Digitizing sustainable process development: From ex-post to ex-ante LCA using machine-learning to evaluate bio-based process technologies ahead of detailed design

Abstract

Life Cycle Assessment is a data-intensive process holding great promise to benefit from advanced analytics and machine learning technologies. The present research aims at the development of a data-science based framework with capabilities to estimate LCA metrics of bio-based and biorefinery processes in early design phases. Life cycle inventories may combine experimental (pilot and lab scale) data, property and thermodynamic databases, and model-derived data from simulations and design studies. The framework applies advanced analytics such as classification trees and artificial neural networks (ANN) with a scope to produce input–output relationships through predictor variables that refer to the molecular structure of bio-chemical or bio-fuel products of interest, the feedstocks used, and the process technologies characteristics. The combined use of ANNs and trees demonstrates a coordinated level of complementarity between the approaches, while it improves robustness and streamlines LCA estimations in the early-stage design.