Computing systems that emulate the biological neural networks of animal and human brains can potentially save both money and time as scientists at Los Alamos National Laboratory seek to convert nonfood biomass into new engine fuels. These fuels could be used in existing transportation infrastructure and engine technologies—and meet government regulations. Researchers are using these artificial neural networks to estimate the combustion characteristics of biofuel constituents and new fuel molecules.
“We think this could become a powerful tool to guide the next generation of renewable transportation fuels,” said Andrew Sutton, project leader. The specific network that William Kubic, Jr. of the Los Alamos team developed accounts for the complex, nonlinear interactions among the chemicals, called functional groups, and is applicable to a broad range of hydrocarbons and oxygenated organic compounds. (Functional groups are sets of either atoms or bonds between atoms that determine how molecules behave during chemical reactions.)
Historically, the challenge has been that the chemistry of some of these future fuels wasn’t fully understood, and expensive testing was the only path known for perfecting them. Furthermore, synthesizing these fuels via novel chemical pathways often yields hydrocarbons and oxygenated organic compounds that are not found in current fuel supplies and whose physical and chemical properties have not been measured or published, though their molecular structures are known. However, using the neural network, the team demonstrated that they could effectively predict the contributions of various functional groups to the overall characteristics of the fuels and fuel blends. The new prediction capability provides a workaround to avoid costly and time-consuming physical testing, especially if the chemicals are not available commercially.
The objective of the Los Alamos research was to develop a reliable method for distinguishing the promising fuels from mediocre ones by understanding the physical and chemical properties of fuel components through the details of their molecular structure, a quantitative approach. In this instance, that meant figuring out the connection between how a molecule looks and how efficiently it burns. This quantitative structure−property relationship, or QSPR, is a standard means of determining thermodynamic and transport properties of molecules but it is only now being applied to fuels.
Determining QSPR is a complex task, and so the team pulled in the artificial neural networks. By mimicking biological neural networks, the artificial neural net algorithms learn and progressively improve such things as predictions about the behavior of fuel constituents or blends.
The new method was recently published in the in the journal Industrial and Engineering Chemical Research and could have significant impact on the design of the next generation of biofuels. The team has offered their network tool to other researchers via an easy-to-use, Excel-based implementation that will predict how functional groups contribute to overall fuel characteristics.