高速轨道超声成像伤损检测及其参数学习方法

Damage detection and parameter learning method for high speed rail ultrasonic imaging

针对高速轨道伤损检测问题,提出一种基于0°、37°、70°超声探头探伤的检测方法。该方法基于B型图像显示分析了各伤损的颜色、面积、倾斜角度、长度、质心坐标等特征,并根据其伤损特征的内在逻辑关系设计了检测算法。此外,由于超声成像过程受多种不确定因素的影响,同类伤损的图像特征常出现较大差异而影响检测准确率,为了提高算法检测的准确率,提出一种参数学习方法,该方法可实时调整检测参数的阈值。首先,基于建立的检测算法模型提取伤损判定过程的检测参数;其次,结合伤损的关键检测参数,以相同特征约束下的同区域轮廓其类别特征间隔最大化为准则,基于支持向量机建立了一种检测和学习相结合的学习模型,并基于该模型对参数阈值进行优化调整。实验结果分析表明,采用所提方法,其轨道伤损检测准确率可达97.5%;并对初检中检测率较低的伤损进行学习再检测,其准确率得到了明显提高,从而验证了所提方法的有效性。

Considering the problem of high-speed rail damage detection,a damage detection method was proposed based on 0°,37°and 70°ultrasonic probes.Based on B-scan image of damage,the color,area,tilt angle,length,and center of mass coordinates were analyzed in the proposed method,and the damage detection algorithm was designed based on their logical relationships of damage features.Furthermore,the image features of the same damage type usually show obviously differences due to uncertainty disturbance factors in the ultrasonic imaging process,which will affect its detection accuracy.For these reasons,a parameter learning method was proposed to improve the detection accuracy of the proposed algorithm,which could adjust the thresholds of the detection parameters in time.Firstly the detection parameters were extracted based on the established detection algorithm model.Secondly combining with the key detection parameters of damage,a learning model based on support vector machine was established to maximize the class feature interval of contours in the same region under the same feature constraints,and the parameters threshold was optimized and adjusted based on the model.The experimental results showed that the accuracy of rail damage detection could reach 97.5%by using the proposed method,and the accuracy could be significantly improved by learning and re-detecting the damages with low detection rate in the initial inspection.The experimental results verified the effectiveness of the proposed method.