Predicting rare earth elements concentration in coal ashes with multi-task neural networks.

Area of Science:

• Materials Science
• Environmental Science
• Data Science

Background:

• Growing demand for rare earth elements (REEs) necessitates sustainable sourcing strategies.
• Coal ash presents a potential secondary source for REEs, but efficient extraction is hindered by costly and time-consuming analysis.
• Accurate and rapid screening methods are crucial for identifying REE-rich coal ash feedstocks.

Purpose of the Study:

• To develop a machine learning model for predicting REE content in coal ash using easily measurable bulk composition.
• To enhance the efficiency and reduce the cost of identifying coal ash sources suitable for REE recovery.
• To investigate the use of multi-task learning and transfer learning for improved model performance and adaptability.

Main Methods:

• Development of a multi-task neural network to predict concentrations of various REEs simultaneously.
• Utilizing bulk composition data of coal ashes as input features for the machine learning model.
• Application of transfer learning to adapt the model for coal ashes from different sources.

Main Results:

• The multi-task neural network demonstrated improved accuracy and noise reduction compared to single-task models.
• Key data patterns were identified for effectively screening coal ashes with high REE concentrations.
• Transfer learning successfully improved the model's adaptability to diverse coal ash samples.

Conclusions:

• Machine learning offers a viable and efficient solution for predicting REE content in coal ash.
• The proposed model facilitates rapid identification of promising coal ash sources for sustainable REE extraction.
• This approach supports the development of alternative REE supply chains, reducing reliance on conventional mining.