超声波钢轨探伤车B显数据伤损模式分类技术研究

BP Neural Network Based on Rail Flaw Classification of RFD Car's B-scan Data

既有钢轨探伤车超声检测系统均为国外引进。中国铁道科学研究院自主化钢轨探伤系统目前已经通过试验验证,所采用的伤损模式分类技术是其中的一项关键技术。根据自主化钢轨探伤系统的探轮设置,规定伤损的类别设置,并将伤损类别划分为四大类:核伤类、螺孔裂纹类、轨底裂纹类、水平裂纹类。归属于不同大类的伤损采用不同神经网络进行识别。以螺孔裂纹类为例进行说明,采用标定线数据进行测试,验证算法的有效性。

Currently the ultrasonic inspection systems equipped on domestic rail flaw detection car are all imported from foreign countries. The rail flaw detection system independently developed by China Academy of Railway Sciences has passed the test and verification, of which the flaw classification system is a key technology. The probe of the rail flaw detection system is set to define different defect types, including transverse defect, bolt hole crack, rail base crack and horizontal crack. Different neural networks are used to identify different defects. The bolt hole defect is also taken as an example to verify the effectiveness of the algorithm using calibration line data.