Summary:

Separation processes make up 40%-70% of capital and operating costs of chemical manufacturing. Distillation accounts for more than 60% of the total process energy for the manufacture of commodity chemicals. Complex column configurations are estimated to harness energy savings up to 70%. Therefore, distillation synthesis with complex is a meaningful target for energy improvements on an industry-wide scale. However, design of complex separation system requires (i) structural decisions such as sequencing of distillation columns for solvent recovery, and (ii) determination of associated operating conditions (e.g. distillate/.bottoms compositions, reflux, ratio etc). For computer tools this process synthesis constitutes a very formidable challenge. Technically MINLP incorporating binary as well as continuous variables can address synthesis problem. However, infeasible operation following from ineffective structural decision may jeopardize the robustness and convergence of MINLP. In this presentation, we demonstrate a hybrid algorithm combining stochastic with gradient-based search techniques. An evolutionary algorithm will construct automatically structurally different complex separation network. Feasibility of the design will be delegated to an advanced feasibility tests based on temperature collocation of finite elements. Our search algorithm is capable of solving complex distillative synthesis problem robustly and reliably without user interactions. Application will demonstrate the algorithms performance in complex column sequencing problems for solvent recovery.

Scope

We propose a novel bi-level hybrid algorithm for the synthesis of complex column sequences. The master problem will make structural decisions using evolutionary algorithms, while the subproblem will rigorously assess the feasibility of each of these structures using our novel MInimum Bubble Point DIstance (MIDI) Algorithm. The minimum distance algorithm computes the section temperature profiles by uses orthogonal finite-element collocation technique.

Each specification e.g. flowsheet structure, reflux ratio, distillate and bottoms composition, can be represented by a novel chromosome. At every generation, the chromosomes are cultivated by crossover and nonuniform dynamic mutation. In order ensure feasibility of the chromosomes, our MIDI algorithm computes the minimum bubble point distance for each specification, i.e. a chromosome. This minimum bubble point distance is obtained by solving a gradient based NLP optimization problem. A minimum distance of zero ensures feasibility, whereas positive values indicate infeasible specifications. Penalizing the genetic algorithm's objective for infeasible specifications ensures that the chromosomes are feasible after a few generations.

Significance:

The major innovation of the proposed approach is an evolutionary program combined with a novel temperature collocation algorithm to systematically build and optimize complex column configurations. Massive problem size reductions due to temperature collocation ensure the realistic composition profiles of each column in the network without sacrificing the computational and thermodynamic rigor. The evolutionary algorithm constructs automatically different separation sequences; feasibility of the synthesized sequences is ascertained by means of the algorithm as a subroutine to the master GA. In contract to existing approaches, this method synthesizes several clusters of designs solutions, each one corresponding to regions of local optimality.

Extended Abstract Status: File Uploaded