论文部分内容阅读
目的:面向铁路货车关键机械部件的健康状态监控,针对铁路货车闸瓦钎这种复杂机械部件,实现基于视觉图像的户外全天候实时自动故障检测。创新点:针对闸瓦钎这种复杂目标机械部件的故障检测,提出一种新颖的实时精确故障检测方法。鉴于目标部件故障样本和无故障样本存在极强的类间相似性和类内差异性,情况相对复杂,提出采用多特征多层级方式。多特征避免单一特征的局限性和片面性,满足系统高精度要求,而多层级级联方式可事先排除大量无关背景信息,满足系统实时性需求。方法:采用层次化故障检测思路,在ROI分割上(图10),提出采用多尺度中心变换编码(MSCT),通过构建改进的空间金字塔方式实现。在闸瓦钎定位上,在梯度域对闸瓦钎部位进行中心变换编码,以梯度编码直方图(HEG)特征构建特征向量,采用SVM训练生成定位分类器。故障状态分类器的构建与之相似,但编码是建立在灰度图像基础上,最终在分割出的ROI中通过定位和判别分类器级联方式实现闸瓦钎丢失故障的全自动检测,无需任何人工参与过程。结论:针对现有铁路故障检测技术存在的不足,提供一种铁路货车闸瓦钎丢失故障的自动检测方法,既可降低铁路货车故障检测成本,又可提高铁路货车故障检测效率,为铁路提速提供了可靠的安全保障。相应实验表明该系统故障检测率达到了99.2%(表2),而检测速度接近5帧/秒,具有很好的实时性和很高的检测精度。
OBJECTIVE: To monitor the health condition of key mechanical parts of railway wagons and to realize the real-time automatic fault detection of outdoor weather based on visual images in view of the complicated mechanical components of railway wagons. Innovation: Aiming at the fault detection of such complex target mechanical parts, this paper proposes a novel real-time accurate fault detection method. In view of the fact that there is a strong similarity and intra-class difference between the faulty sample and the non-faulty sample of the target component, the situation is relatively complicated, and a multi-feature multi-level approach is proposed. Multi-feature avoids the limitations and one-sidedness of a single feature, to meet the system’s high-precision requirements, and multi-level cascaded way to exclude a large number of unrelated background information to meet the real-time system requirements. Methods: Using hierarchical fault detection approach, on the ROI segmentation (Figure 10), we propose to use multi-scale center transform coding (MSCT), which is achieved by constructing an improved spatial pyramid. In the brake shoe brazing position, the center of the Brake shoe brazing position is transformed and encoded in the gradient domain, and the feature vector is constructed by the gradient encoding histogram (HEG) feature. The SVM training is used to generate the positioning classifier. The construction of the fault state classifier is similar, but the coding is based on the grayscale image. Finally, the automatic detection of the braze braze loss fault can be realized by positioning and classifying the classifier in the segmented ROI without any Artificial participation process. Conclusion: In view of the shortcomings of existing railway fault detection technology, this paper provides an automatic detection method for railway truck brake fault, which can not only reduce the cost of railway truck fault detection but also improve the efficiency of railway truck fault detection, Reliable security. The corresponding experiments show that the system failure detection rate reached 99.2% (Table 2), and the detection speed close to 5 frames / second, with good real-time and high detection accuracy.