<scp>BayesCAMPD</scp> : Data‐efficient and closed‐loop integrated molecular and process design using Bayesian optimization

Abstract Data‐driven techniques leverage surrogate models to enable efficient computer‐aided molecular and process design (CAMPD). However, accurately modeling complex systems across a big design space often requires substantial data. To reduce data demand and improve design efficiency, the BayesCAMPD approach is proposed for the integrated design of molecules and processes using Bayesian optimization. It provides a data‐efficient and closed‐loop solution to data‐driven CAMPD through iterative data‐driven modeling, model‐based optimization, and validation of the solutions obtained. Based on limited data, BayesCAMPD systematically identifies and validates promising molecular and process solutions, finally converging to an optimal design. Through its application to the integrated design of solvents and extractive distillation processes, the proposed BayesCAMPD approach is demonstrated to be practically relevant and very efficient. Although BayesCAMPD incurs increased computational costs due to model updating and sequential optimization, it significantly reduces the need for large training datasets, offering a highly efficient solution for data‐driven CAMPD tasks.