Soo Hyoung Choi

A stochastic approach to global optimization of chemical processes

Abstract Chemical process optimization often leads to large nonconvex nonlinear programming problems that have many nonlinear equality constraints. Since the global optimization of such a problem is one of the toughest NP-hard problems, large problems in many cases cannot be solved in a reasonable time span if we rely solely on deterministic algorithms that are theoretically guaranteed to find the global optimum. Generally, stochastic algorithms, which do not guarantee the global optimality of the obtained solution, are suitable for large problems, but not efficient when there are too many equality constraints. Therefore, an algorithm suitable for general chemical process optimization problems is proposed in this paper, which is based on a feasible point strategy and combination of a stochastic global algorithm and a deterministic local algorithm.

GLOBAL OPTIMIZATION OF CHEMICAL PROCESSES USING THE INTERVAL ANALYSIS

Optimization of chemical processes often leads to nonlinear programming problems that are nonconvex. Such problems may possess many local optima, whose objective function values vary significantly from one to another. Thus identifying the global optimum is an important, albeit difficult, endeavor. A deterministic algorithm based on interval analysis branch and bound is proposed in this paper to be suitable for global optimization of chemical processes.

Global Optimization Methods for Chemical Process Design: Deterministic and Stochastic Approaches

Process optimization often leads to nonconvex nonlinear programming problems, which may have multiple local optima. There are two major approaches to the identification of the global optimum: deterministic approach and stochastic approach. Algorithms based on the deterministic approach guarantee the global optimality of the obtained solution, but are usually applicable to small problems only. Algorithms based on the stochastic approach, which do not guarantee the global optimality, are applicable to large problems, but inefficient when nonlinear equality constraints are involved. This paper reviews representative deterministic and stochastic global optimization algorithms in order to evaluate their applicability to process design problems, which are generally large, and have many nonlinear equality constraints. Finally, modified stochastic methods are investigated, which use a deterministic local algorithm and a stochastic global algorithm together to be suitable for such problems.

Stochastic dynamic simulation of chemical processes with changing uncertainties

Abstract Process simulation is essential to the economic evaluation and the reliability or safety analysis of a chemical process. However, conventional similators do not provide users with the information on the accuracy of the results or on the effects of the uncertainties in the data used in the simulation. Therefore, a new simulation method is required which deals with uncertainties in the input variables and in the model parameters. A methodology for stochastic simulation is proposed in this paper, which is based on Monte Carlo simulation. The results of sensitivity analysis numerically and graphically show the trend of the change in the uncertainties of the process variables, changes in the importance of the variables, and the relations between the variables. The proposed approach was implemented in a general purpose dynamic process simulator, MOSA, and showed good applicabilities for chemical processes with various uncertainties. One of the most important advantages of the proposed method is that it can use the deterministic models used in the conventional simulators without any modifications.

A reliability model for process systems under changing operating conditions

Reliability analysis of process systems, which is often based on a model of Weibull distribution, is semi-quantitative at best because it uses constant parameters, requiring assumption of steady state operating conditions. A reliability model based on a variable scale parameter Weibull distribution is proposed in this work, in which a power law, the Arrhenius factor, and instantaneous amplitudes and frequencies of the operating condition variables are introduced. Numerical experiment indicates that when an operating condition variable fluctuates, the assumption of an average steady state operating condition can cause a serious error in reliability analysis. Therefore, the proposed method is expected to contribute to more quantitative risk assessment, and thus more rigorous safety analysis of process systems under changing operating conditions.

A Method for Optimal Operation of BOG Compressors in a LNG Gasification Plant

An algorithm for optimal operation of BOG (boil off gas) compressors in a LNG gasification plant is proposed in this paper, which uses an empirical BOR (boil off rate) model, a MILP (mixed integer linear programming) formulation, and a simplified dynamic tank model. Given the values for a set of process state variables, the proposed algorithm generates an optimal operation schedule for BOG compressors, which minimizes the power consumption while preparing against the potential failure of one of the operating compressors. The performance of the proposed method is compared with that of the conventional method in terms of safety and energy.

Reliability and Availability Analysis of Process Systems under Changing Operating Conditions

Abstract Quantitative risk analysis for process safety often requires reliability and availability analysis of process systems. Although most systems are operated under changing conditions, conventional methods use reliability models with constant parameters only, and thus average operating conditions should be applied. Furthermore, a simplified availability equation is frequently used. Case studies indicate that using an average operating condition for reliability or using a simplified availability equation can cause a serious error. A rigorous method is proposed in this work which is suitable for solving the exact availability equation using the reliability under changing operating conditions.

Effects of Graphene on a Resin Transfer Molding Process Using Bisphenol A Based Epoxy Resin

Resin transfer molding is a popular process to fabricate polymer composites reinforced with a large amount of glass or carbon fibers. In general, fiber reinforcements are put in a mold, and a liquid resin such as epoxy resin is injected into the mold after preheating. For successful production of polymer composites via a resin transfer molding process, the filling and curing stages of the liquid resin as well as the mold design should be optimized. Recently, polymer composites reinforced with nanoparticles are attracting attention of researchers in academia and industries because efficient reinforcement can be achieved by small loading of nanoparticles such as carbon nanotubes and exfoliated clays. In this work, as an effort to develop light weight automotive parts, graphenes were investigated as a nano size reinforcement of epoxy resin for resin transfer molding. Graphenes were prepared from graphites by microwave irradiation. Addition of graphenes to bisphenol A based epoxy resins such as YD-128 from Kukdo Chemical results in an increase in viscosity and shear thinning behavior, affecting the filling process. The curing of epoxy resins is also affected by graphenes. In order to develop a model for simulation of the filling and curing of epoxy resins containing different amounts of graphenes in the resin transfer molding, FLUENT and MATLAB have been used in this study, which are a finite element based computational fluid dynamics analysis tool and a general purpose numerical analysis tool, respectively. The effects of graphenes on the mold filling pattern and curing profile are discussed for the resin transfer molding of bisphenol A based epoxy resins.

Calculation of probability distributions of output variables in process simulation

Abstract Stochastic process analysis is often based on Monte Carlo simulations. As a more rigorous alternative, a deterministic algorithm based on numerical integration is proposed in this paper, which calculates the probability distributions of dependent random variables using the results of simulation at grid points of independent random variables. For performance evaluation, the proposed algorithm is applied to an example problem which can be analytically solved, and the result is compared with that of Monte Carlo simulation. The proposed algorithm is suitable for general process simulation problems with a few independent random variables, and expected to be applicable to areas such as safety analysis and quality control.

An Operating Procedure Synthesis Method for Shutdown of Boiler Plants

Abstract A method is proposed which automatically synthesizes a safe operating procedure based on the knowledge of human operators that are generalized and classified hierarchically. The proposed method automatically generates and orders subgoals that must be achieved between the initial state and the goal state. It searches for the primitive operation device to achieve each subgoal subject to the operational constraints, and evaluates the safety of the operation. A case study of automatic synthesis of a boiler plant shutdown procedure shows the industrial applicability of the proposed method.