Periodic Reporting for period 2 - SmartOptoelectronics (Machine-learning-guided design of perovskite lanthanum oxide cathodes for solid oxide fuel cells)
Reporting period: 2023-06-01 to 2024-05-31
Microscale thin-film-based solid oxide fuel cells (μSOFCs) are an emerging alternative for portable power supply due to their high efficiency, fuel flexibility and high volumetric and specific power densities. Promising cathode materials for μSOFCs are perovskite lanthanum-based oxide materials which have improved oxygen transport properties and resistance to the high operating temperatures. However, the physicochemical factors influencing the performance of these materials are yet to be well understood. The SmartOptoelectronics project will develop machine learning (ML) methods to establish trends between the structural properties and the electrochemical performance of perovskite lanthanum-based oxides based on high-throughput experimental data. These techniques will be used to explore the chemical space of lanthanum-based oxides with the goal of undestanding and designing lanthanum-based materials with enhanced performances for μSOFC applications. Machine learning methods will be validated in
three main steps: (1) deriving structure-property relationships in lanthanum-based oxides from spectroelectrochemical data of combinatorial ternary and quaternary maps; (2) demonstrating new lanthanum-based oxides with enhanced electrochemical properties and performance; (3) optimising the operation of devices based on top-performing materials with operando monitoring of spectroelectrochemical properties.
three main steps: (1) deriving structure-property relationships in lanthanum-based oxides from spectroelectrochemical data of combinatorial ternary and quaternary maps; (2) demonstrating new lanthanum-based oxides with enhanced electrochemical properties and performance; (3) optimising the operation of devices based on top-performing materials with operando monitoring of spectroelectrochemical properties.
Data on La0.8Sr0.2(Mn,Co,Fe)3± with different combinations of Mn, Co and Fe was cleaned and processed. Different parameters were extracted from the data to be used to build machine learning models. Models were then built and investigated using different groups of features and machine
learning techniques. Good performing models have been investigated to identify correlations between variables and the optimal composition. In-depth discussions have been dedicated to elucidate the physicochemical meaning. Experiments and discussions are underway to verify some of the observations.
learning techniques. Good performing models have been investigated to identify correlations between variables and the optimal composition. In-depth discussions have been dedicated to elucidate the physicochemical meaning. Experiments and discussions are underway to verify some of the observations.
Random forests have been found to be the most robust methods to build structure-porperty-performance models. Accuracies over 83% and 90% have been achieved in the modelling of the performance and composition, respectively. This will allow to better understand these materials and will facilitate their characterization in the future.