Design And Implementation Of The Crack-detection System For Brake Disks Of High Speed Railway

Brake disks, the core of train braking system, will influence trains’braking performance. So it is significant to detect brake disks effectively to guarantee high-speed railway’s safe operation. An important index to measure the brake disk’s safety is the quantity and length of cracks on it. Compared with other countries, China lags relatively behind in technology to detect the cracks. Besides, the traditional methods of manual inspection and detection with miniature instruments can no longer meet the needs of mass detection because of such disadvantages as low efficiency, high labor intensity, high risk and error rate. Therefore, it is very necessary to develop a fast and accurate system to detect cracks on brake disks. The development of digital image processing technique provides a new idea in detecting cracks on brake disks, based on which, this thesis is written.After a careful study on the techniques related to the digital image processing, the author has designed and developed, according to the requirements of software and hardware for detecting cracks on the brake disks of high-speed railway, a complete set of system that can detect the cracks on brake disks. The system consists of a hardware system and a software system. The hardware system will be briefly introduced while the realization of the software system will be emphasized and discussed in details. As for the software system, there will be two parts to discuss. One is processing detection, which will be introduced in details, including four modules, respectively of calibration, image preprocessing, crack detection and crack calculation; the other is system management, thanks to which, workers can register and log in, inputting brake disks’ models and querying historical detection records; system managers can manage information of workers and brake disks.The innovation of this system is that the digital image processing is used in detecting trains’brake disks. As for the calibration in detection process, a reference-free calibration method is raised through employing an image edge detection technology, simplifying the process of practical detection. In addition, the purification and revision of preliminary result is available through employing classical camera calibration technology in machine vision, which ensure the calibration accuracy; By analyzing the crack image noise, the author select the appropriate method for pre-processing images, meanwhile, by comparison of different methods of segmentation, select the optimal solution to segment the Crack image by using the method for quantitative evaluation of artificial composite image segmentation algorithm. In the part of crack detection, the authors evaluate the effect of different image feature operators by Forward-Backward model, and chooses Haar feature as the feature of detection and recognition. In the meantime, the author construct the training of Adaboost classifier which is based on the extracted Haar features. Finally, the detection accuracy of the system is improved effectively. At the same time, the authors propose the idea that the model can be trained separately according to the possible situations of the crack and the crack detect by the characteristics of scoring.This research originates from a program by the Institute of Information of Beijing Jiaotong University, in which the author finished all work of developing the software system of this thesis. The software system is still in the experimental stage, because the hardware system hasn’t been processed into finished products. Though some improvements are still needed to be made, this research, compared with traditional detection methods, is an innovative attempt and does obviously improve detection efficiency and accuracy, having a positive influence on further researches in relevant research fields.