Train wheel tread will produce scrapes, peelings and other defects due to the friction between wheel and rail surface for its long-running process. Tread defects not only have a bad affect for the stability and security of the operation of the vehicle, but reduce the service life of the bearing and rail facilities and do harm for the safety and efficiency of rail transport. Among them tread scrapes and peelings are the two main defects of train tread. In order to achieve the detection and classification of tread scrapes and peelings, a method based on image processing and BP Neural Networks model was presented for detection and classification of scrapes and peelings in train wheel tread. First we preprocess the acquired images, and extract the defects. Next calculate four characteristic parameters including energy, entropy, moment of inertia and correlation, and eventually we calculate the mean and standard deviation of those characteristic parameters as the 8 texture parameters. Then we adopt principal component analysis method to turn 8 texture characteristic parameters of these two types of defects into three unrelated comprehensive variables. Finally by extracting and analysis the texture features of tread defects, the recognition correct rate reaches to 93.3%. The result shows that the method can meet the requirement of train wheel tread defects online-measurement.
|