Online Rail Defect Detection Method Based On Multi-Sensor Information Fusion

Railway, as an important type of national infrastructure and a popular means of transportation for the masses, plays a very critical role in facilitating the development of national economy. As the railway technologies continues to develop toward the goals of "high-speed passenger transport, heavy duty freight transport and high density traffic", the greater train weight, the larger wheel/rail force and the increased frequency of track use lead to higher speed of damage to the rail and incur defects. Occurrence of rail defects not only affects the normal and smooth running of train, resulting in less passenger comfort, but also in seriously cases causes fatal accidents such as train derailment and overturning, which endanger the train safety. Therefore, in order to adapt to the healthy and rapid development of the high-speed rail and ensure safe and smooth operation of train, strengthening the rail defect detection is of great significance. However, the existing rail defect detection methods mostly study on the single-sensor information, which has limitations and one-sidedness and is difficult to efficiently and reliably achieve the rail defect detection. In this thesis, the rail defects are studied and the online rail defect detection method based on multi-sensor information fusion is developed. While the train is running, it can better achieve the online judgment, locating and identification of rail defects and make up for the deficiencies of the existing rail defect detection method with only single-sensor information. In order to make up for the deficiencies of the single sensor information and produce high-quality data more suitable for vibration signal processing and image processing techniques to extract rail defect feature information, the thesis builds up a rail defect detection model for single type multi-sensor information fusion and a rail defect detection model for multi-type multi-sensor information fusion oriented by the multi-sensor data fusion after studies the fault type and characteristics of the vehicle-track coupling system, the key structural parameters and the sensor type selection, layout and networking solutions. In order to overcome the limitations and one-sidedness of the single sensor information studies, based on the rail defect detection model for single type multi-sensor information fusion, the information coupling between the sensors of the same type is fully dug out and a rail defect detection method on the basis of the spatial information reconstruction is proposed. The method reconstructs the space impact through the processing such as multi-sensor data preprocessing, information alignment and shifting, energy enhancement and superposition, the aperiodic impact of the surface rail defect is converted to the reconstruction data with the characteristics of periodic impact in the local space and the rail defects are detected through the time frequency analysis method. Simulation analysis and application research show that this method can significantly improve the SNR of train vibration signals, reconstruct the rail defect feature with local periodic characteristics, and more accurately and reliably achieve the judgment and locating of rail defects based on the vibration signal processing techniques. For eliminating the interference information and highlight the rail defect feature information, based on the rail defect detection model of single type multi-sensor information fusion, the information relevance among the sensors of the same type is fully dug out and a rail defect detection method on the basis of the two-dimensional impact reconstruction is proposed. The spectrum is obtained through the shifting time frequency analysis of more relevant multi-sensor information, the background signals of noise are removed by the image reconstruction of spectrum, the periodic impact signals are blanked, and thus the train fault information and noise interference information are eliminated. The aperiodic impact of the feature rail defects is converted to the reconstruction data with the characteristics of periodical impact in the local space, which are highlighted, and thus achieving the rail defect feature information extraction. Simulation analysis and application research indicate that this method can effectively remove background noise signals, partially eliminate the periodical impact caused by the wheel and rail, reconstruct and highlight the rail defect features with the local periodical characteristics, and can more directly, accurately and reliably realize the judgment and locating of rail defects based on vibration signal processing technique. For circumventing the limitations of the single type sensor information research, based on the rail defect detection model of the multi-type multi-sensor information fusion, the information relevance among multiple types of sensors is fully dug out and a rail defect detection method on the basis of the impact guide technique is proposed. The method conducts the preliminary detection of rail defects through the information captured by the vibration impact sensors, guide the image sensors to capture the images at fixed points by using the extracted impact feature information, then analyzes and processes the images and ultimately realizes the accurate judgment of rail defects. Theoretics study and simulation analysis research reveals that this method brings into full play to the advantages of the vibration signal analysis and image processing technologies and can more efficiently and reliably achieve judgment, locating and identification of rail defects. Finally, the research achievements are put into application, focusing on the multi-sensor data information algorithm, mainly by the software development, the rail defect detection system based on the multi-sensor information fusion is developed. The research results in this thesis indicate that the online rail defect detection method based on multi-sensor information fusion makes up for the one-sidedness of the vibration signal analysis based on the single sensor information and can more accurately, reliably and directly achieve judgments and locating of rail defects. It avoids the limitations of the image processing technique in need of covering the entire image acquisition, and can more efficiently and reliably realize judgment, locating and identification of rail defects. Therefore, it is an effective rail defect detection method.