Research And Optimal Design Of Detection Algorithm For Digital Track Detection System

It is important for railway engineers to control tack geometry variations in some railway by track geometry detection system which is installed on an inspection vehicle.According to detection results,the maintenance plan can be drawn up in details and maintenance cost can be estimated accurately.The thesis presents some advanced algorithms about detection models based on GJ6 D,the digital platform for tack geometry detection system.In addition,some new detection methods are proposed in the thesis,such as detection of slope and vertical curve.Specially,the contents are as follows:(1)The high pass filter with smaller pass band ripple and faster attenuation of transition band are designed based on three trapezoidal windows in parallel.For different design cutoff wavelengths,the corresponding band-limited track irregularities are artificially generated to evaluate the detection accuracy of different types of filters.The experimental show that the detection result for track irregularity of high pass filter based on three trapezoidal windows in parallel is more accurate.(2)For large radius curve recognition,this thesis shows the reason that the curve recognition algorithm has a large deviation in the detection result,and then puts forward the corresponding optimization scheme.By means of adjusting the scale coefficient of the yaw gyro and filtering the curvature and curvature variation with appropriate low-pass filter parameters,as well as resetting the parameters of curve recognition algorithm,the results show that the accuracy of the algorithm for large radius curve detection is greatly improved.In addition,a new curve feature point detection algorithm based on convolutional neural network is explored and tested.The algorithm has higher universality and accuracy,but it’s shortcoming lies in the large amount of computation and poor real-time performance.(3)For the cross level detection,this thesis shows the main reason for the bulging phenomenon of the cross level detection results at the curve feature point,then two optimization scheme are proposed.The first scheme is to design a digital filter with smaller pass band ripple and faster attenuation of transition band to detect the cross level,and the results show that this method can greatly reduce or even eliminate the bulge phenomenon if the cut-off wavelength of the filter remains unchanged.The second scheme is based on discrete wavelet transform,and the bulge phenomenon can be completely eliminated by selecting appropriate wavelet basis function and decomposition layer number as well as threshold values.(4)Establishing the mathematical model of slope and vertical curve detection based on the basic principles of strapdown inertial measurement.For slope detection,three kinds of slope detection algorithms are designed and experimented,and the detection results of the three algorithms are analyzed and compared.The results show that the detection algorithm based on complementary filtering has the best detection results for slope detection.For vertical curve detection,the corresponding detection algorithm is designed based on its mathematical model,and the accuracy of the algorithm on vertical curve detection results is analyzed in combination with the slope detection results.