| The existing freight train fault detection system adopts a man-machine combination for fault diagnosis.On the one hand,the detection efficiency needs to be improved;on the other hand,the accuracy is affected by artificial factors,and the reliability is poor.According to the idea of positioning firstly and detection,based on the existing TFDS system and related research on fault detection algorithms,this thesis combines traditional image processing algorithms with algorithms based on features and machine learning,proposes an automatic detection algorithm for the detection of several typical freight train bottom faults,which meets high accuracy and real-time requirements of the train inspection.The specific research contents are as follows:1.Research on the train bottom low-quality image processing algorithms.First,a method for discriminating the train bottom low-brightness and low-contrast images is realized,the lowcontrast image enhancement is realized by the histogram specification,and the lowbrightness image enhancement is realized by the improved Lee enhancement algorithm.The enhancement algorithm in this thesis is proved by comparing with the traditional enhancement method and has better effect.Secondly,the discrimination method of bogie blurred image based on the variance of gradient map is realized,the blurred image restoration is realized by using Wiener filter,and the PSF scale parameter estimation is realized by using image differential autocorrelation.The slightly motion blurred images have better restoration effect.The average time of several algorithms is about 10 ms,meets the real-time requirements of the train inspection.2.Research on the fault detection algorithm for the block key missing based on shape similarity.Firstly,the step-by-step positioning process for the block key area is realized by using the geometric features of the axles in the bogie area and the positional relationship between the areas.Then,a fault discrimination algorithm based on the similarity of contour and shape is adopted,and the Hu’s moment feature,the shape distance and the Hausdorff distance are used to detect the block key missing,and the performance indicators of several algorithms are comprehensively compared.Finally,the step-by-step positioning combined with the contour matching algorithm based on the Hu’s moment feature is proposed as the fault detection method of the block key.The false detection rate and the missed detection rate are both less than 5%,and the average time is less than 100ms;the algorithm is accurate and fast.3.Research on the fault detection algorithms for the safety chain fasteners falling off and the brake shoe missing based on the GA-SVM.Firstly,the step-by-step positioning process for the safety chain and the brake shoe area is realized by using the geometric characteristics of the break beam body in the brake shoe area and the positional relationship between the areas;then a sample set of the fault area is constructed,and the SVM classifier is used to classify and predict the characteristics of the fault area,and the genetic algorithm(GA)is used to optimize parameters.Finally,the GLCM feature,the HOG feature and the LBP feature combined with the GA-SVM are used to detect the failure of the safety chain fasteners falling off and the brake shoe missing,and the performance of several algorithms is comprehensively compared.Finally,the step-by-step positioning combined with the LBP feature and the GA-SVM classifier is proposed as the safety chain fault detection method,and the step-by-step positioning combined with the HOG feature and the GA-SVM classifier is proposed as the brake shoe fault detection method.The false detection rate and the missed detection rate are both less than 5%,and the average time is less than 100ms;the algorithm is accurate and fast.4.The software platform of the train fault detection system based on C++ is built. |
