Research On The Key Technologies Of Wheel Tread Defect Detection Based On Optoelectronic

The operating condition of the train wheel and the tread is directly related to the operating safety and quality.Therefore,the monitoring of the surface damage and defects is an important measure for the safety.In this paper,according to the application requirements of tread dynamic complete-cycle defect detection,the key technologies of wheel-to-tread fault detection were studied.First of all,according to the practical application scene and testing requirements of tread testing,the wheel tread detecting system is designed and constructed in this paper.It presented a systematic approach to complete-cycle dynamic detection for wheelset,and designed and optimized the key parameters in the system architecture.Aiming at the problem that the multi-line structured light is applied to the wheelset detection,this paper analyzes the data fusion of multi-light strips.Secondly,the key problems are analyzed in this paper that the light center of line structure is affected by the detection accuracy.In view of the problem that the existing methods are not sensitive to the complex steepness,it provides a new center line extraction method based on the adaptive template.By locating the center of the light strip and constructing the adaptive weight template and the correction template,the morphological change of the light strip of the algorithm is more sensitive and the influence caused by the scattered light is reduced.Compared to the steger method,the error is reduced by 30%,and the algorithm consumes 10% of the steger method.In order to solve the problem of three-dimensional reconstruction of translational and rolling objects at the same time,depending on the spatial morphological characteristics of the detection data in this application scenario,the three-dimensional reconstruction of the tread is transformed into the matching of multi-frame detection data in this paper.And on the basis of the existing matching methods,the algorithm is improved.Compared with the original method,the matching accuracy of the improved method is increased by 15% after the same number of iterations.The reconstruction results of this method and ICP method are compared and calculated,which proves the superiority of this method in reconstruction effect and robustness.Finally,in the aspect of tread defect detection,to solve the problem that the tread profile standard signal is difficult to be obtained precisely,a sparse representation overcomplete atomic library was designed according to the tread signal characteristics and reconstructs the defect-free template signal by the sparse representation algorithm based on the detection signal.There is no need to pre-obtain the template for the tread profile.In order to solve the issue of deep image defect detection,the defect signal is obtained by matching with the standard signal,and the depth image isgenerated by two-dimensional interpolation of the three-dimensional defect signal.A method of quantitatively detecting the defect based on the depth image is proposed.For the issue of detecting and classifying the defects in the tread by image,the defect detection and classification of the tread image are carried out through the deep learning network.After experimental verification,the error of quantitative calculation of defects based on depth images is less than 5%,which takes 5% of the conventional three-dimensional method.In this paper,the accuracy of image extraction is about 85%,which is at least 12% higher than the traditional convolutional neural network method.