Machine learning assisted optimization of synthesis parameters for Ni-rich cathode materials

Abstract: Optimizing synthesis parameters is a key to success for designing ideal Ni-rich cathode materials with satisfied principal electrochemical specifications. We herein implement machine learning algorithms on 330 experimental datasets, obtained from the controlled environment for reliability, to construct the prediction model. First, correlation values exhibit that the calcination temperature and the size of the particles are determining factors for achieving a long cycle life. Then, the accuracy of seven different machine learning algorithms for predicting the initial capacity, the capacity retention rate, and the amount of residual Li is compared. Remarkable predictive capability is obtained with the average value of coefficient of determinant, R2=0.833, from the extremely randomized tree with adaptive boosting algorithm. Furthermore, we propose the reverse engineering to search for experimental parameters satisfying the target specification. The current result demonstrates that the machine learning has a great potential to accelerate the optimization process for commercialization of cathode materials.