中文摘要：

随着列车运行速度和轴重增加,车轮踏面擦伤现象不断加剧。提出一种对列车经过时钢轨振动信号进行高阶谱特征提取并结合粒子群-支持向量机（PSO-SVM）进行车轮擦伤识别的算法。通过建立车辆轨道垂向耦合模型和车轮擦伤模型,计算正常车轮与擦伤车轮作用下的钢轨振动响应。利用高阶谱方法对两种情况下钢轨振动信号进行信号处理得到二维等高线图和三维双谱图,通过灰度-梯度共生矩阵提取其二维等高线图的6个纹理特征,与车速共同输入PSO-SVM模型识别车轮是否擦伤;再结合三维双谱图对角切片峰值、二维等高线图内扩对角频率,对擦伤等级进行识别。结果表明：利用高阶谱进行特征提取的方法识别正常与擦伤车轮准确率可以达到100%,擦伤等级的准确率可以达到94.6%。最后将该方法与EMD方法进行特征提取做比较分析,EMD方法识别正常与擦伤车轮准确率为98.3%,而擦伤级别准确率仅为56.4%。研究结果表明,基于高阶谱的PSO-SVM方法更能有效识别擦伤车轮并确定其擦伤等级。

英文摘要：

With the increasing of train speed and axle load, the wheel flats occur more frequently. A detection method for wheel flats is proposed based on PSO-SVM, in which high-order spectrum is used to extract the features of the rail vibration responses when the train runs on the rail. A vehicle-track vertical coupling system and wheel fiat are modeled and the rail dynamic responses are calculated with the condition of normal wheel and wheel fiat. The two-dimensional contour and three-dimensional double spectrum diagrams are obtained by analyzing two kinds of rail vibration responses with high-order spectrum method. Six texture features of the 2D contour map, which are extracted from gray-gradient co-occurrence matrix, are combined with the train speed to detect whether there is a wheel fiat by PSO-SVM. Then the peak value of the diagonal slice of the three-dimensional double spectrum map and the frequency of the inner spread of the two-dimensional contour map are added into PSO-SVM to recognize different wheel-fiat levels. The results show that the detection accuracy of normal wheel and wheel fiats can be up to 100% and the recognition accuracy of wheel fiat different levels is 94.6%. Finally, the recognition accuracy are compared and analyzed between high-order spectrum and EMD methods which are used to extract the feature signals. For EMD method, the detection accuracy for wheel fiats is 98.3%, and for different wheel fiat levels is only 56.4%. The investigation shows the detection method based on high-order spectrum and PSO-SVM can be used to identify wheel fiat and its levels effectively.