Machine learning-based discrete element reaction model for predicting the dechlorination of poly (vinyl chloride) in NaOH/ethylene glycol solvent with ball milling

The thermal treatment of poly (vinyl chloride) (PVC) during plastic waste management can result in undesirable chlorine-based compounds. Dechlorination (de-Cl) of PVC waste by ball milling in NaOH/ethylene glycol solvent can be an effective method for recycling chlorine and valorizing the hydrocarbons present. The de-Cl behavior versus reaction time can be well fitted by a shrinking core model for a single treatment under certain conditions. However, the change of the fitted kinetic parameters have not clear law under various mechanical conditions so that the reaction cannot be predicted, especially for the PVC particles in heterogeneous shape. To optimize the de-Cl process for highly heterogeneous waste with a complex composition, we developed a novel discrete element reaction model based on machine learning for predicting the de-Cl behavior of PVC. First, the fundamental experiments for generating the training and validation data resulted in up to 99% of de-Cl degree for PVC pellets with 300 1.27-cm balls at 30 rpm. The model can make predictions regarding the de-Cl reaction based on the ball-to-sample impact energy. The model parameters were successfully optimized by the training data, and the model predictions during verification were in keeping with the experimental data, especially for the high prediction accuracy fixing the rotation speed at 30 rpm. The model suggested that the generation of additional reactive area by sufficient ball-to-sample impact energy (>0.5 × 10⁻¹ J/s in this study) is vital for the enhancement of the de-Cl efficiency. Thus, the proposed method should be suitable for integration with industrial de-Cl processes.