关键词: 仿真实验, 全模型辅助法, 贝叶斯, 信号期望

Abstract:

Aiming at the problem in the traditional Bayesian-based full-model assisted probability of detection （FMA-POD） assessment method where the posterior POD model variance is distorted due to mismatches between computer simulations and the physical process of non-destructive detection, a Bayesian FMA-POD assessment method is proposed based on signal expectation. This method selects the variance of the test signal as the POD model variance and focuses the assessment object on the signal expectation. It optimizes the traditional Bayesian method’s problem of estimating two POD model parameters into a Bayesian assessment problem for the signal expectation with a known model variance. This avoids the distortion risk inherent in traditional methods where data fusion contaminates the signal variance. Consequently, a Bayesian FMA-POD assessment model based on signal expectation is constructed. To verify the method’s effectiveness, case studies on eddy current detection of defects in conductive plates and ultrasonic detection of defects in stainless steel plates are conducted. The results indicate that for the two detection cases, the proposed method achieves a₉₀, 90% probability of detection for defect size in the POD curve, with deviations of 0.37% and 1.21% relative to the experimental data, respectively, while narrowing the confidence interval width, satisfying the conservatism strategy, and avoiding over-conservatism. These deviations are significantly lower than the 2.23% and 15.3% observed with the traditional method. Moreover, compared to the traditional method, the proposed method demonstrates superior capability in maintaining the stability of a₉₀, the key characteristic parameter of the POD curve, and effectively prevents the occurrence of distortion phenomena.

Key words: simulation experiment, full-model assisted (FMA), Bayesian, signal expectation