This study presents a machine learning-assisted Surrogate-Particle Swarm Optimization (SOO) framework to optimize the uncatalyzed esterification of acetic anhydride with isoamyl alcohol in a microreactor for energy-efficient process intensification. A multiple-input multiple-output (MIMO) surrogate model, using temperature, flow rate, and retention time as inputs, predicts isoamyl acetate concentration, residual alcohol, and anhydride concentration as outputs. Various machine learning models, including ANN, SVM, and decision trees, were evaluated, with bootstrap resampling to enhance robustness. The best-performing model, i.e. ANN, was integrated with single-objective PSO (SOO) to optimize ester formation while minimizing energy consumption and unreacted alcohol. Additionally, multi-objective PSO (MOPSO) explored trade-offs between ester maximization and energy minimization. Results highlight that energy minimization occurs at low flow rates and temperatures (65.0 °C, 7.0 µL/min, 22.5 min), while ester maximization requires higher flow rates and longer residence times (72.36 °C, 10.53 µL/min, 100 min). The findings offer significant insights for improving microreactor design and operational efficiency in esterification reactions.
Rohman et al. (Mon,) studied this question.