基于RBF神经网络的尖轨导波裂纹信号定量方法

打开文本图片集

中图分类号：U216.3 文献标志码：A 文章编号：1671-5276（2025）05-0122-04

Quantitative Method for Crack Signal of Switch Rail Guided Waves BasedonRBFNeuralNetwork

JINGLixuan，WANG Ping，YANG Yuan，XU Yihang，LIURenbao （Colege of Automation Engineering，Nanjing Universityof Aeronautics and Astronautics，Nanjing 21106，China）

Abstract：Switchrail，categoriedasvariable cross-sectionrail，its irregular structuralcharacteristicsresult inthe complexity of ultrasonic guided wave detection signals，andwhen reflection methodappliedto detecting rail bottom cracks，theamplitudeof theecho signal failsto characterize thesizeof thecracks.To quantifycrack signals，a quantitative method basedon RBF neural network was proposed.For training and testing，a model was established， whose final prediction effect indicated that the root mean square error reached O.54，the coeficientof determination R2 achieved O.998 2，and the relative error of the test results was less than 10% and 5% ，accounting for 91% and 82% respectively.Comparison was made with traditional BP neural network，andthe resultsshowed that the performanceof theRBF prediction model was better than that of the BP model.

Keywords：ultrasound guided wave；switch rail crack；RBF neural network；feature extraction

0 引言

发生漏检与误判[7]。（剩余5984字）