A Data-Driven Optimization Approach and Software Toolkit for Modular Process Intensification Synthesis

Modular process intensification aims to develop more efficient, more environmentally friendly, and more compact technologies by combining multiple process tasks into a single equipment, maximizing driving forces, and maximizing heat and mass transfer rates. A key research question is how to systematically generate innovative process designs. To this purpose, phenomena-based process synthesis approaches have been proposed which offer the potential to re-invent unit operations by finding optimized process solutions without pre-postulation of equipment/flowsheet alternatives. Herein, phenomena are fundamental chemical process functions such as heat transfer and mass transfer, which can serve as building blocks to represent unit operations from a lower aggregated level. A major challenge in applying phenomena-based synthesis is the computational complexity due to the large-scale optimization problem and the nonlinearities introduced by modeling the first principles physical phenomena.

To address this challenge, we propose a data-driven optimization approach for phenomena-based process intensification synthesis using the Generalized Modular Representation Framework (GMF). In GMF, two types of phenomenological building blocks are used: pure heat exchange modules (HE) and mass/heat exchange modules (M/H). The mass transfer feasibility in a M/H module is characterized by the Gibbs free energy-based driving force constraints, which contribute to the key GMF representation capability to identify multifunctional separation and/or reaction tasks while also render the major mathematical complexities with logarithmic terms and multivariable polynomials. In view of this, the rectified linear units (ReLU) approach is applied to generate a data-driven model to correlate driving forces with key process variables (e.g., temperature, molar fractions) in a mixed-integer linear formulation. This offers the advantages to: (i) reduce the computational strain to identify optimal solutions, and (ii) extract simplified mathematical expressions to spotlight the synergistic relations of multifunctional phenomena. The data-driven model is then integrated with other GMF modeling constraints on mass balances, energy balances, and superstructure combination rules to synthesize modular and intensified process systems. The proposed hybrid data-driven/mechanistic GMF synthesis approach will be showcased on two case studies: (i) reactive distillation optimization to demonstrate the solution optimality and computational efficiency with respect to continuous optimization, and (ii) modular reactor network optimization for decision making with mixed-integer variables. The automation and integration of this strategy as a software toolkit to the SYNOPSIS prototype platform will also be presented.