Integration of\nNeural Networks and First-Principles\nModel for Optimizing l‑Lactide Branched Polymerization

Addressing the growing demand for sustainable materials,\nthis research\npaves the way for the efficient consumption and sustainable production\nof branched polylactide (PLA). A novel hybrid modeling approach combines\nfirst-principles (FP) model with artificial neural network (ANN) for\nring-opening polymerization (ROP). The hybrid ANN, trained with FP\nmodel data, demonstrated optimal performance with a hidden layer of\n20 neurons, achieving a root mean square error (RMSE) of 0.004 and\na regression coefficient (<i>R</i><sup>2</sup>) of 0.99.\nThe hybrid model accurately predicted key polymer properties, including\naverage molecular weights (Mn and Mw), polydispersity index (PDI),\ndegree of branching (DB), monomer conversion, and polymerization time.\nValidation was performed on various branched PLA compositions (PLLH80,\nPLLH94, and PLLH97). Multiobjective optimization (MOO) using NSGA-II\nshowed strong agreement between FP model and hybrid ANN across six\ncase studies, highlighting their effectiveness in predicting polymerization\noutcomes.