Explore how AI meets polymer chemistry... hands on.
Two interactive demos showcasing informed neural networks and reinforcement learning for reactor control.
Free radical polymerization of MMA in solution, modeled in three ways. The mechanistic model uses first-principles kinetic equations (mass and energy balances) but has imperfect rate constants. The pure neural network was trained on data within a limited region and fits well there, but has no knowledge of the underlying chemistry. The PCINN (polymer chemistry-informed neural network) combines both: it learns from data while respecting the kinetic equations, staying reliable even outside the training region. Push the sliders beyond the highlighted training zone and watch what happens.
You're running a semibatch free radical polymerization of MMA. Your job: control the monomer feed rate over time to reach the target conversion and the target number-average molecular weight (Mn) simultaneously. Feed too fast? You'll overshoot Mn. Too slow? You won't reach the target conversion in time. Design your feed profile below, then hit "Run My Strategy." When you're ready, let the RL agent show you how it's done.
Click and drag the control points to shape the monomer feed rate over time. The total monomer added is shown below.