Exploiting connectivity structures for decomposing process plants

Abstract Process plants are typically characterized by a large number of variables which renders traditionally deployed process systems algorithms computationally intractable for online applications. As parallelization and distribution of computational methods are becoming increasingly important and feasible, this paper proposes a method for structural analysis of plants in order to estimate connectivity strengths among various sub-processes making the process system algorithms amenable for distributed computing. In this work, analogy is drawn to the neuroscience literature where connectivity of neuronal population is established using data from magnetic resonance imaging. By using an input-state-output deterministic model for process systems and parameterizing this model to reflect connectivity and coupling, a Bayesian scheme is developed to estimate connectivity while incorporating priors. The algorithm is successfully applied to three case studies- one to demonstrate computational efficacy, one for which exact quantitative information of the structural connectivity is available and another where only qualitative information of the structural connectivity can be deduced.