Abstract
Simultaneous synthesis and design of reaction–separation–recycle processes using rigorous models is highly desirable to improve process efficiency. However, it often leads to a large-scale highly challenging optimization problem. In this work, we propose a computationally efficient optimization framework for the challenging problem. The reactor and separator networks are modeled using the generalized disjunctive programming, which are reformulated into a highly nonconvex mixed-integer nonlinear programming (MINLP) formulation using the convex-hull method. To solve the complex MINLP model, a systematic solution approach is proposed in which an initialization strategy is first proposed to generate a feasible solution for a partially relaxed synthesis problem using the hybrid steady-state and time-relaxation optimization algorithm. A successive relaxed MINLP solution strategy is then adopted to solve the original model to local optimality. The computational results demonstrate that the proposed framework obtains better solutions with less computational effort, by ∼1 order of magnitude, than the existing algorithms.
Highlights
Developing sustainable chemical processes is a key issue in the current world.[1]
We proposed a computationally efficient optimization-based framework for simultaneous synthesis and design of the reaction−separation−recycle processes using rigorous models
The superstructure was modeled through generalized disjunctive programming (GDP), which was reformulated to a mixedinteger nonlinear programming (MINLP) model, using the convex hull method
Summary
Developing sustainable chemical processes is a key issue in the current world.[1]. Since chemical processes are usually composed of reactors, separators, and recycle streams to recover raw materials, we often need to determine the types of reactors and separators, their connection structure, and optimal operating conditions when designing a chemical process. The reactor type is first selected based on existing experience and the operating parameters and reactor sizes are determined,[4] which is difficult to be used for complex reaction systems.[5] A more general method, the concept of attainable region, was first proposed by Horn[6] referring to the set of points in concentration space that can be obtained through reaction and mixing. The superstructure for the reactor network and distillation system was modeled using GDP, which was converted to a nonconvex MINLP problem using the big-M method This framework using rigorous models can find a locally optimal solution and the generated design has higher accuracy than the surrogate-based optimization approach.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
