Design of multipurpose batch chemical plants using mathematical programming techniques

Savoula Papageorgaki, Purdue University

Abstract

Multipurpose batch plants feature general purpose equipment items shared among multiple products whose manufacture follow complex recipes. Such plants are particularly suitable for the production of high value added chemical and biochemical products because they provide the necessary flexibility to accommodate a large number of low volume chemicals in the same processing facility. The increased flexibility of multipurpose plants, however, leads to increased complexity of operation which imposes significant computational and analytical burdens to the related design problem. To date none of the existing solution approaches enable the designer to formulate the multipurpose plant design problem in its entirety. Two new mathematical formulations are developed in this work to describe the grassroots and retrofit design of multipurpose batch chemical plants, respectively. Both problems are posed as mixed integer nonlinear programs (MINLP) in which the binary variables are the structural choice variables. The proposed models are able to accommodate equipment used in- and out-of-phase, units available in multiple sizes, multiple choices of equipment types for each product task, allocation of products to campaigns and determination of the campaign lengths and sizing of the processing equipment. The retrofit design model incorporates additional aspects such as potential changes in the product demands and prices, revisions in the product slate, additions of new units in- and out-of-phase and elimination of old inefficient units. The capital cost of equipment and the net profit are selected as optimization criteria for each case, respectively. The complexity of the proposed models makes the underlying problems computationally intractable for direct solution using existing MINLP solution techniques. Consequently, a novel optimization strategy is devised for each case which partitions the original MINLP into a lower bound master problem and an upper bound subproblem and then solves an alternating sequence of these subproblems. The effectiveness of the proposed decomposition procedures is demonstrated with a number of test problems which were solved in reasonable computation times.

Degree

Ph.D.

Advisors

Reklaitis, Purdue University.

Subject Area

Chemical engineering|Industrial engineering

COinS