Research On Denoising And Recognition Of Rail MFL Signal Based On Adaptive Filtering And Random Forest

Railway safety is related to national economic development and people’s life safety,so it is of great significance to conduct rail flaw detection and damage assessment.Rail MFL(magnetic flux leakage)detection is one of the fast,accurate and non-destructive testing technologies that can adapt to high-speed patrol conditions.It is widely used in rail surface inspection.Based on the analysis of the current research status of rail MFL detection technology at home and abroad,this paper makes a related research on how to further improve the accuracy of the MFL detection signal filtering and damage identification.The main research works are as follows:(1)Studying the research literature of rail magnetic leakage detection technology at home and abroad in recent years,summarizing the current achievements and existing problems in this research field,on the basis of which the research directions and method;(2)Research on filtering of MFL signals.Aiming at the filtering of MFL signals in the field orbit detection process,combined with wavelet denoising and adaptive filtering methods,respectively,for the filtering problem of high-speed MFL detection signals at rail treads and gauge angles,a horizontal and vertical array sensor noise reconstruction algorithm is proposed.An adaptive interference canceller is constructed to filter the multi-channel MFL signals of the rail tread and gauge angle.The filtering results show that the average noise intensity is reduced by90.5%,and the signal-to-noise ratio is significantly improved;(3)Research on the location of MFL signals of defects.Aiming at the problems of noisy defect MFL signal localization accuracy and so on,a natural defect MFL signal localization algorithm based on difference degree and defect phase difference was proposed.The results show that the positioning effect is good;(4)A rail surface defect classification model based on random forest was developed.Aiming at problems such as weak interpretation power of traditional rail surface recognition algorithms,multi-channel and multi-directional information is combined,and a rail forest surface defect classification model based on random forest is designed.The accuracy of identifying 18 rail surface defects can reach 99.54%.The influence of different combination of feature vectors on recognition accuracy is analyzed,and some suggestions are made for the selection of feature vectors.