SPLIT : a separation process designer

Abstract The development of chemical processes is a complex design problem, commonly performed as an evolution among synthesis, analysis and evaluation steps. Experience and skilled use of existing tools for process analysis and optimization allow designers to focus on the development of relatively few promising flowsheets. However, there is still a need for software systems supporting this activity. The blackboard-based integrated software environment SPLIT (“Separation Process Layout by Invention and Testing”) is conceived to automate the design of separation systems, while allowing the user the option to guide the solution process. The system can handle problem specifications at various levels of detail, using any available knowledge to restrict the number of alternative designs investigated. Resembling the control architecture AKORN-D proposed by Lien (Ph.D. Thesis, Univ. of Trondheim, Norway, 1988), SPLIT combines data-driven and goal-oriented strategies to provide a framework for opportunistic reasoning. Representations of the problem and the current state of the solution are embedded in the blackboard as several semantic hierarchies, some of which are dynamically created. Knowledge sources (KS) of two types screen the common memory blackboard for data patterns indicating their ability to be executed. Domain KS implement procedural knowledge about physical properties, the technologies available to perform required tasks, etc. Control KS are used to guide the problem solving behavior of the system; they are integrated in the same recognition/bidding cycle, but when selected by the scheduling function, only change foci of attention on tasks, partial solutions, or in general any object on the blackboard. Using a distributed problem solving framework (Cardozo, Ph.D. Thesis, Carnegie Mellon University, Pittsburgh, U.S.A., 1987), the blackboard system is being integrated with existing analysis/optimization tools. This framework is used within several domain knowledge sources to invoke such external programs as the flowsheeting program Aspen Plus (Aspen Tech, Aspen User Guide to Release 8.2 and 8.3 , Cambridge, MA, 1988) or the optimization package DICOPT + + (Viswanathan and Grossmann, Computers chem. Engng , 14 , 769–782, 1990) (a mixed-integer nonlinear programming package) for the analysis and optimization of partial and complete flowsheets. The focus of this paper is the way SPLIT addresses the presynthesis problem, i.e. the generation of a reasonable set of alternative separation processes, which may be incorporated in a superstructure in order to determine the best solution. The system first decomposes separation problems initially described through specifications of feed and product streams into explicitly represented binary split tasks. A classification of the components according to a criterion such as molecular structure is used to identify abstract tasks. This abstraction aids in the selection of technologies that can accomplish the necessary separations. Combined with novel representations of separation and mixing functionality, this approach enables the current prototype version of SPLIT to systematically develop promising flowsheet alternatives, including complex processes with recycles.