Artificial Intelligence Approaches

Using next-generation artificial intelligence (AI) approaches—such as deep learning and machine learning (ML) as well as robust AI tools—NREL accelerates data-mining for bioenergy and bioeconomy research.

The data is mined from:

• Research databases (biological, chemical, polymer, and physical)
• Multivariate spectroscopic, spectrometric, and imaging data
• Real-time measures of titers, rates, and yields in bioreactors.

Machine Learning

NREL's AI/ML capabilities are applied across:

Flagship Project: Artificial Intelligence-Enabled Hyperspectral Imaging for Municipal Solid Waste Analysis

Currently, only 38% of municipal solid waste (MSW) is recycled, and 12% is used for energy recovery; whereas 50% of MSW ends up in landfills. This abundant waste stream is restricted as a feedstock for fuel production because of its heterogeneity and the lack of fast and robust sensing technology capable of detecting and characterizing MSW components accurately and efficiently.

To address this bottleneck, NREL researchers, together with North Carolina State University, are developing and demonstrating a fully functional smart MSW management system. The system combines spectroscopy, computer vision, and AI that enables spectroscopy/object recognition-based technology for rapid identification and characterization of organic fractions (food, plastic, paper and paperboard, rubber, leather, and textile) of MSW in real time. This is accomplished by developing and training deep learning neural networks to perform chemical-based rapid identification and characterization through the combined use of hyperspectral imaging and computer vision of organic fractions.

The technology enables high-throughput characterization of low-cost heterogenous MSW to produce conversion-ready feedstock for biofuels and bioproducts.

Significance and Impacts

Combining advanced analytical and imaging tools, computer vision, and AI, NREL enables the prediction of new products with new properties and process optimization for a broad range of domestic energy technology applications. Our work in deep learning and AI significantly reduces time spent on bench-scale research, eliminating many barriers to bringing products and energy technologies to market.

Contact