High performance fuel design is key to creating cleaner, more efficient combustion engine systems. To increase efficiency and reduce carbon emissions when designing liquid fuels for combustion engine applications, they present a data-driven artificial intelligence (AI) framework. The fuel design approach is a limited optimization challenge that combines two components:
- A deep learning (DL) model for predicting attributes of single components and mixtures.
- Search algorithms for moving quickly through chemical space.
Their method integrates the mixture operator (MO) into the network architecture and provides the hidden vector of the mixture as a linear combination of the vectors of each individual component in each mixture.
They show that the DL model predicts pure component attributes with an accuracy comparable to competing computational techniques, while the search tool can produce a variety of candidate fuel combinations. The integrated framework was evaluated to demonstrate the development of a high octane, low smoke fuel that meets gasoline specification requirements. Using an AI fuel design approach, fuel compositions can be developed quickly to increase engine efficiency and reduce emissions.
Most of the increase in global temperatures can be attributed to greenhouse gas emissions. The combustion of hydrocarbons, such as gasoline, which power most vehicle engines, is a major source of CO2 emissions. Designing transportation fuels with greater efficiency and lower carbon emissions is a viable response to these environmental issues.
Many fuel screening methods have been developed; however, they are usually only proven on smaller mixtures or require additional pre-treatment, making these combinations unsuitable for reverse fuel design. According to the research group, “the main bottleneck is the screening of large mixtures involving hundreds of components to predict the synergistic and antagonistic effects of species on the resulting combination attributes.”
To filter efficiently, the researcher built a deep learning model composed of many smaller networks dedicated to particular tasks. According to one researcher, “this problem was perfectly suited to deep learning, which allows capturing non-linear interactions between species”. The researchers used the reverse design method to identify possible fuels by first defining combustion-related characteristics, such as fuel ignition quality and propensity for soot formation.
The researchers created a large database to train the model using experimental measurements from the literature. The database included all kinds of pure substances, mixtures of alternative fuels and complex mixtures, such as gasoline.
The researchers had to incorporate vector representations into the model because no model could be modified for reverse fuel design. They created a mixture operator that directly connects the hidden terms of pure compounds and mixes them by linear combinations. This mixing operator is inspired by word processing methods that use hidden vectors to connect words to sentences. They also included search algorithms to find fuel mixtures in chemical space that match predefined parameters.
The model correctly predicted the ignition quality of the fuel and the propensity for soot formation of different molecules and mixtures. Additionally, he found several gasoline blends that met predetermined standards.
This Article is written as a research summary article by Marktechpost Staff based on the research paper 'Artificial intelligence-driven design of fuel mixtures'. All Credit For This Research Goes To Researchers on This Project. Check out the paper and reference article.
Please Don't Forget To Join Our ML Subreddit
Ashish Kumar is an intern consultant at MarktechPost. He is currently pursuing his Btech from Indian Institute of Technology (IIT), Kanpur. He is passionate about exploring new technological advances and applying them to real life.