Researchers at the Yale School of the Environment published a study “Machine learning for sustainable development and applications of biomass and biomass-derived carbonaceous materials in water and agricultural systems: A review” in Resources, Conservation, and Recycling that analyzed current machine learning applications for biomass and biomass-derived materials (BDM).

The goal was to determine if machine learning is advancing the research and development of biomass products. The study found that machine learning has not been applied across the entire life cycle of BDM, limiting its ability for development.

“Biomass-derived materials (BDM) have broad applications in water and agricultural systems. As an emerging tool, machine learning (ML) has been applied to BDM systems to address material, process, and supply chain design challenges. This paper reviewed 53 papers published since 2008 to understand the capabilities, current limitations, and future potentials of ML in supporting sustainable development and applications of BDM,” write the investigators.

Identifying critical factors for optimizing BDM

“Previous ML applications were classified into three categories based on their objectives–material and process design, end-use performance prediction, and sustainability assessment. These ML applications focus on identifying critical factors for optimizing BDM systems, predicting material features and performances, reverse engineering, and addressing data challenges for sustainability assessments.

“BDM datasets show large variations, and ∼75% of them possess < 600 data points. Ensemble models and state-of-the-art neural networks (NNs) perform and generalize well on such datasets. Limitations for scaling up ML for BDM systems lie in the low interpretability of the ensemble and NN models and the lack of studies in sustainability assessment that consider geo-temporal dynamics.

“A workflow is recommended for future ML studies for BDM systems. More research is needed to explore ML applications for sustainable development, assessment, and optimization of BDM systems.”

Biomass is widely considered a renewable alternative to fossil fuels, and many experts say it can play a critical  role in combating climate change. Biomass stores carbon and can be turned into bio-based products and energy that can be used to improve soil, treat wastewater, and produce renewable feedstock.

Limited large-scale production

Yet large-scale production of it has been limited due to economic constraints and challenges to optimizing and controlling biomass conversion.

“There are so many combinations of biomass feedstock, conversion technologies, and BDM applications. If we want to try each combination using the traditional trial-and-error experimental approach, this will take a lot of time, labor, effort, and energy. We already generate a lot of data from these past experiments, so we are asking, can we apply machine learning to help us to figure out how we can better design BDM,” explains Yuan Yao, PhD, assistant professor of industrial ecology and sustainable systems.

Doctoral student Hannah Szu-Han Wang notes that the study has led to further research on data gaps in machine learning on biomass-derived materials.

The two researchers said they were interested in pursuing this study to find out if machine learning could help with best practices for creating BDM, a chief component of a bio-based economy, as well as predicting their performance as sustainable materials.

“We found a future direction that people have not yet explored in terms of sustainability assessments for BDM,” she says. “There needs to be a full pathway prediction to enhance our understanding of how various factors regarding BDM interact and contribute to sustainability.”

Previous articleModeling Uncertainty in Process Development
Next articleT Cells Could Represent Therapeutic Target for Alzheimer’s Disease