Machine-learned digital phase switch for sustainable chemical production

Batch or semi-batch chemical reaction units often requires multiple operational phases to convert reactants valuable products. In various production facilities, the switching decision of such has be confirmed and registered by operating personnel. Imprecise can waste a significant amount time energy for unit, which gives negative plant sustainability costs. Additionally, automation phase is rarely used due challenges batch-to-batch variance, sensor instability, process uncertainties. Here, we demonstrate that using machine learning approach includes optimized noise removal methods neural network (that was architecture searched), real-time completion could precisely tracked (R 2 > 0.98). Furthermore, show latent space evolved transferred from predicting classifying via optimal transfer learning. From optimally learned network, novel switch index proposed act as digital shown capable reducing total reactor operation time, with verification an operator. These intelligent analytics studied on reactive distillation unit monomers acids polyester in Netherlands. The combined gave potential 5.4% batch saving, 10.6% savings, 10.5% carbon emissions reduction. For operator, this method also saves up 6 hours during end discharge reaction. • Explainable track multi-step systems. Evolving Hampel filter provide precise tracking Optimal provides system. system improved consumption, emissions. Prediction stabilizes after 800 mins operator monitoring time.