a b s t r a c tIndustrial needs are evolving fast towards more flexible manufacture schemes. As a consequence, it is often required to adapt the plant production to the demand, which can be volatile depending on the application. This is why it is important to develop tools that can monitor the condition of the process working under varying operational conditions. Canonical Variate Analysis (CVA) is a multivariate data driven methodology which has been demonstrated to be superior to other methods, particularly under dynamically changing operational conditions. These comparative studies normally use computer simulated data in benchmark case studies such as the Tennessee Eastman Process Plant (Ricker, N.L. Tennessee Eastman Challenge Archive, Available at 〈http://depts.washington.edu/control/LARRY/TE/down load.html〉 Accessed 21.03.2014).The aim of this work is to provide a benchmark case to demonstrate the ability of different monitoring techniques to detect and diagnose artificially seeded faults in an industrial scale multiphase flow experimental rig. The changing operational conditions, the size and complexity of the test rig make this case study an ideal candidate for a benchmark case that provides a test bed for the evaluation of novel multivariate process monitoring techniques performance using real experimental data. In this paper, the capabilities of CVA to detect and diagnose faults in a real system working under changing operating conditions are assessed and compared with other methodologies. The results obtained demonstrate that CVA can be effectively applied for the detection and diagnosis of faults in real complex systems, and reinforce the idea that the performance of CVA is superior to other algorithms.

Kernel principal component analysis (KPCA) is an effective and efficient technique for monitoring nonlinear processes. However, associating it with upper control limits (UCLs) based on the Gaussian distribution can deteriorate its performance. In this paper, the kernel density estimation (KDE) technique was used to estimate UCLs for KPCA-based nonlinear process monitoring. The monitoring performance of the resulting KPCA-KDE approach was then compared with KPCA, whose UCLs were based on the Gaussian distribution. Tests on the Tennessee Eastman process show that KPCA-KDE is more robust and provide better overall performance than KPCA with Gaussian assumptionbased UCLs in both sensitivity and detection time. An efficient KPCA-KDE-based fault identification approach using complex step differentiation is also proposed.

No abstract

No abstract

Dynamic principal component analysis (DPCA) is commonly used for monitoring multivariate processes that evolve in time. However, it is has been argued in the literature that, in a linear dynamic system, DPCA does not extract cross correlation explicitly. It does not also give the minimum dimension of dynamic factors with non zero singular values. These limitations reduces its process monitoring effectiveness. A new approach based on the concept of dynamic latent variables is therefore proposed in this paper for extracting latent variables that exhibit dynamic correlations. In this approach, canonical variate analysis (CVA) is used to capture process dynamics instead of the DPCA. Tests on the Tennessee Eastman challenge process confirms the workability of the proposed approach.