In order to study the frequency domain calculation method of non-Gaussian excitation, Gaussian mixture model（GMM） is introduced. The measured non-Gaussian excitation is transformed into probabilistic power spectrum（PPSD） through GMM model, so that the non-Gaussian excitation is introduced into the frequency domain. The corresponding non-Gaussian rain flow distribution is obtained by simulation and GMM-Dirlik model, and the damage of the specimen is calculated. At the same time, according to the Gaussian hypothesis, the excitation is directly converted into PSD for damage calculation. Then, the bench test is carried out to obtain the stress response of the specimen, the rain flow distribution in the stress range is obtained by rain flow counting, and the damage of the specimen is calculated. After comparing the test results with the simulation results, it is found that the relative error between the results obtained by GMM-Dirlik model and the test is 10.8%, while the relative error between the results obtained by Gaussian hypothesis and the test is large, which is 45.3%, which further explains the risk of non-Gaussian excitation damage evaluated by Gaussian hypothesis. Finally, the difference between non-Gaussian excitation and Gaussian distribution probability density function is compared to explain the concave phenomenon of measured stress rain flow distribution at the middle stress level and the reason why the damage of non-Gaussian excitation is larger than that of Gaussian excitation.