制备晶粒尺寸不同的In718合金试样，采用经验模态分解(EMD)研究超声背散射信号的时频域特征，分析晶粒尺寸对不同频段固有模态函数(IMF)的影响，并探求各IMF信号的功率与晶粒尺寸之间的相关性。结果表明，原始背散射信号和IMF1信号的频带宽，受晶粒尺寸变化影响均不显著；IMF2信号频域分布较为集中，峰值频率的幅值随晶粒的增大而增大，其功率与晶粒尺寸相关系数达0.995，显著高于其它模态，EMD过程实质上滤除了与晶粒尺寸无关的成分。以充分反映晶粒散射强度的IMF2成分作为晶粒尺寸评价的特征信号，建立面向In718晶粒尺寸的超声背散射EMD评价模型。晶粒尺寸实测结果显示，本方法的灵敏度是传统背散射法的3.74倍；对两验证试样的评价误差分别为-3.72%和2.87%，精度显著高于声速法；与衰减法相比，本方法无需厚度信息，评价结果不受测厚误差的影响；此外，相比于金相法具有效率及无需破坏待测件的优势。

Superalloy In718 enjoys wide application in such crucial parts as turbine engine disks due to high strength, great toughness and corrosion resistance in different temperature environment. Since the mechanical properties of superalloy In718 are greatly influenced by the grain size, a nondestructive detection method is studied in order to determine the grain size quickly and effectively. In this research, superalloy In718 test samples of different grain sizes were produced and the empirical mode decomposition (EMD) method was employed to find the characteristics of the time-frequency domain of the ultrasonic backscattering signals. Then the effects of the grain size over the intrinsic mode function (IMF) of different frequency bands was analyzed to seek the relations between the power and the grain size of the IMF signals. The original backscattering signals and IMF1 (the first IMF) signals barely respond to the change of the grain size because of their wide frequency bandwidths; the distribution of the frequency domain of the IMF2 signals is centralized and the amplitude of the peak frequency increases with the grain size, and the correlation coefficient between the power and the grain size is 0.995, much higher than that of other modes. This method eliminates the components irrelative to the grain size and takes the IMF2 components which fully reflect the intensity of the grain scattering as the characteristic signals of the grain size evaluation to build an ultrasonic backscattering EMD model evaluating the grain size of superalloy In718. The actual measurement results of the grain size show that the sensitivity of this method is 3.74 times the traditional backscattering method; the evaluation errors over the two verification test samples are -3.72% and 2.87%, apparently more accurate than the ultrasonic velocity method; compared with the attenuation method, this method requires no information of the thickness so that the evaluation results are independent of the thickness measuring error; compared with the metallographic method, this method is more efficient and requires no damage on the components to be evaluated.