Process Condition Optimization for Li3PO4–Li3BO3–Li2SO4 Lithium-Ion Conductors Using Gaussian Process and Genetic Algorithm

This paper reports on a method of material development with multiple recommendations using a Gaussian process and a genetic algorithm(GA). Conventional Bayesian optimization(BO) was an inefficient method because it updates the predictive model sequentially with one recommendation and one experiment, furthermore there was a possibility of reaching local solutions in the huge search space. Therefore, we devised an evaluation index based on GA, which is different from the evaluation function used in BO, and developed a multiple recommendation method to efficiently visualize and grasp the enormous material search space. Multiple recommendations were able to efficiently explore a huge search space, and furthermore, deviant experimental prosses conditions were constrained by introducing a prior knowledge model. We found the optimum composition and sintering temperature of a Li3PO4–Li3BO3–Li2SO4 ternary mixture system to be 22:16:62 (mol%) and 723°C, respectively, and Lithium-ion conductivity is measured to be 1.3 × 10−3 S/cm at 300°C, which is more than twice the maximum conductivity (4.9 × 10−4 S/cm) observed in a previous report.