Superstructure Optimization with Embedded Neural Networks for Sustainable Aviation Fuel Production

Alexander Klimek, Christoph Plate, Sebastian Sager, Kai Sundmacher, Caroline Ganzer · 2025

Read Superstructure Optimization with Embedded Neural Networks for Sustainable Aviation Fuel Production by Alexander Klimek, Christoph Plate, Sebastian Sager, Kai Sundmacher, Caroline Ganzer (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-09796.txt.

Introduction

Alexander Klimek and 4 co-authors optimise sustainable aviation fuel plants with neural networks embedded inside the planning maths in a 2025 engineering paper. Choosing a fuel route means picking discrete process steps and tuning continuous operating settings at once, while stream compositions vary along the way. Ordinary superstructure methods strain under that mix, so the team folds learned surrogate models directly into a mixed-integer optimisation problem.

Applied to Fischer-Tropsch kerosene, the cheapest layout with no carbon limit leans on fossil autothermal reforming. Tighten the emissions cap and biomass gasification plus direct air capture with storage move in, cutting net emissions at rising cost. At zero emissions, a hybrid of the fossil and biomass routes wins on price.

Catalog record

CREATORAlexander Klimek, Christoph Plate, Sebastian Sager, Kai Sundmacher, Caroline Ganzer
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE156 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-09796

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Superstructure Optimization with Embedded Neural Networks for Sustainable Aviation Fuel Production Alexander Klimek, Christoph Plate, Sebastian Sager, Kai Sundmacher, Caroline Ganzer Abstract This study presents a multi-objective optimization framework for sustainable aviation fuel (SAF) production, integrating artificial neural networks (ANNs) within a mixed-integer quadratically constrained programming (MIQCP) formulation. By embedding data-driven surrogate models into the mathematical optimization structure, the proposed methodology addresses key limitations of conventional superstructure-based approaches, enabling simultaneous optimization of discrete process choices and continuous operating parameters. The framework captures variable input and output stream compositions, facilitating the joint optimization of target product composition and system design. Application to Fischer-Tropsch (FT) kerosene production demonstrates that cost-minimizing configurations under unconstrained CO 2 \text{CO}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} emissions are dominated by the fossil-based autothermal reforming (ATR) route. Imposing carbon emission constraints necessitates the integration of biomass gasification and direct air capture coupled with carbon sequestration (DAC-CS), resulting in substantially reduced net emissions but higher production costs. At the zero-emission limit, hybrid configurations combining ATR and biomass gasification achieve the lowest costs ( \textapprox 2.38 $ / kg ​ kerosene 2.38\text{\,}\$\mathrm{/}\mathrm{k}\mathrm{g}\textsubscript{kerosene} ), followed closely by biomass gasification-only ( \textapprox 2.43 $ / kg 2.43\text{\,}\$\mathrm{/}\mathrm{k}\mathrm{g} ), both of which outperform the ATR-only pathway with DAC-CS ( \textapprox 2.65 $ / kg 2.65\text{\,}\$\mathrm{/}\mathrm{k}\mathrm{g} ). In contrast, DAC-only systems relying exclusively on atmospheric CO 2 \text{CO}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} and water electrolysis are prohibitively expensive ( \textapprox 10.8 $ / kg 10.8\text{\,}\$\mathrm{/}\mathrm{k}\mathrm{g} ). The results highlight the critical role of process adaptability: configurations exploiting flexible process parameters, facilitated by embedded ANNs, consistently outperform fixed setups, achieving up to 20 % 20\text{\,}\% cost savings. Sensitivity analyses elucidate the influence of process conditions, such as FT reactor pressure and gasification temperature, on economic and environmental performance. keywords Sustainable Aviation Fuel, Superstructure Optimization, Surrogate Modeling, Artificial Neural Networks 1 Introduction The defossilization of aviation represents a formidable challenge in the global transition toward net zero greenhouse gas (GHG) emissions. In 2019, civil aviation was responsible for over 900 million metric tons of \text{CO}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} -equivalent emissions, accounting for approximately 10\text{\,}\% of global transportation-related emissions. 1 Forecasts anticipate a four- to six-fold increase in air traffic over the coming decades, driven by population growth, urbanization, and globalization. 2 , 3 Whilst \text{H}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} -fueled aircrafts and electrification may present viable options at small scale and for niche applications, it is expected that long-chain liquid sustainable aviation fuels (SAFs) are crucial at large scale. 4 SAFs are hydrocarbon-based fuels synthesized from renewable carbon ( C ) and \text{H}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} sources via certified thermochemical or biochemical conversion pathways. Unlike alternative propulsion technologies, SAFs are compatible with existing aircraft engines, fuel handling systems, and airport infrastructure. As such, they can serve as drop-in replacements for conventional jet fuels (e.g., Jet-A1), facilitating a smoother and less disruptive transition to low-carbon aviation. The American Society for Testing and Materials (ASTM) International has codified the production and compositional criteria for SAFs under Standard D7566, with multiple conversion pathways such as Fischer–Tropsch Synthetic Paraffinic Kerosene (FT-SPK), Hydroprocessed Esters and Fatty Acids (HEFA-SPK), and Alcohol-to-Jet Synthetic Paraffinic Kerosene (ATJ-SPK) having already received certification for commercial deployment. 5 , 2 , 6 However, the design of economically and environmentally optimal SAF production pathways remains a multi-dimensional optimization problem. This complexity arises from the wide array of feasible raw material sources, including biomass, captured \text{CO}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} , and renewable \text{H}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} , as well as the multiplicity of intermediate chemistries (e.g., syngas, methanol, FT liquids) and energy integration schemes. 7 These process network synthesis problems are further complicated by nonlinear thermodynamic relationships, discrete design choices, and sustainability constraints such as \text{CO}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} emissions and land use. Identifying viable SAF production configurations requires a systematic modeling approach that accounts for process interdependencies and interactions among material and energy flows as well as emissions. 3 Recent techno-economic analyses have assessed various power-to-liquid production routes, revealing significant cost disparities and efficiency trade-offs. For instance, Salem 8 emphasizes the wide cost range of methanol-to-jet (MtJ) pathways ( 650\text{\,}\euro to 4630\text{\,}\euro per tonne) primarily driven by the price of \text{H}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} and \text{CO}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} . Sacchi et al. 9 present a time-dependent life-cycle assessment comparing fossil jet fuel with DAC-CS offsets to synthetic jet fuel from DAC and electrolysis. They show that climate outcomes depend strongly on the carbon footprint of electricity, with synthetic fuels reducing \text{CO}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} storage needs if powered by low-carbon electricity. Do et al. 10 develop a techno-economic framework to evaluate \text{CO}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} -to-fuel pathways, including aviation fuels, using process simulation and optimization to assess various technologies and scenarios. Meanwhile, Albrecht et al. 11 demonstrate that the viability of different SAF pathways depends heavily on electricity prices, with biomass-to-liquid favored under higher electricity prices. Comparative studies by Bube et al. 12 , 7 and Voß et al. 13 indicate that MtJ offers higher kerosene selectivity and lower energy input, while hybrid routes based on biomass and electrolytic \text{H}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} achieve higher C efficiency at a premium cost. Eyberg et al. 14 underline the persistent economic gap with fossil jet fuel, suggesting that policy support and technological advancements will be pivotal. A growing body of research has applied optimization-based approaches to the design and integration of SAF production pathways, adopting a variety of system boundaries and modeling strategies. Gonzalez-Garay et al. 3 formulate a superstructure optimization of solar-powered SAF production based on Aspen Plus ® 15 simulations, using fixed operating conditions and lumped component modeling. Demirhan et al. 16 develop a multi-scale optimization framework for fuels, chemicals, and power generation, incorporating multiple \text{H}{\vphantom{\text{X}}}_{\vphantom{\text{2}}\smash[t]{\text{2}}}^{\vphantom{\smash[t]{\text{2}}}\hphantom{\text{2}}\text{}} production routes and policy incentives. Martín and Grossmann 17 and Zhang et al. 18 analyze renewable fuels and power integration in Spain using linear models with temporal discretization. Kenkel et al. 19 integrate Power-to-Jet and Biomass-to-Jet production in a refinery superstructure including algae-based feedstocks, while Restrepo-Flórez and Maravelias 20 extend ethanol upgrading studies with a richer representation of fuel components and property constraints. Niziolek et al. 21 examine municipal solid waste conversion to fuels and chemicals via a broad reaction network, and Wu et al. 22 present a systematic representation framework for process superstructures based on unit–port connectivity. Finally, Wang et al. 23 propose a multiobjective Mixed-Integer Nonlinear Programming (MINLP) model for biomass gasification-based hydrocarbon biorefineries, integrating economic and environmental criteria. Across all these studies, simplifications are common in the treatment of component mixtures, operating conditions, and carbon management. Most approaches rely on lumped components and linear process models, fixed operating parameters, and predetermined stream compositions. Limited consideration of biomass, atmospheric carbon, and carbon capture further restricts flexibility in design. As a result, opportunities remain to advance modeling frameworks that incorporate nonlinear process behavior, detailed mixture representations, variable operating conditions, broader feedstock options, and emissions management strategies to more fully capture the trade-offs in SAF production. Surrogate modeling has become a vital tool in chemical process engineering, especially for enabling tractable optimization of complex systems. As summarized by McBride and Sundmacher, 24 commonly used surrogate models include polynomials, Gaussian processes (Kriging), ANNs, and radial basis functions. While Kriging dominates surrogate-based optimization studies, ANNs are increasingly used for high-dimensional, nonlinear problems. Surrogates are typically trained via sequential data generation from simulation platforms such as Aspen Plus ® . 15 Several studies have explored the integration of surrogate models into superstructure or flowsheet optimization. Fahmi and

Read the complete text