Machine Learning Predicts Conversion and Molecular Weight Distributions in Computer Controlled Polymerization

Abstract

The skew and shape of the Molecular Weight Distribution (MWD) of polymers have a significant impact on polymer physical properties. Standard summary metrics statistically derived from the MWD only provide an incomplete picture of the polymer MWD. Machine learning (ML) methods coupled with high-throughput experimentation (HTE) could potentially allow for the prediction of the entire polymer MWD without information loss. In our work, we present a computer controlled HTE platform that is able to run up to 8 unique variable conditions for the free radical polymerizations of styrene. The segmented-flow HTE system was equipped with an inline Raman spectrometer and offline size exclusion chromatography (SEC) to obtain time dependent conversion and MWD respectively. In a supervised learning exercise, the gradient-boosted decision trees ML algorithm was used to predict monomer conversion from reagent concentrations and reaction time with high accuracy, using data obtained from the Raman spectra. A second ML algorithm uses the random forest regressor to predict entire MWDs of the synthesized polymer. Each algorithm can navigate the complexity of multiple parallel reactions occurring in a polymerization. The algorithm accurately predicts monomer conversion in the first case despite variations in the polymerization kinetic parameters over time. In the second case, we predict the polymer MWD where fine details such as the shape and skew of the MWD are captured without information loss. SHAP values were calculated to examine the dependence of the ML model output on the experimental conditions. A transfer learning approach was also used to enhance the predictive power of a Deep Neural Network (DNN) model in predicting batch polymerization MWDs. Overall, we demonstrate that the combination of HTE and ML provide a high level of predictive accuracy in determining polymerization outcomes, providing polymer chemists with the ability to target the synthesis of polymers with desired properties via the predicted conversion and MWD.

An Object-Oriented Framework to Enable Workflow Evolution across Materials Acceleration Platforms

“…In this case study, we demonstrate our evolved workflow to create a model that predicts Molecular Weight Distribution (MWD) of polystyrene for a given set of input reaction parameters such as monomer and initiator concentrations 22 . Figure 2 shows our workflow mapping from the initial state represented by the precursors class to the output state which is a prediction of polystyrene MWD.…”

Data-Driven Prediction of Janus/Core–Shell Morphology in Polymer Particles: A Machine-Learning Approach

“…Besides answering the critical question of how different features affect the predicted target, XAI also makes it clear whether we can trust the prediction results or not. Shapley’s values are based on the game theory, which mathematically discusses how players’ actions in an interactive decision-making game influence the outcome. , In polymer informatics, Shapley additive explainers have been shown to interpret the factors affecting molecular weight distribution (MWD) in computer-controlled polymerization . Da Tan and co-workers successfully applied ensemble ML classifiers in combination with XAI analysis to predict a polymer’s MWD diagram, taking into account only the reaction parameters .…”

“…Shapley’s values are based on the game theory, which mathematically discusses how players’ actions in an interactive decision-making game influence the outcome. , In polymer informatics, Shapley additive explainers have been shown to interpret the factors affecting molecular weight distribution (MWD) in computer-controlled polymerization . Da Tan and co-workers successfully applied ensemble ML classifiers in combination with XAI analysis to predict a polymer’s MWD diagram, taking into account only the reaction parameters . Also, in industrial processes or production lines, the combined application of Shapley classifiers and ML algorithms has shown improved reliability and interpretability. , Liang et al applied XAI analysis to uncover the factors affecting the ML-assisted discovery of low dielectric-constant polymers …”

“…36,37 In polymer informatics, Shapley additive explainers have been shown to interpret the factors affecting molecular weight distribution (MWD) in computer-controlled polymerization. 38 Da Tan and co-workers successfully applied ensemble ML classifiers in combination with XAI analysis to predict a polymer's MWD diagram, taking into account only the reaction parameters. 38 Also, in industrial processes or production lines, the combined application of Shapley classifiers and ML algorithms has shown improved reliability and interpretability.…”

“…38 Da Tan and co-workers successfully applied ensemble ML classifiers in combination with XAI analysis to predict a polymer's MWD diagram, taking into account only the reaction parameters. 38 Also, in industrial processes or production lines, the combined application of Shapley classifiers and ML algorithms has shown improved reliability and interpretability. 39,40 Liang et al applied XAI analysis to uncover the factors affecting the ML-assisted discovery of low dielectric-constant polymers.…”