Computational Science Approach for Design and Optimization of Sustainable Process and Its Industrial Applications

Abstract

Recently, in the field of chemical engineering, many types of research based on high-performance computing have been combined with computer-aided process systems engineering. Therefore, various techniques of computational science such as computational fluid dynamics, optimization methodology, and machine learning have been applied to the problems of chemical reactor modeling and process optimization. Notably, in this advance computational science approach, the scope of research extends to non-traditional fields such as reactive research according to the 3D shape of the reactor that has not been easily solved in the past and surrogate model based optimization using machine learning. In this thesis, various methods are proposed to obtain the maximum profit with minimum cost by making a breakthrough design.

In parallel, there is a growing demand for sustainable chemical processes in chemical engineering. Conventional chemical processes are highly dependent on oil prices, and unless a diverse portfolio is designed, the sustainability of their chemical industries can be violated because of the oil controlling from the Middle East or US. In addition, these crude oil based chemical processes and power plants generate a great deal of CO2. Therefore, it is not necessary to capture these CO2 and make only meaningless storage but to reproduce it as a product that can be used and make it economical carbon capture, utilization, and storage (CCUS) technology. To solve this series of processes, the Gas-to-Liquid (GTL) process and CCUS are being researched and developed in various ways.

In this thesis, I will discuss the process modeling, optimizing, and designing the reactor and process using CFD, mathematical programming, machine learning, deep learning, and derivative-free optimization techniques in computational science. First of all, the Fischer-Tropsch microchannel reactor and 3-phase carbonation reactor, which are the key reactor of two most important processes of the sustainable process, the gas-to-liquid process (GTL) and the carbon capture, utilization, and storage (CCUS), are modeled by CFD. Also, we propose an integration platform of CFD model and process simulator and conduct research from the point of view of combining with existing process engineering.

With these advanced reactor model, we propose a multi-objective optimization methodology using a stochastic optimization algorithm, a genetic algorithm (GA) with e-constraint method for simultaneously maximizing C5+ productivity and minimizing the temperature rise of a Fischer-Tropsch microchannel reactor. The main mixed integer nonlinear programming (MINLP) optimization problem is decomposed into an external CFD reactor model function and internal optimization constraints. The methodology is applied to the catalyst packing zone division, which is divided and packed with a different dilution ratio to distribute the heat of reaction evenly. The best solutions of the proposed optimizer are reproducible with different crossover fractions and are more efficient than other traditional non-convex constraint local solvers. Based on the Pareto optimal solution of the final optimizer with 4 zones, discrete dilution increases C5+ productivity to 22% and decreases ∆Tmax to 63.2% compared to the single zone catalyst packing case. Finally, several Pareto optimal solutions and sub-optimal solutions are compared and the results are documented in terms of C5+ productivity and maximum temperature increase.

In process scale optimization platform, a modified DIRECT algorithm with a sub-dividing step for considering hidden constraints is proposed. The effectiveness of the algorithm is exemplified by its application to a cryogenic mixed refrigerant process using a single mixed refrigerant for natural gas liquefaction and its comparison with a well-known stochastic algorithm (GA, PSO, SA), and model based search algorithm (SNOBFIT), local solver (GPS, GSS, MADS, active-set, interior-point, SQP), and other hidden constraint handling methods, including the barrier approach and the neighborhood assignment strategy. Optimal solution calculated by the proposed algorithms decreases the specific power required for natural gas liquefaction to 18.9% compared to the base case. In the same chapter, heat exchanger network synthesis (HENS) has progressed by using mathematical programming-based simultaneous methodology. Although various considerations such as non-isothermal mixing and bypass streams are applied to consider real world alternatives in modeling phase, many challenges are faced because of its properties within non-convex mixed-integer nonlinear programming (MINLP). We propose a modified superstructure, which contains a utility substage for use in considering multiple utilities in a simultaneous MINLP model. To improve model size and convergence, fixed utility locations according to temperature and series connections between utilities are suggested. The numbers of constraints, discrete, and continuous variables show that overall model size decreases compared with previous research. Thus, it is possible to expand the feasible search area for reaching the nearest global solution. The models effectiveness and applications are exemplified by several literature problems, where it is used to deduce a network superior to that of any other reported methodology.

In the case of plant-wide scale systems, a non-linear surrogate model based on deep learning is proposed using a variational autoencoder with deep convolutional layers and a deep neural network with batch normalization (VAEDC-DNN) for real-time analysis of the probability of death (Pdeath). VAEDC can extract representation features of the Pdeath contour with complicated urban geometry in the latent space, and DNN maps the variable space into the latent space for the Pdeath image data. The chlorine gas leak accident in the Mipo complex (city of Ulsan, Republic of Korea) is used for verification of the model. The proposed model predicts the Pdeath image within a mean squared error of 0.00246, and compared with other models, it exhibits superior performance. Furthermore, through the smoothness of image transition in the variable space, it is confirmed that image generation is not overfitting by data memorization.

Finally, a pilot scale (1.0 BPD) compact GTL process comprising of reforming section, CO2 separating section and Fischer -Tropsch (FT) synthesis section is presented. Systematic design procedure adopted for the design of a modular 0.5 BPD microchannel FT reactor block design consisting of 528 process channels is described. On average 98.27% CH4 conversion to syngas in reforming section comprising of a pre-reformer unit and a tri-reformer unit, CO2 separation rate of 36.75 % along with CO/H2 reduction from 2.67 to 2.08 in CO2 membrane separation section comprising of three membrane separators, for the entire plant operation duration of 450 hr demonstrated successful and stable operation of pre-processing sections of the present pilot-scale compact GTL process. Parallel operation of FT microchannel reactor and multitubular fixed bed type FT reactor proved failure for latter due to reaction runaway, while the former showed stable operation with high CO conversion of 83% and successful temperature control (at 220 oC, 230 oC and at 240 oC during the 139 hr operation), which demonstrated the appreciable performance of KOGAS-SNU novel microchannel FT reactor. Furthermore, a tank agitator carbonation reactor in which the reaction between calcium oxide and carbon dioxide takes place is studied to understanding that how 6 design variables (the number of impeller, impeller type, D/T, clearance, speed, baffle) affect to the solid dispersion using CFD simulation.