压气机叶片加工几何不确定性问题十分突出，作为不确定性量化系统的输入，准确表达其统计分布对系统输出尤为重要。首先，针对某100个压气机转子叶片同一叶高截面前缘半径误差，进行统计分析发现其分布呈现左偏特征，且为尖峰态，非正态性显著；其次，为补充正态分布的局限性，使用条件期望最大化的方法获得具有偏峰特征的误差分布，并与正态分布对比发现偏峰分布拟合程度更佳；再次，将服从2种分布的前缘半径误差分别作为不确定性量化输入，以叶栅总压损失系数和静压比作为响应，对比量化结果发现，不同分布对应的总压损失系数和静压比均值的差异不到1%，可考虑忽略，而分散度差异显著，且总压损失系数分散度变化量较静压比更大，最大约20%；最后，若以正态分布作为输入，会显著高估前缘半径误差不确定性对叶栅气动性能分散度的影响与性能变化范围，而低估性能恶化的可能性。研究说明了“考虑加工误差偏峰特征”的必要性，以期对叶片精细化加工提供更准确的参考。

The issue of geometric uncertainty of compressor blade machining is very prominent. As the input of the uncertainty quantification system， the accurate expression of its statistical distribution is particularly important for the system output. According to the statistical analysis of leading-edge radius errors at the same blade height section of 100 compressor rotor blades， it is found that the distribution is left-skewed and in a peak condition， with significant non-normality. Then， to overcome the limitations of the normal distribution， the expectation conditional maximization either method is used to obtain the error distribution with skewness and kurtosis， which fits the error data better than the normal distribution. Finally， the leading-edge radius error fitting the two distributions are used as the uncertainty quantification input separately， and the total pressure loss coefficient and static pressure ratio of the cascade are used as the response. Comparison of the quantification results shows that the total pressure loss coefficient and static pressure ratio mean variable values，the mean variable values of the total pressure loss coefficient and static pressure ratio corresponding to different statistical distribution are less than 1%， which can be negligible， while the scatter difference is significant. Besides， the variation of the total pressure loss coefficient is larger than that of the static pressure ratio， about 20% at most. In addition， if the normal distribution is input， the influence of the uncertainty of the leading-edge radius error on the scatter of the cascade aerodynamic performance and the performance variation range are significantly overestimated， while the possibility of performance deterioration is underestimated. This research illustrates the necessity of considering skewness and kurtosis characteristics of machining errors to provide more accurate reference for blade fine machining.