Machine learning-driven prediction and multi-objective optimization of pyrolysis behavior for waste printed circuit boards

Predicting the pyrolysis behavior of waste printed circuit boards (WPCBs) and optimizing its parameters are crucial for establishing large-scale efficient pyrolysis systems for resource recovery. This study focuses on elucidating the pyrolysis mechanism, kinetics, and thermogravimetric analysis of WPCBs using model-free approaches and predicting pyrolysis behavior through machine learning (ML) models. The apparent activation energy (E_α) was determined to use the Flynne‒Walle‒Ozawa (FWO), Kissinger‒Akahira‒Sunose (KAS), and Friedman methods, which E_α of 175.29 kJ/mol, 174.38 kJ/mol and 170.67 kJ/mol for the three methods, respectively. The master plot method indicated that the pyrolysis reaction mechanism conforms to the chemical reaction model F3. The mass loss prediction model was developed based on four ML models, with the artificial neural network (ANN) model demonstrating superior performance (R² = 1, MSE = 24574 × 10⁻⁴). Multi-objective optimization based on the ANN model revealed that 500 °C and a heating rate of 10 °C/min represent the optimal operating parameters, simultaneously maximizing energy efficiency and minimizing environmental impact. The findings of this study provide fundamental and practical insights into the optimized pyrolysis of WPCBs.