Abstract: The cutting force and tool wear were measured through the zirconia turning experiment， and the ratio of workpiece material removal to tool wear was as a quantitative index of tool utilization. The single-factor experimental values were trained and predicted by BP neural network that was improved by particle swarm optimization (PSO).The one-dimensional models describing the relationship of tool utilization/cutting force and the geometric parameters of each tool were established by least-squares fitting， and the reliability of the models was tested by the correlation coefficient. The multivariate models based on the one-dimensional models are proposed too. The multivariate models were solved by PSO combined with orthogonal experimental values， and was proved to be more accurate through experiments. Taking the multivariate models as the objective function and the maximum tool utilization/minimum cutting force as the optimization goals， the tool geometry parameters were optimized by PSO， and the experiments show that the optimized tool geometry parameters are reasonable.

Key words: zirconia ceramic, tool geometry parameters, numerical fitting, multi-objective optimization, turning