A wheel polygon recognition model based on improved statistical geometric feature and support vector machine

Addressing the problems of long processing time, low accuracy and poor real-time performance in wheel polygon recognition, a new recognition model is proposed. The vertical vibration data of the axle box is converted from 1D to 2D and normalised into grey degree images. Subsequently, the grey degree image is decomposed into binary images through bit plane decomposition, and an improved statistical geometric feature (ISGF) method is utilised to extract textural features from the binary image. Finally, these features are used for the training and classification process of support vector machine (SVM) to diagnose and recognise wheel polygon faults. Through dynamics simulation verification and double-wheel test bench experiments, the results show that the model organically combines the fault data with the image recognition method, effectively restrains the noise interference in vibration data, and significantly improves the recognition accuracy, reduces recognition time and demonstrates strong generalisation performance.