基于频率下降率的结构损伤自适应神经网络识别

笔者探讨了动量系数和学习率自适应调整的神经网络算法及结构裂纹损伤识别特征参数的选取,提出以反映结构损伤位置和程度的频率下降率作为结构裂纹损伤识别的特征参数,利用有限元网格细化法对结构裂纹损伤进行数值模拟,获取训练样本数据,通过自适应神经网络对结构裂纹损伤问题进行识别研究。从结构裂纹损伤识别实例的结果中可以看出,采用频率下降率和自适应神经网络技术对结构裂纹进行损伤识别分析具有较高的精度和可靠性。

Self-Adaptive Neural Networks Identification of Structural Damage Based on Frequency Drop Rate

The self-adaptive neural networks of momentum vector and learning rate were discussed, and the selection of characteristic parameters of structural crack damage identification was studied. The frequency drop rates were put forward as characteristic parameters. The training sample data were obtained through numerical simulation applying fine grid method of finite element. The problem of structural crack damage could be identified by self-adaptive neural networks with sufficient precision and reliability.