Research On On-line Detection System Of Wheelset Scratch Based On Image Processing

With the continuous promotion of the "One Belt,One Road" strategic layout,China's rail transit industry has developed rapidly,and the annual freight volume and passenger traffic of the railway have been continuously grown.The safe driving of trains is closely related to the national economy and people's livelihood.The train wheel is a very important part of the bogie,and its working condition will affect the sTab.ility and safety of the train.At present,the detection of the wheelset state mostly uses manual detection and Photoelectric measurement.These methods need to be carried out when the train is out of service,and it is time consuming and complicated,and the real-time performance is poor,which is very likely to cause hidden dangers.Therefore,real-time detection of wheel tread5s scratches is very necessary.In this paper,a combination of image processing and neural network pattern recognition is used to design a set of wheel scratch online detection devices,which is aiming at realizing online detection and making early warning of wheel.The main research contents are as follows:1.Image acquisition hardware system.This paper selects STM32F407 MCU as the control chip to improve the processing speed and robustness of the acquisition system.The camera uses a high-resolution COMS lens and is calibrated using the Zhang Zhengyou's calibration method.The microcontroller starts to collect image by receiving an external trigger.The system uses high-speed Ethernet transmission to ensure the real-time and concurrency of multiple acquisition systems.2.Algorithl and flow of tread image processing.In order to improve the processing speed of the system,the acquired image is first grayed out;the image is restored by the Lucy-Richardson algorithm to remove the motion blur;the Gaussian filtering method is used to remove the Gaussian and salt and pepper noise in the image;The grayscale transform method is used to compensate the image for illumination,.Until this step,the image pretreatment is not completed.In this paper,four edge detection operators are compared.The Canny operator is used to detect the edge and then fit the edge of the wheel to extract the tread portion.Finally,the texture feature of the tread image is extracted by the method of gray level co-occurrence matrix.3.Improve the pattern recognizer of the GA-BP neural network.The extracted tread image texture features are input into the pattern recognizer of BP neural network for supervised learning,and the output classification result is used as a criterion for positioning scratches.In this paper,genetic algorithn and BP neural network are combined,and the improved GA-BP neural network is used for pattern recognition to improve the recognition rate and sTab.ility of the network.The results show that the optimized neural network recognition rate can be as high as 80%or more,which is much higher than the traditional BP neural network algorithm.4.Tread detect online software system.The software part uses Labview and MATLAB hybrid programming to design a friendly interactive interface.The former implements the control of the hardware acquisition system,and the latter completes the image processing algorithm and neural network pattern recognition.When the tread image that has been collected is input during the detection,the automatic recognition of the tread scratches areas can be realized and circled on the interface,and the area of the scratched areas is calculated and the classification early warning is performed.The system also realizes splicing multiple tread images of the same wheel to show the scratching of the entire tread.