Acoustic Detection Method And Limit Of Short-pitch Rail Corrugation On Metro System

Rail corrugation is one of key issues in urban rail transit.It affects the running safety and ride comfort of the train.With the increase of running speed,operation density,passenger capacity of the trains,and the use of the damping tracks,rail corrugation is becoming more and more serious.Rail grinding is the most effective means to remove rail corrugation.Timely knowledge about the rail corrugation condition and making reasonable corrugation limit are the basis for formulating the rail grinding plan.At present,the static methods such as straight edge and corrugation analysis trolley are mainly used to measure rail corrugation.The biggest advantage of this type of method is that it can describe the characteristics of rail corrugation more accurately,but the test efficiency is very low and requires a lot of manpower and material resources.With the increasing density of metro network in major cities,the traditional static measurement methods of rail corrugation can no longer meet the increasing demand of detection.Thus,it is urgent to develop more rapid and more efficient dynamic detection methods.The dynamic detection method is an indirect method of rail corrugation measurement,which can quickly assess the rail corrugation state by monitoring the vibration or noise or wheel–rail force caused by rail corrugation during train operation.This type of method faces many challenges,among which the accurate identification of the corrugation characteristics is the key and difficult point.In terms of corrugation limit,there is no unified standard based on metro interior noise control,and the existing limits do not fully consider the actual operating conditions of trains.It is of great significance for rail grinding and riding comfort of trains to combine corrugation limit with metro noise control under different operating conditions.The state-of-the-art overview of the rapid acoustic detection and limit of rail corrugation is presented.The necessity and feasibility of rapid detection of metro rail corrugation as well as the key points and difficulties in the dynamic detection of rail corrugation are discussed.Then,the research direction of this thesis is clarified.In this thesis,aiming at the detection and control of short-pitch rail corrugation on metro lines,the metro noise signal is used as the analysis object.Field test and numerical simulation are combined to carry out the following research works.(1)Field tests are conducted to summarize the characteristics of rail corrugation and the effect of rail corrugation on metro noise,and to determine the relationship between metro noise and rail corrugation.The above work provides a data basis for the subsequent characteristic recognition and classification of short-pitch rail corrugation based on metro acoustic signals.A prediction model of wheel–rail rolling noise is developed,which consists of wheel–rail interaction model,wheel–rail vibration model and wheel–rail acoustic radiation model.The prediction model is verified with experimental data.It provides a reliable tool for quantitative research on the relation between rail corrugation and noise.(2)Aiming at the shortcomings of the current rail corrugation dynamic detection method in terms of rail corrugation amplitude identification,a characteristic extraction method based on wheel–rail rolling noise wavelet packet decomposition is presented.The node energy level of the corresponding wavelet packet is used as an index to measure the rail corrugation amplitude at the corresponding wavelength.By using the time-domain wheel–rail rolling noise prediction model and numerical fitting/kriging surrogate model,the quantitative relationship between the short-pitch rail corrugation of different amplitudes and the corresponding wheel–rail rolling noise characteristics is established.The validity of the method is verified,including the following three parts.First,the adaptability of the numerical fitting is verified.The result shows that the quantization relation established by corrugation with single wavelength can still estimate the amplitude of corrugation with compound wavelengths,and the average error between the designed amplitude and the estimated amplitude is approximately 3.9%.Second,the accuracy of the interpolation points generated by the Kriging model is verified.The result shows that the average error between the design amplitude and the interpolated amplitude is approximately 2.7%,indicating that the Kriging model can accurately predict the unsampled point.Third,the effectiveness of the proposed method is verified by the field test,which shows that the average error between tested results in several sections and the estimated amplitude is approximately 10%.In the end,the error source and issues in practical application are discussed.(3)In terms of characteristic classification of rail corrugation,the wheel–rail noise is used as classification sample,a complete classification system of short-pitch corrugation is established using test samples.For the wavelength classification,rail corrugation with wavelengths in the range of 31.5–50 mm on metro lines are considered,including single wavelength case and compound multi-wavelength case.For the amplitude classification,the common amplitude range in the lines is divided into several different classes.A complete characteristic recognition framework of rail corrugation based on support vector machine is built.According to the classification characteristics including the wavelength and amplitude of rail corrugation,a characteristic extraction method based on time-domain statistical characteristics and wavelet packet node energy is proposed.Combined with the D-score–SFS–SVM characteristic selection method,the characteristics of samples,which are most important for corrugation classification,are extracted and selected,and the dimensionality is reduced.Due to the limit of the field test,the tested samples are imbalanced.For this situation,the accuracy of wavelength is 84.4%,and the average accuracy of amplitude classification is84.2%.To solve the problem of low classification accuracy caused by imbalanced sample,an data enhancement method based on wavelet decomposition packet coefficient editing is proposed.It can effectively improve the classification accuracy of the model.After the data enhancement,the accuracy of wavelength is 97.3%,and the average accuracy of amplitude classification is 98%.(4)By using the prediction model of wheel–rail rolling noise,combined with the relationship between the measured interior noise and wheel–rail rolling noise,a fast calculation method of interior noise based on the body noise transfer function is proposed.This method has been validated by the test data,which indicates that the proposed method can accurately and quickly calculate the interior noise.The quantitative relationship among interior noise,wheel–rail rolling noise and rail corrugation is established.On this case,a method for determining the amplitude limit of rail corrugation based on the control of interior noise is proposed,and the rail corrugation limits under different operating conditions including running speed,trackform and operating environment are investigated.