Chemical Process Design under Uncertainty-Models and Algorithms for Global Optimization

2018-02-08T00:00:00Z (GMT) by Jeonghwa Moon
Uncertainty is a fundamental factor we should consider in process design because it is inherent characteristics of any process systems. For instance, physical properties of reactants, kinetics, or value of transfer coefficient are uncertain in the design stage. Also we expect disturbances – changes of flowrate, compositions, pressure, and temperature- during the operation. However, classical process design uses only nominal information of physical properties, transport phenomena, and reactions to find design variables in order to optimize process performance. In reality, the classical design procedure is not suitable because these parameters are not exactly known but have unavoidable variations, leading to uncertainty in the system. In addition, in practical operation the processes are exposed to changing conditions called dynamic disturbance, such us cooling water temperature or raw material quality that are dynamically changing. In order to accommodate uncertainties and dynamic disturbance in classic design, the process is usually oversized to minimize risk of operating outside specifications. However, this arbitrary overdesign does not guarantee feasibility and optimality of the process. Thus it is clear that consideration of uncertainty is necessary and important for the optimality and feasibility of operation of the chemical plant. The first aim of this thesis is to develop novel methodologies to tackle problems of classical approach for design under uncertainty. Two main topics in design under uncertainty –flexibility analysis and integrations of design and control dealt with this thesis. Part A addresses flexibility analysis of process. A new hybrid algorithm for flexibility analysis problem is suggested. Flexibility analysis is to quantify flexibility of a given process design to handle uncertainty in process parameters as well as variations in operating conditions. It is one of important problem in “design under uncertainty”. It is formulated as a multistage global optimization problem, whose search space is discontinuous and non-differentiable. Traditional local deterministic approaches cannot solve this problem properly, so I used a new approach based stochastic method and project technique to tackle this problem. This approach can be easily parallelized, so it reduces computational time when we solve large size problems. In part B, the problem of integrating design and control is addressed. Integration of design and control is finding an optimal design considering dynamic controllability. It aims at pursuing the synergetic power of a simultaneous approach to guarantee the economical and robust operation of the process in spite of any disturbance and uncertainty. However integration of design and control renders a complex combinatorial optimization problem which cannot be solved directly with existing mathematical methods. Thus we suggested a decomposition technique which eases the problems of this integration called embedded control optimization. In this thesis, I will extend embedded control optimization for integration of design and control. A new identification method is adopted to produce a better performance, and this methodology will be applied to large-scale processes successfully. The second area this thesis considers is global optimization. Global optimization applications are widespread in all disciplines. Despite there are many challenging and important problems that require global solutions, relatively little effort has been made in this area compared to the area of local optimization. Specially, the problem of finding all solutions in nonconvex search area remains as still challenging and difficult area in applied mathematics, engineering, and sciences. Part C addresses global optimization for multimodal objective functions. A novel hybrid sequential algorithm is suggested in this part. It aims to find multiple global solutions as well as local solutions. To locate multiple optimal points, it uses niche concept. It also adopts a local deterministic method to accelerate finding solutions. This algorithm was applied to tackle multiplicity problems in engineering problems such as finding multiple optimal parameters of distributed systems in problem inversion