Abstract: According to the characteristic parameters of ultrasonic sound velocity, attenuation coefficient and nonlinear coefficient, the detection error of grain size of GH4169 is large. The exponential function is used as the mapping function. The ultrasonic parameters are normalized and then merged to construct new characteristic parameters. The optimization problem with the minimum absolute error is optimized, and the genetic algorithm is used to solve the problem, so as to determine the optimal ultrasonic multi-parameter evaluation model, and develop a real-time ultrasonic detection and grain size evaluation combined detection device and verify the experiment. The results show that the constructed ultrasonic multi-parameter evaluation error is small and monotonous; based on the known model, the ultrasonic detection device can extract the feature of the ultrasonic A-scan signal collected in real time and calculate the average grain size of the detected position.