrial processes are nonlinear and the Gaussian assumption is invalid for such processes so that CVA with a UCL based on this assumption may not be able to correctly identify underlying faults. In this work, a new monitoring technique using the CVA with UCLs derived from the estimated probability density function through kernel density estimations (KDEs) is proposed and applied to the simulated nonlinear Tennessee Eastman Process Plant. The proposed CVA with KDE approach is able to significantly improve the monitoring performance and detect faults earlier when compared to other methods also examined in this study.      

A branch and bound method for isolation of faulty variables through missing variable analysis

Fault detection and diagnosis is a critical approach to ensure safe and efficient operation of manufacturing and chemical processing plants. Although multivariate statistical process monitoring has received considerable attention, investigation into the diagnosis of the source or cause of the detected process fault has been relatively limited. This is partially due to the difficulty in isolating multiple variables, which jointly contribute to the occurrence of fault, through conventional contribution analysis. In this work, a method based on probabilistic principal component analysis is proposed for fault isolation. Furthermore, a branch and bound method is developed to handle the combinatorial nature of problem involving finding the contributing variables, which are most likely to be responsible for the occurrence of fault. The efficiency of the method proposed is shown through benchmark examples, such as Tennessee Eastman process, and randomly generated cases.