Nonlinear dynamic process monitoring: the case study of a multiphase flow facility

Data-driven statistical process condition monitoring techniques have enjoyed remarkable growth over decades. However, although the source of data sets used for validating these monitoring algorithms may vary from computer simulations, experimental rigs and industrial processes, the fault being monitored are straightforward. Hence a benchmark data set acquired from an industrial-scale plant with complicated faults is of great interest. The multiphase flow facility presented in this work, which is a fully automated rig with adjustable configurations, can be a promising candidate. This paper applies a novel Multimode Kernel Latent Variable Canonical Variate Analysis (mKLV-CVA) algorithm to the benchmark data set collected from the multiphase flow facility in varying normal operating conditions as well as the slugging situation.