DeepFuel-kit

DeepFuel-kit

DeepFuel-kit is an open-source Python library for AI-guided formulation of next-generation sustainable fuels.

The reduction of greenhouse gases (GHG) is becoming a critical priority for energy utilisation reliant on fossil fuels. Liquid fossil fuels — including petrol, diesel, and kerosene — constitute over 30% of the global total energy supply and remain the most significant single fuel source. Due to their advantageous properties, such as high energy density and ease of storage and transportation, liquid fuels are expected to continue their utilisation. In sectors involving heavy transportation — such as aviation, road freight, and shipping — kerosene, diesel, and heavy fuel oil (HFO) are predominantly employed, collectively accounting for more than 50% of GHG emissions from the transport sector. However, these applications are categorised as “difficult-to-decarbonise", since electrification and hydrogen are unlikely to supplant these fuels in the near future. Nevertheless, it is imperative to adopt low-carbon alternative fuels to steadily replace fossil fuels.

Reducing the negative environmental impacts of fossil fuel utilisation necessitates a transition towards more sustainable energy carriers. The automotive sector, particularly road (passenger) transportation, is rapidly progressing away from fossil fuels. The “difficult-to-decarbonise” transportation applications will need to depend on liquid fuels for several decades to come; however, these fuels must become significantly cleaner and greener. Liquid fuels are characterised by property specifications rather than composition. As substitutes for fossil-derived kerosene, diesel, and HFO, theoretically, any feedstock-technology combination can be employed to produce sustainable alternative fuels, provided that the fuels comply with sustainability criteria and meet technical property specifications.

Formulation of next-generation sustainable fuels can be significantly advanced through the application of artificial intelligence, which possesses the capability to rapidly identify optimal fuel compositions. DeepFuel has been developed to facilitate intelligent fuel formulation. By mapping molecules with properties relevant to fuel blends for engines, it expedites the design of high-performance, low-carbon fuels, including sustainable aviation fuel (SAF) for the aviation sector and diesel alternatives for road freight and maritime shipping.

DeepFuel-kit empowers researchers and innovators to utilise advanced AI, molecular feature extraction, and predictive modelling tools, thereby promoting the efficient, insightful, and environmentally sustainable design of next-generation fuels. The application of DeepFuel includes (but not limited to):

1. “Drop-in” alternative fuel & fuel additive design
• Formulation of “drop-in” fuel mixtures (such as SAF): This involves designing fuels that meet performance and regulatory standards and are fully compatible with existing infrastructure and engines without requiring modifications. Using components produced from renewable feedstocks, the designed fuel blends reduce lifecycle greenhouse gas (GHG) emissions. It might involve selecting or developing new fuel components or molecules.
• Development of fuel additives for blends with user-defined properties: This involves selection or development of new fuel components or molecules, and adding them to existing fuels (including conventional fossil fuels), achieving specific outcomes such as targeted physicochemical properties (e.g. lower freezing point & viscosity) for enhanced performance.
2. Low-carbon multi-fuel design
• Formulation of low-carbon fuel mixtures based on existing low-zero carbon fuels: This involves combining existing low or zero-carbon fuels (e.g. ammonia, biodiesel, dimethyl ether, methanol), to produce a final multi-fuel product that offers optimal burning and reduces lifecycle GHG emissions without requiring new components/molecules. New technologies may be needed to burn the formulated multi-fuel.

Requirements & installation

DeepFuel-kit runs on:

• Python >=3.11
• NumPy >=1.24
• scikit-learn >=1.4
• ... >=1.4

Developers