基于频率下降率的刚性路面脱空自适应神经网络识别研究

Study on Self-Adaptive Neural Network Identifying of Void on Rigid Pavement Based on Frequency Drawdown Ratio

为快速而又有效地进行无损检测 (NDT) ,探讨了动量系数和学习率自适应调整的神经网络算法 ,及刚性路面脱空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脱空识别的特征参数。利用有限元方法对刚性路面脱空进行数值模拟 ,同时采用声振法研究了刚性路面板声学特征变化的关系 ,分别获取训练样本数据 ,通过自适应神经网络对刚性路面脱空进行了识别研究。从中可以看出 ,采用频率下降率和自适应神经网络技术对刚性路面脱空进行缺陷识别分析具有较高的精度和可靠性。

The self-adaptive neural network about adjusting mom en tum vector and learning rate, and the selection of characteristic parameters of void identification on rigid pavement for nondestructive testing (NDT) fast and validly are discussed, and the frequency drawdown ratio as characteristic parame ters put forward in this paper. The training sample data are obtained through nu merical simulation of finite element Relation of acoustic signature changes of rigid pavement is studied by acoustic-vibration The problems about void on rigid pavement are identified by means of self-adaptive neural network. It can be proved from the examples that this method is precise and reliable T he bases of theory and experiments are guided for NDT of defect identifying of void on rigid pavement with acoustic signature inspection