Lignin Biorefinery Design and Optimization using Machine Learning

A novel lignin biorefinery concept developed at Aalto university [1] facilitates high-value lignin extraction from biomass with an inexpensive and green hydrothermal treatment followed by aqueous acetone extraction. The extraction process is flexible and capable of producing materials with varied chemical composition, structure, and properties. To take full advantage of the flexibility, the lignin extraction process should be optimized to identify the best combination of adjustable experimental settings for a given application. This can, however, be a time-consuming endeavor involving a significant number of experiments. In this presentation, I demonstrate how we can efficiently optimize this new lignin biorefinery with machine learning. Our approach employs Bayesian optimization to guide the data collection process so that predictive models can be established from as few experiments as possible. Our results show that with less than 20 experiments, we can find the process settings that maximize lignin yield or the β-O-4 content. We also obtain intuitive maps that correlate the process variables with the yield and β-O-4 content over the entire range of feasible experiment settings. We are now investigating multi-objective optimization techniques to determine optimal processing conditions for given applications.[1] T. V. Lourencon, L. G. Greca, D. Tarasov, M. Borrega, T. Tamminen, O. J. Rojas, and M. Y. Balakshin, ACS Sustainable Chem. Eng. 8, 1230 (2020).