Abstract: The coarseness of crystalline grains in water turbine material decreases the UT signal SNR, which leads to undetected fault or misjudgment in actual UT testing. To solve this problem, UT signals was collected from a test block which is made of the same material as water turbine rotor. After the signals were processed by EMD and Hilbert transform, IMFs that include information of defects were chosen to rebuild signal. SNR was increased up to 2 times. The best sampling frequency, 40 times as the probe center frequency, for real time UT testing was determined considering SNR and calculation quantity.