Introduction
To meet global decarbonization and climate goals, the rapid deployment of chemical technologies—such as carbon capture and water desalination—is essential. However, the traditional chemical engineering paradigm relies on "bespoke" designs, where every plant is uniquely optimized for its specific location, feedstocks, and capacity. While this maximizes local efficiency, it is inherently slow and expensive; engineering costs alone can account for 10% to 30% of a project's total expenditure. Furthermore, manufacturing unique parts for every installation prevents "economies of numbers," the cost savings achieved through mass-producing standardized components.
This paper proposes a transition from one-off engineering toward a "platform" model, similar to the automotive industry's approach of building multiple vehicle models on a single shared chassis. The authors introduce a scalable mathematical framework to design "process families" that utilize standardized, modular components, drastically reducing engineering effort and manufacturing costs.
Methodology
The core of the research is a transition from complex, nonlinear models to a highly efficient Mixed-Integer Linear Programming (MILP) formulation. The methodology follows a three-step process:
- Nonlinear GDP Formulation: The problem is initially framed as a Nonlinear Generalized Disjunctive Programming (GDP) model. This captures the logic of choosing between "common" platform designs and "unique" site-specific designs.
- Discretization and Pre-computation: To manage the computational complexity of solving nonlinear problems for dozens of plant variants simultaneously, the authors discretize the design space. They create a "catalog" of candidate designs for key units (e.g., specific sizes of heat exchangers or compressors) and run thousands of independent, small-scale simulations to determine the feasibility of these designs across different plant variants.
- The MILP "P-median" Model: The pre-computed data is fed into an MILP model. This formulation is mathematically analogous to the "P-median" problem used in logistics (e.g., determining the optimal location of warehouses to serve a set of customers). A key technical advantage of this model is its scalability: the number of binary variables scales with the number of candidate designs in the catalog, rather than the number of plant variants.
Case Studies and Datasets
The framework was evaluated using two high-fidelity case studies representing critical infrastructure technologies:
- Water Desalination: Utilizing the PARETO framework, the authors modeled 76 different process variants defined by varying inlet salt concentrations and water flow rates.
- Carbon Capture: Based on the CCSI2 (Carbon Capture Simulation for Industry Impact) model in Aspen Plus®, the researchers modeled 63 variants with different CO2 concentrations and flue gas flow rates.
Key Insights
The Platform Analogy
Just as an automotive manufacturer might use one engine across five different vehicle models to save costs, chemical engineers can use the same absorber column or compressor across dozens of industrial sites. This shift moves the industry away from local optimization toward global fleet efficiency.
Computational Scalability
The authors highlight a unique property of the P-median problem: even when binary requirements are relaxed (allowing variables to exist between 0 and 1), the solver naturally converges to integer values. This makes the optimization significantly faster than typical integer problems. In both case studies, once the pre-computation was complete, the Gurobi solver identified the optimal family design in less than one second.
The Standardization Trade-off
There is an inherent trade-off between standardization and performance. A standardized module might be slightly "off-design" for a specific site compared to a custom-built unit. However, the research suggests that the savings in manufacturing and engineering time likely far outweigh the minor losses in local thermodynamic efficiency.
Results
The MILP approach demonstrated massive simplification of complex systems:
- Desalination: 76 unique potential designs were reduced to a platform of just 3 evaporator designs and 3 compressor designs that covered the entire range of requirements.
- Carbon Capture: For 63 variants, the model successfully identified an optimal platform consisting of 3 absorber and 3 regenerator designs.
- Efficiency: By avoiding dozens of unique engineering cycles, the process family approach significantly lowers the barrier to mass deployment of green technologies.
Limitations and Future Work
While the framework proves the viability of the method, several areas for future research remain. Currently, the number of common designs (e.g., choosing "3" designs) is selected manually; future iterations aim to make this an optimization variable balanced against "economies of numbers" discount factors. Additionally, the current model focuses on steady-state operations. Future work must integrate dynamic behaviors—such as startup, shutdown, and control—into the off-design analysis to ensure the robustness of modular plants. Finally, the authors aim to explicitly quantify the exact economic savings achieved through mass production in subsequent studies.
Summary updated