Machine learning for efficient prediction and optimization of platinum group metal recovery from spent automotive catalysts

Abstract

Spent automotive catalysts (SACs) represent a valuable resource for recycling platinum group metals (PGMs). Roasting pretreatment followed by acid leaching is a widely used process for this purpose. However, traditional optimizing methods are often challenged by the variable composition of waste feed and numerous process parameters. This necessitates extensive experiments, leading to increased costs and environmental risks. To overcome these issues, this study employed machine learning (ML) to enhance the recovery of PGMs from SACs. Based on a dataset of 18,877 collected data points, the model included 34 input features covering the waste properties and technological parameters. By comparing four ML algorithms, the extreme gradient boosting demonstrated superior predictive performance for PGMs leaching efficiency across training, testing, and 5-fold cross-validation. Feature importance analysis revealed the influence of the multiple parameters on PGMs leaching behaviors. Additionally, a user-friendly graphical user interface (GUI) was developed to enable the rapid prediction of PGMs leaching results solely by experimentally measuring the particle size and composition of the waste, thereby minimizing the need for repeated experimental optimization. Finally, the practicality and effectiveness of the GUI were confirmed through experimental validation. This study provides an intelligent and efficient strategy for PGMs recovery from SACs, which can enhance both economic efficiency and environmental sustainability.