Frontiers in Materials (Feb 2022)
Machine Learning Assisted Design of Experiments for Solid State Electrolyte Lithium Aluminum Titanium Phosphate
Abstract
Lithium-ion batteries with solid electrolytes offer safety, higher energy density and higher long-term performance, which are promising alternatives to conventional liquid electrolyte batteries. Lithium aluminum titanium phosphate (LATP) is one potential solid electrolyte candidate due to its high Li-ion conductivity. To evaluate its performance, influences of the experimental factors on the materials design need to be investigated systematically. In this work, a materials design strategy based on machine learning (ML) is employed to design experimental conditions for the synthesis of LATP. In the variation of parameters, we focus on the tolerance against the possible deviations in the concentration of the precursors, as well as the influence of sintering temperature and holding time. Specifically, models built with different design selection strategies are compared based on the training data assembled from previous laboratory experiments. The best one is then chosen to design new experiment parameters, followed by measuring the corresponding properties of the newly synthesized samples. A previously unknown sample with ionic conductivity of 1.09 × 10−3 S cm−1 is discovered within several iterations. In order to further understand the mechanisms governing the high ionic conductivity of these samples, the resulting phase compositions and crystal structures are studied with X-ray diffraction, while the microstructures of sintered pellets are investigated by scanning electron microscopy. Our studies demonstrate the advantages of applying machine learning in designing experimental conditions by the synthesis of desired materials, which can effectively help researchers to reduce the number of required experiments.
Keywords
