458643 Implementing Robust Vapor-Liquid Equilibrium Calculations in Nonsmooth Multi-Stream Heat Exchanger Models

Wednesday, November 16, 2016: 12:30 PM
Carmel I (Hotel Nikko San Francisco)
Matias Vikse1, Harry A. J. Watson2, Truls Gundersen1 and Paul I. Barton3, (1)Energy and Process Engineering, Norwegian University of Science and Technology, Trondheim, Norway, (2)Process Systems Engineering Laboratory, Massachusetts Institute of Technology, Cambridge, MA, (3)Chemical Engineering, Massachusetts Institute of Technology, Cambridge, MA

Optimization of the liquefaction process for natural gas has proven to be a challenging task in particular for two reasons. First, the refrigeration cycles introduce computational loops that cause convergence problems. Second, the refrigerants are mostly multicomponent mixtures with strong interactions in the search for optimal heating profiles to match the cooling profile of the natural gas. In addition, the driving forces in the heat exchangers are very tight, asking for rigorous modeling to make sure the simulation mimics the true behavior of the process.

Both gradient based methods and stochastic search methods have been used to try to optimize processes for Liquefied Natural Gas (LNG) production, where commercial simulators are used to provide the function values in the optimization, however, with limited success beyond the most simple liquefaction processes. Our research groups have been involved in the development of a new model for these processes with the inherent features required to be able to utilize the latest developments in the field of Global Optimization [1].

As an important part of this new approach to LNG optimization, Barton and coworkers at MIT [2] has developed a new non-smooth multi-stream heat exchanger model accounting for phase changes in the process. The model was successfully used to simulate the Poly Refrigerant Integrated Cycle Operations (PRICO) process for liquefied natural gas production. However, it suffered from convergence difficulties when studying more complex single mixed refrigerant (SMR) processes. The primary challenge was flash calculations, which frequently failed even for initial guesses close to the solution. These problems will of course increase even further when moving to Dual Mixed Refrigerant (DMR) processes and the Mixed Fluid Cascade (MFC) process.

In this project, an alternative model structure is studied, replacing the equation-oriented framework with a hybrid solution. Rather than converging the vapor-liquid equilibrium calculations simultaneously with the other model equations, they are included as nested subroutines and solved sequentially. Nonsmooth Newton-type methods and the implicit function theorem for lexicographically smooth functions are used for computing analytical derivatives in each subroutine [3, 4]. In addition, more robust flash calculations are attempted by developing a nonsmooth version of the “inside-out” algorithm by Boston and Britt [5] that can be incorporated in the existing model framework. The same SMR processes are then simulated in order to investigate whether the alternative framework is more robust.

[1] Floudas, C.A. and Gounaris, C.E., A review of recent advances in global optimization, Journal of Global Optimization, vol. 45(1), pp. 3-38, 2009.

[2] Watson, H.A.J., Khan, K.A. and Barton, P.I., Multistream heat exchanger modeling and design, AIChE Jl., vol. 61(10), pp. 3390-3403, 2015.

[3] Facchinei, F., and Pang, J.S., Finite-dimensional variational inequalities and complementarity problems, vol. 2. Springer Science & Business Media, 2007.

[4] Khan, K. A., and Barton, P.I., Generalized derivatives for hybrid systems, Submitted, 2015.

[5] Boston, J.F. and Britt, H.I., A radically different formulation and solution of the single-stage flash problem, Computers & Chemical Engineering, vol. 2(2), pp. 109-122,


Extended Abstract: File Not Uploaded
See more of this Session: Energy Systems Design and Operations II
See more of this Group/Topical: Computing and Systems Technology Division