基于频率识别纤维增强树脂复合材料加筋板的分层损伤

以纤维增强树脂（FRP）复合材料加筋板为研究对象，通过对比分层损伤发生前后FRP复合材料加筋板的振动频率变化，来识别FRP复合材料加筋板中的分层损伤。构建了人工神经网络（ANN）和基于有代理模型的优化算法（SAO）两种逆向检测算法，利用FRP复合材料加筋板在损伤前后发生的一系列频率变化值来逆推出FRP复合材料加筋板中的分层位置和大小。分别采用数值验证和实验验证来双重检验ANN和SAO两种算法的识别精度和效率。数值验证结果表明:ANN和SAO两种逆向检测算法对分层损伤位置和大小的识别最大误差分别是5.04%（ANN）和5.24%（SAO），证明方法在理论上可行。实验验证结果表明:ANN在使用实测频率数据进行识别时预测精度很差，无法得到有效的分层损伤信息；而采用SAO可以较好地预测试件中的分层损伤，且对分层大小的预测比对分层位置的预测精度更高，其中，对贯穿损伤和底板损伤的大小预测误差分别不超过2.05%和9%，而四个试件中有两个试件预测的分层与实际的损伤部位存在重合（重合率分别为34%和32.65%）。因此，当前提出的ANN和SAO在理论上可行，但实际应用时都会受到不同程度实测数据误差的影响，相比ANN而言，SAO算法有更好的鲁棒性，在采用实测频率时也可以较为准确地预测出试件中的分层损伤。 

Frequency-based delamination detection in stiffened fiber reinforced polymer composite plates

The present work focuses on assessment of delamination damage in stiffened fiber reinforced polymer (FRP) composite plates through the changes in frequencies after delamination occurring. The two inverse algorithms, namely artificial neural network (ANN) and surrogate assisted optimization (SAO) were developed to predict the location and size of delamination in the stiffened FRP plates using a series of frequency shifts. The efficiency and accuracy of the frequency-based detection algorithms were validated both numerically and experimentally. The results of numerical validation show that the two proposed inverse detection algorithms can successfully identify the delamination in stiffened FRP plates with good accuracy (maximum errors are 5.04% for ANN and 5.24% for SAO, both in the prediction of delamination location). Compared to using the genetic algorithm directly, SAO can greatly enhance the prediction efficiency and maintain good accuracy. The experimental results show that the prediction accuracy of ANN is reduced greatly compared with the numerical validation due to the existence of the measurement noise in the testing, and ANN cannot give useful information of delamination in stiffened FRP plate specimens. But SAO still can obtain reasonable prediction accuracy, with maximum prediction errors of 2.05% and 9% for through-flange delamination and the embeded delamination in the base plate, while two out of four specimens have predicted the delamination to have overlapping areas with the actual delamination, of which the overlapping areas are 34% and 32.65%. In conclusion, frequency-based method using ANN and SAO as inverse algorithms is validated numerically to successfully predict delamination, but only SAO can give reasonable prediction in delamination size and location when using the measured frequencies of the stiffened plate specimens.