Flame-assisted spray pyrolysis to size-controlled LiyAlxMn2−xO4: a supervised machine learning approach

Abstract

The demand for lithium manganese oxides for cathodes is growing faster than that for lithium cobalt oxides because of their smaller environmental footprint, lower cost, and flammability. Their capacity, cyclability, and time between charges require a precise control of the physiochemical properties including the particle size distribution. We developed a novel one-step synthesis route for Al-doped LiMn₂O₄ by flame assisted spray pyrolysis (FASP) that controls the particle size and morphology. A combination of intrinsic and extrinsic parameters governs the process with synergistic and antagonistic interactions and non-linear relationships. We identified the responses to 10 operational factors that affect the size and specific surface area of Li_yAl_xMn_2−xO₄ powders. The surface areas of powders doped with Al are doubled versus standard formulations, which we attribute to carbon formation on the surface. Elemental mapping of the surface shows that Al, Mn, and O are evenly distributed in the particles. The shifts in the XRD peaks prove that Al is incorporated into the structure. We trained three machine learning algorithms to predict the particle size given the experimental factors. Artificial neural networks (ANNs), support vector regression (SVR), and random forest regression (RF) adequately predict the particle size and each model accounts for 0.97, 0.86, and 0.49 of the variance (R-squared), respectively. The ANN model identifies the ethanol to precursor ratio, precursor type, and fuel flow-rate as the most significant factors to account for the particle size. The RF model assigns the greatest effect to the vacuum pressure, followed by the precursor type and fuel flow-rate.