Machine learning in life cycle assessment for process modeling of emerging technologies

Main Presenter:    Nicolás Martínez Ramón 

Co-Authors:   Diego Iribarren     Javier Dufour                                          

The use of life cycle assessment (LCA) is instrumental in identifying environmental hotspots in intricate systems and guiding the design and selection of environmentally friendly production methods. However, the amount of time, resources, and expertise required to compile an accurate life cycle inventory dataset is among the main concerns for LCA practitioners. This is especially true in the early design phase given the lack of commercial-scale primary plant data, while design choices strongly affect industrial-scale production [1]. Although process simulation is a useful tool for estimating inventory data in this phase, the sole use of first-principle models can sometimes be unfeasible due to high complexity and/or high computational demand.
Surrogate or substitute models aim to capture essential system responses through simplified versions of a model that significantly alleviate computational demands. They can be developed, amongst others, with machine learning (ML) techniques using data from real-world experiments and/or high-fidelity simulations [2].
This study aims to illustrate the possibilities of incorporating ML into the LCA workflow, especially regarding inventory data generation in the prospective scale-up of a process involving emerging technologies. To do so, a first case aimed at abstracting a complete process into a two-layer artificial neural network was proposed. In this case, a set of four gasification simulation models was prepared to be flexible and provide synthetic data by conducting sensitivity analysis for 19 input variables (composition, temperature, pressure…) and 18 responses such as CO2 emissions, electricity consumption, and external heat demand. This simplified version of the model was then used to assess the environmental impact of integrating a gasification unit into a complete waste management system.
Additionally, a second case focused on substituting a complex part of a process for an ML surrogate model. In this case, the pyrolysis reactor was simulated using a three-layer artificial neural network capable of predicting pyrolysis product yields based on data from the literature. This simulation enabled a flexible model (8 input variables and 18 responses) for predicting inventory data such as emissions and utility consumption for a technology that is otherwise complex to model.
Overall, this work paves the way for future studies relying on implementing ML to predict inventory data, consequently facilitating the completion of LCAs, especially when process modeling poses a challenge in terms of complexity or computational load.

©2024 Forum for Sustainability through Life Cycle Innovation e.V. | Contact Us | Legal Info


If you would like to get in touch with us, please feel free to send us a message. Thank you very much in advance.

Subscribe To Our Newsletter
Join our newsletter and stay up-to-date on everything happening in the life cycle community! We'll send you 3-4 newsletters per year
Subscribe Now!

Log in with your credentials


Forgot your details?

Create Account