A complementarity‐based vapor‐liquid equilibrium formulation for equation‐oriented simulation and optimization

Vibhav Dabadghao, Jaffer Ghouse, John Eslick, Andrew Lee, Anthony P. Burgard, David C. Miller, Lorenz T. BieglerAIChE 2023 · DOI

TL;DR

This paper introduces a "square-flash" formulation for Vapor-Liquid Equilibrium (VLE) that uses complementarity constraints to handle phase transitions and supercritical conditions seamlessly. By replacing discrete "if-then" logic with smooth mathematical relationships, the authors enable robust, high-speed simulation and optimization of complex chemical processes within the IDAES Integrated Platform.

What Problem Does This Paper Address?

Vapor-Liquid Equilibrium (VLE) is the bread and butter of chemical engineering, appearing in everything from distillation columns to natural gas pipelines. However, calculating VLE is notoriously difficult for computers. Traditionally, these calculations are "procedural"—they rely on nested loops and discrete logic (e.g., "if the temperature is above the dew point, the liquid phase disappears").

This discrete behavior creates "non-smoothness," which is a nightmare for Equation-Oriented (EO) solvers. EO solvers prefer to solve all equations in a flowsheet simultaneously using gradient-based methods (like Newton’s method). When a phase disappears or a system goes supercritical, traditional models often crash or require "heuristics" (educated guesses) that slow down the process. The gap addressed here is the need for a VLE formulation that is mathematically smooth, robust enough to handle disappearing phases, and efficient enough for large-scale industrial optimization.

Methods

The authors propose a "square-flash" formulation based on complementarity constraints.

  • Complementarity Constraints: Instead of using logic gates, they use relationships like $0 \le s_V \perp V \ge 0$. This essentially says that either the vapor flow ($V$) is zero, or a "slack" variable ($s_V$) is zero.
  • Pseudo-Temperature ($\bar{T}$): They introduce a relaxed equilibrium temperature. When a system becomes all-liquid or all-vapor, $\bar{T}$ stays "pinned" at the bubble or dew point. This ensures that thermodynamic property libraries (which might fail if asked for properties of a non-existent phase) are always evaluated in a valid two-phase region.
  • Cubic Equation of State (CEOS) Integration: They leverage the mathematical properties of CEOS (like the Peng-Robinson model). By tracking the first and second derivatives of the cubic roots, the model can "know" which root corresponds to which phase without manual selection.
  • Smoothing: To make these constraints "solver-friendly," they use smooth approximations (controlled by a parameter $\epsilon$). This turns sharp corners into differentiable curves, allowing standard solvers like IPOPT to glide through phase transitions.
  • Pseudo-Roots: When a phase physically disappears, the cubic equation might not have a real root for it. The authors developed a "translated cubic" method to provide continuous "pseudo-roots," preventing the solver from seeing a mathematical void.

Datasets

The researchers evaluated their formulation using three primary case studies rather than traditional training datasets:

  1. Benzene-Toluene Binary Mixture: Used for $PQ$-flash simulations (constant pressure and heat duty) across subcritical and supercritical regimes.
  2. Nitrogen-Oxygen-Argon Ternary Mixture: A more complex system representing cryogenic air separation units.
  3. 30-Tray Distillation Column: A large-scale optimization problem (approx. 2,200 equations) involving the ternary mixture, specifically designed to test "disappearing phases" on individual trays.

Key Insights and Ideas

  • Modularity is King: Unlike previous optimization-based approaches, this "square" formulation doesn't require a custom objective function to handle the math. It can be dropped into existing process models as a modular block.
  • The "Aha" Moment on Supercriticality: By using complementarity constraints on pressure variables ($P^+$ and $P^-$), the model automatically forces vapor flow to zero above the critical point, treating the fluid as "liquid-like" without needing a separate set of equations for supercritical states.
  • Initialization Matters: The paper details a hierarchical initialization strategy. By solving sub-models (state blocks) in isolation before tackling the whole flowsheet, they significantly improve the chances of the solver finding a solution.

Key Results

  • Superior Robustness: In head-to-head comparisons with previous state-of-the-art methods (e.g., Burgard et al., 2018), the new formulation converged successfully across the entire pressure range. The previous method failed consistently near the critical point due to numerical degeneracy.
  • Significant Speedups: For the benzene-toluene case, the proposed formulation was twice as fast in solving and more than twice as fast during the initialization stage.
  • Handling "Dry Trays": The model successfully optimized a distillation column even when the specifications forced some trays to have zero liquid or vapor flow—a scenario that typically causes commercial simulation tools to fail.

Limitations and Future Work

  • Beyond VLE: Currently, the model focuses on Vapor-Liquid Equilibrium. The authors acknowledge that future work needs to extend this to more complex multi-phase systems (e.g., Liquid-Liquid-Vapor).
  • Distributed Systems: They aim to apply this logic to "distributed" models, such as long-distance pipelines or complex heat exchangers where phase behavior changes continuously over space.
  • Parallelization: Because the modular "square" approach lends itself well to parallel computing, the authors plan to explore parallel solution strategies to further speed up massive dynamic optimization problems.

Summary updated

Abstract

Abstract Vapor‐liquid equilibrium (VLE) is a cornerstone of computer‐aided process engineering (CAPE). Embedded within process system models, VLE calculations are inherently procedural with non‐smooth behavior that frequently requires discrete decisions. Traditionally, these features resist the incorporation of VLE within efficient, large‐scale equation‐oriented (EO) process simulation and optimization strategies. On the other hand, recent reformulation of VLE models through the incorporation of complementarity constraints has broadened its scope to deal seamlessly with phase transitions and even supercritical excursions in process simulation and optimization. In this study, we extend these VLE complementarity models to EO frameworks where procedural thermodynamic property libraries are still required. Here we develop an efficient, non‐intrusive, and intuitive “square‐flash” equation system that has been implemented within the IDAES Integrated Platform (IDAES‐IP). The effectiveness of this modular approach is demonstrated on case studies for non‐ideal flash calculations and distillation optimization, with disappearing phases and supercritical transitions.
BibTeX
@inproceedings{dabadghao2023complementaritybased,
  author = {Vibhav Dabadghao and Jaffer Ghouse and John Eslick and Andrew Lee and Anthony P. Burgard and David C. Miller and Lorenz T. Biegler},
  title = {{A complementarity‐based vapor‐liquid equilibrium formulation for equation‐oriented simulation and optimization}},
  booktitle = {AIChE},
  year = {2023},
  doi = {10.1002/aic.18029},
}

Loading related content...

Figure