Brightcon 2026, hackathon and courses

Life-cycle assessments are rarely applied as often to the operational decision-making in production as to the planning stage, despite the significant cumulative environmental impacts of high-volume processes. The bottleneck here is that an LCA is too complex for daily use, and decision-makers in production are not trained to interpret the results.

For the ENCIRCLE project, parameterised LCAs are conducted, using as parameters operational variables. The parameters are varied and used to train a surrogate model with the output, i.e. the life-cycle impacts. To determine the stopping criterion for the calculations, the CO2e footprint from the LCA calculations is compared with the optimisation potential derived from more accurate LCI results.

The approach is being tested in two industrial use cases: galvanising and aluminium recycling. As the surrogate model's output is integrated into an Reinforcement Learning AI agent as an optimization objective, it must aggregate all impacts into a single normalized value to form its reward function, a requirement present both in the simulation environment where the agent trains, and the live production line in which it will imminently be deployed.
To this end, the values in the damage categories of ReCiPe 2016 are normalised so that a standard use case corresponds to 100%.

An interactive notebook is used to demonstrate how the surrogate model is generated and ported in Brightway 2.5, using the joblib package.