Modeling Techniques For Integration Of Process Systems

Increasing social, economic and environmental pressure force the process industry to look for new ways to improve the overall operation of their process systems.Since these systems are operated at different levels of decision (apparatus, plant, enterprise, etc.) integration of the different levels is expected to lead to significant improvements in system efficiency (energy, waste, costs, quality, product distribution, logistics, etc). Integration also results in increased mathematical complexity that can not be handled with the current numerical methods. This leads inevitably to a paradox: the integrated problem needs to be decomposed again into simpler sub-problems that are solved independently providing sub-optimal solutions. This paper will discuss why integration fails and which steps are needed to break the paradox. Focus will be on new modeling techniques for the different decision levels. INTRODUCTION The process industry (petrochemical-, pharmaceuticaland consumer goods industries) is an economic key sector in the world. During the last ten years, the process industry sensed an increasing pressure to reduce costs and inventories as a result of environmental, social and economic changes. The overall optimization of R&D, manufacturing and supply chain is seen as the ‘holy grail’ and as a complete solution to deal with these emerging changes (Grossmann (2005), Varma et al. (2007)). In the past, such process systems were relatively simple and straightforward models could be used for the optimization of the process efficiency. However, in the last decades processes became increasingly complex and are continuously subject to changes. The integration of various levels in company functions (purchasing, manufacturing, distribution and sales), distributions (markets, facilities, vendors) and decisions (strategic tactical and operational) is crucial to realize overall optimization resulting in dramatic increases in operating efficiency. At this moment modeling and optimization at each of the levels is mostly done sequentially and this leads to the obvious question whether the optimal solutions of the individual levels are the same as the optimal solutions of the overall problem (Harjunkoski et al. (2008)). For the overall optimization, integration of the different levels into a monolithic structure is applied. Integration could mean that targets are closer to their global optima, but it also encompasses the increasing complexity of the overall optimization problem. Most of these monolithic structures are NP complete problems, i.e. there is no known way to locate the solution. This conclusion leads to the integration paradox: for most industrially scaled optimization cases, the integrated problem is too complex to be solved and needs to be decomposed again into sub-problems that can be handled with the available methodologies. The integration paradox can only be broken by resolving four fundamental flaws in the integration process: 1. Different levels speak different languages (the models used for each of the layers differ structurally) (Wang et al. (2007)) 2. There is no proper interface to communicate information between the levels (Stehphanoupoulos & Han (1996)) 3. Methods for active re-optimization and experimental validation are not incorporated in the integrated optimization problem, making the solution insensitive to external changes (Shobrys & White (2002)) and 4. Numerical algorithms to deal with complex problems are far from perfect (large computational times and/or sub-optimal solutions). (Klatt & Marquardt (2008)) In this proceeding we will address the first flaw in the integration process: modeling issues. We will identify a modeling technique that is suitable for process systems integration and on the basis of two examples we will project what the capabilities of the envisaged modelin technique should be. PROCESS MODELING Fig. 1 shows schematically how a process system can be described by models at several levels. At this moment, each of the levels uses different models with distinction in: mechanistic/black-box, deterministic /stochastic, distributed/lumped, discrete/continuous, linear/nonlinear, event/data-driven, etc. The development of a modeling technique that captures the essence of each of the decision levels adequately is of crucial importance to solve the integration paradox. Proceedings 23rd European Conference on Modelling and Simulation ©ECMS Javier Otamendi, Andrzej Bargiela, José Luis Montes, Luis Miguel Doncel Pedrera (Editors) ISBN: 978-0-9553018-8-9 / ISBN: 978-0-9553018-9-6 (CD) Elementary system Sub system System Input Output Figure 1: Decomposition of a process system into several sub problems and levels, after Marquardt (Klatt