Steel rail damage detection method, system and terminal

Abstract

The invention relates to the technical field of railway traffic detection, in particular to a steel rail damage detection method, system and terminal.The steel rail damage detection method comprises the steps that original ultrasonic signals of a non-destructive steel rail are obtained multiple times through a lead breaking experiment, and segmentation preprocessing is conducted to obtain a plurality of ultrasonic signal sample training sets; inputting the plurality of ultrasonic signal sample training sets into an adversarial auto-encoder model for training to obtain a steel rail damage detection model based on multi-dimensional Gaussian probability distribution; and obtaining an ultrasonic signal of a to-be-detected steel rail, inputting the ultrasonic signal into the steel rail damage detection model for processing after segmentation preprocessing, and outputting a detection result of the to-be-detected steel rail. After the ultrasonic signal of the steel rail to be detected is input into the steel rail damage detection model, the detection result of the steel rail to be detected can be obtained. The detection method has the advantages of being extremely low in cost, high in detection efficiency, reliable in result and high in universality, the requirement for the professional degree of operators is low, and meanwhile the detection method is suitable for rapid track inspection of a large-size railway system. 本发明涉及铁路交通检测技术领域，尤其涉及一种钢轨损伤检测方法、系统及终端，钢轨损伤检测方法包括步骤：通过断铅实验多次获取无损钢轨的原始超声信号，进行分割预处理得到多个超声信号样本训练集；将多个超声信号样本训练集输入对抗自编码器模型中进行训练，得到基于多维高斯概率分布的钢轨损伤检测模型；获取待测钢轨的超声信号，经分割预处理后输入钢轨损伤检测模型中进行处理，输出待测钢轨的检测结果。通过将待测钢轨的超声信号输入钢轨损伤检测模型后，即可得到待测钢轨的检测结果。该检测方法具有成本极低、检测效率高、结果可靠且泛用性强的特点，对操作人员专业程度需求低，同时适用于大体量铁路系统的轨道快速巡检。