CAJ | 학술논문

为准确进行结构损伤预警,提出基于前馈BP网络实现新奇检测技术的斜拉桥损伤预警方法.以斜拉索局部振动模态的基频作为网络的基本输入,定义新奇指标为网络输出,根据训练阶段和检测阶段的新奇指标间的比较指示结构的健康状态.相比一般的损伤检测,其不依赖于数值模型,避免了对数值模型高精度的要求,提高了实用价值.对汲水门大桥14种损伤的模拟结果表明,在1%的噪声水平下,该方法可达到较好的预警率.
위준학진행결구손상예경,제출기우전궤BP망락실현신기검측기술적사랍교손상예경방법.이사랍색국부진동모태적기빈작위망락적기본수입,정의신기지표위망락수출,근거훈련계단화검측계단적신기지표간적비교지시결구적건강상태.상비일반적손상검측,기불의뢰우수치모형,피면료대수치모형고정도적요구,제고료실용개치.대급수문대교14충손상적모의결과표명,재1%적조성수평하,해방법가체도교호적예경솔.
A novelty detection technique was proposed based on feed-forward BP neural network for accurate alarm of structural damage in cable-stayed bridge. Natural frequencies of local vibration modes of stay cable were taken as the inputs of the network and a novelty index was defined by the outputs of the network. The structural state was indicated by comparing both the novelty index of training and testing phases. Comparing with general damage detection, the proposed technique only depends on non-structural model, which avoiding the high-precision demand and its utility value was improved. A total of 14 potential damage cases in a cable-stayed bridge were simulated, and results show that the proposed method is practicable for damage alarming under the noise of 1%.