Heterogeneous continuous flow hydrogenation is important in the chemical industry, yet the simultaneous optimization of yield and productivity has historically been complex. This study introduces a heterogeneous continuous flow hydrogenation system specifically designed for preparing 2-amino-3-methylbenzoic acid (AMA), employing FTIR inline analysis coupled with an artificial neural network model for monitoring. We explored two distinct reaction optimization strategies: multi-objective Bayesian optimization (MOBO) and mechanism-based intrinsic kinetic modeling, executed in parallel to optimize reaction conditions. Remarkably, the MOBO approach achieved an optimal AMA yield (99%) and productivity (0.64 g/hour) within a limited number of iterations. In comparison, despite requiring extensive experimental data collection and equation fitting, the intrinsic kinetic modeling approach yielded a similar optimal result. Thus, while MOBO offers a more efficient route with fewer required experiments, kinetic modeling provides deeper insights into the optimization landscape but may be impacted by non-chemical kinetic phenomena and requires significant time and resources.

中文翻译：

---

使用 FTIR 在线分析优化非均相连续流加氢：多目标贝叶斯优化和动力学建模的比较研究


非均相连续流加氢在化工行业中很重要，但同时优化产量和生产率历来很复杂。本研究介绍了一种专为制备 2-氨基-3-甲基苯甲酸 （AMA） 而设计的非均相连续流加氢系统，采用 FTIR 在线分析与人工神经网络模型相结合进行监测。我们探索了两种不同的反应优化策略：多目标贝叶斯优化 （MOBO） 和基于机理的内禀动力学建模，并行执行以优化反应条件。值得注意的是，MOBO 方法在有限的迭代次数内实现了最佳的 AMA 产量 （99%） 和生产率 （0.64 g/h）。相比之下，尽管需要大量的实验数据收集和方程拟合，但本征动力学建模方法产生了类似的最佳结果。因此，虽然 MOBO 提供了一种更高效的路线，所需的实验更少，但动力学建模提供了对优化前景的更深入见解，但可能会受到非化学动力学现象的影响，并且需要大量时间和资源。