识别一个结构在震动状态下的变化，在结构监测中是十分重要的，神经网络就非常适用于这种目的。本文研究了使用可分析的学习样本来训练神经网络的可行性问题。神经网络从损伤状态中训练产生，然后用于诊断一个五层钢框架在一系列震动模拟中的状态。结果表明，使用神经网络可使在线结构诊断更加可行。 Identifying changes in a structure under vibratory conditions is important in structural monitoring, and neural networks are well suited for this purpose. This paper studies the feasibility of training neural networks using analyzable learning samples. The neural network is trained from the damaged state and then used to diagnose the state of a five-story steel frame in a series of shaking simulations. The results show that using NN can make online structure diagnosis more feasible.