Study On Train Faults' Automatic Recognition Based On Digital Image Processing Algorithm

As one of the connected parts of the train coupler and shock-absorber, if the thumb head tail hook dowel is lost or broken, it will be very dangerous. And the round brake beam stanchion dowels'missing is the main reason of train brake beam's breaking off accident. So if we want the train to be safe, it is necessary to ensure the thumb head tail hook and round brake beam stanchion dowels are in good condition. The traditional examination methods for testing these two kinds of parts are Static detection of artificial. But because of the continuous improvement of train speeds, the maintenance cycle's reducing and the traditional examination methods'shortcomings which is low efficient and poor reliable, it is difficult to meet the requirements of railway sector. Therefore, to realize the above two kinds of fault's automatic detection is of vital significance to train traffic safety protection.The thesis is continuity of the study of "the dynamic detection system for train faults". Based on digital image processing, I design the algorithm according to the two kinds of the thumb head tail hook and round brake beam stanchion dowels'image characteristics.The article explains the classical algorithms for digital image processing which will be used in this thesis have explained in detail. Such as image geometric transformation, histogram-based gray-scale transformation, smoothing, gray mapping, edge detection using gradient information, template matching and so on. The reasons of using these algorithms have been expatiated too.According to the features of thumb head tail hook and brake beam stanchion dowels, how to design to extract the features of these two kinds of image correctly, and efficient and accurate algorithm to identify the faults are the important and difficult parts of this issue. Two different faults'recognition algorithms are finished with the classical algorithms for digital image and several new kinds of algorithms.Algorithm function's implementation requires software platform to support. By comparison with other software, the C++ Builder visualization programming software has been used to complete the platform debugging, design and generating DLL files.At last, according to the current image sample, the reliability and efficiency of the algorithms has been analyzed. Actual application shows that the algorithms meet customer's requirements of recognition rate.