(510b) A Machine Learning and Data Driven Holistic LCA Approach for Early-Stage Sustainability Assessment

Global warming has intensified in recent years, resulting in serious environmental impacts such as extreme weather events, rising sea levels, biodiversity loss, and an increasing frequency of wildfires[1] . These impacts endanger global sustainability, making it critical to adopt solutions that reduce environmental impacts. To address this issue, a variety of sustainability evaluation tools have been created to evaluate the environmental impact of products and processes. Among these, Life Cycle Assessment (LCA) has emerged as a popular tool providing a systematic assessment of environmental impacts over a product's whole life cycle—from raw material extraction to disposal as well as materials recovery and recycling[2]. However, traditional LCA is primarily retrospective, relying on historical data of the process or product, making it less applicable to emerging technologies still in the development stage. Traditional LCA relies on foreground and background data to conduct assessments. Foreground data includes mass and energy flow measurements specific to the researched process, which are often gathered through direct observations, laboratory experiments, or process models such as Aspen and SuperPro® Designer. Background data provides standardized industry-average environmental profiles. When analyzing innovative processes, foreground data may be available from design specifications, but the absence of corresponding background data in LCA databases creates a significant challenge.

This study aims to develop machine learning (ML) and data analytics approach to assess the environmental impact of a chemical at the early design stage for its entire life cycle. The first phase of this approach focusses on a cradle-to-gate LCA framework, which assesses the environmental impact of a product from the extraction of raw materials to the point when it enters the manufacturing plant to be processed or used. We employed supervised learning models to forecast four key endpoints Life Cycle Impact Assessment (LCIA) metrics: Resource Utilization Impact (RUI), Ecosystem Quality Impact (EQI), Global Warming Potential (GWP), and Human Health Impact (HHI)[3] . These models use Artificial Neural Networks (ANN) and eXtreme Gradient Boosting (XGBoost) to estimate environmental impacts for novel chemicals using their thermodynamic and molecular descriptor properties[4] .

We extend the analysis to the gate-to-gate phase which evaluates the environmental impact during the operational phase when these chemicals are actively used in industrial processes. Various technologies contribute to the total environmental impact of industrial chemical processes, with separation technologies playing a particularly important role due to their high energy consumption and direct impact on decreasing emissions and waste generation. Given their vital function and environmental impact, this study prioritizes a gate-to-gate evaluation of separation technologies, focusing on how different separation methods affect energy consumption, emission levels, and overall sustainability. This study focusses on technology scale-up analysis and impact assessment, which involves moving from lab-scale to industrial-scale production of greener chemicals. We developed a regression model based on a scaling equation to estimate GWP using fundamental process parameters on an industrial level. The model is trained on a dataset obtained from literature, industrial reports, and computer simulations, resulting in accurate predictions across various separation technologies. Case studies in the pharmaceutical and specialty chemicals sectors have been collected from literature for detailed data on materials, emissions, global warming potential, and energy consumption. Additional case studies are being simulated in Aspen and SuperPro® Designer to evaluate their material and energy balance, while the LCA is conducted in SimaPro to obtain the GWP. These datasets will serve as inputs to identify common scalability indices for environmental emissions and support the development of more sustainable industrial processes.

Open-source AI tools such as ChatGPT, Copilot, Gemini, DeepSeek, and others were employed to accelerate data collecting for the gate-to-gate phase analysis. These tools assisted in finding relevant articles, reports, and industrial case studies related to the technology under consideration that contained the required parameters. Each AI source was carefully checked for accuracy.

To demonstrate the practical effectiveness of the ML-driven Data Analytics approach developed, two different case studies will be analyzed: one in pharmaceutical manufacture and the other in organic waste valorization. Our model will predict the GWP for each case, which will then be compared against the actual GWP values determined by the LCA tool, SimaPro.

[1] “Summary for Policymakers.” Accessed: Mar. 24, 2025. [Online]. Available: https://www.ipcc.ch/report/ar6/wg2/chapter/summary-for-policymakers/

[2] R. Hoogmartens, S. Van Passel, K. Van Acker, and M. Dubois, “Bridging the gap between LCA, LCC and CBA as sustainability assessment tools,” Environ. Impact Assess. Rev., vol. 48, pp. 27–33, Sep. 2014, doi: 10.1016/j.eiar.2014.05.001.

[3] K. M. Yenkie, E. A. Aboagye, A. L. Lehr, J. Pazik, J. Longo, and R. P. Hesketh, “Machine Learning enabled Life Cycle Assessment for Early-stage Sustainable Process Design,” in Computer Aided Chemical Engineering, vol. 53, Elsevier, 2024, pp. 2881–2886. doi: 10.1016/B978-0-443-28824-1.50481-6.

[4] E. A. Aboagye et al., “Machine Learning Methods for the Forecasting of Environmental Impacts in Early-stage Process Design,” presented at the Foundations of Computer-Aided Process Design, Breckenridge, Colorado, USA, Jul. 2024, pp. 621–628. doi: 10.69997/sct.141240.