CNN-based rolling stock brake valve detection method

Abstract

The invention belongs to the technical field of railway rolling stock maintenance, and particularly relates to a CNN-based rolling stock brake valve detection method, which comprises the following steps of: constructing a neural network system of a brake valve body by taking a brake valve as a carrier and taking data of each factor as neurons, and performing correlation analysis on the data of each neuron after importing to obtain a neural network model of the brake valve body; predictive diagnosis on the loading application effect of the brake valve is formed, whether loading application or re-maintenance is carried out or not is judged according to the diagnosis result, and finally the effects of reducing faults of a locomotive vehicle brake system and improving the transportation safety of railway vehicles are achieved. By adopting the brake valve detection method, the influence of a test environment, an application environment and maintenance factors on the performance of the brake valve can be fully considered, the brake valve is ensured to be put into loading and use in an optimal state after maintenance is completed, and the potential safety hazard of railway rolling stock driving is reduced.

Description

Technical Field

[0001] The invention belongs to the technical field of railway locomotive and vehicle maintenance, and in particular relates to a locomotive and vehicle brake valve detection method based on CNN. Background Art

[0002] Railway locomotives are vehicles that complete railway freight transportation tasks, and brake valves are one of the core components of railway freight cars. They are the key actuators for the deceleration and parking of locomotives and vehicles, and are essential devices to ensure the safety of train operation. For modern railways, the importance of braking is not only a safety issue, but braking has become an important factor restricting the further improvement of train speed and traction quality. The traditional railway locomotive brake valve detection method is to disassemble, overhaul, assemble, and test the brake valve before loading and using it. Because the test environment after overhaul is quite different from the actual environment of train operation, brake failures may also occur under different operating conditions during vehicle operation, such as emergency exhaust, exhaust port leakage, natural relief, etc., which bring safety hazards to railway transportation. For the above reasons, simply using the one-time test results of the brake valve in the brake room environment as the basis for loading and operation, especially the impact of marginal test data within the qualified range on the use effect after loading, lacks basic fault prediction methods, which will bring hidden dangers to vehicle driving safety. Therefore, in order to reduce the probability of failure during the operation of locomotives and ensure driving safety, it is necessary to conduct a comprehensive analysis of the brake valve data after the test, find out the key factors that may affect the performance of the brake valve and the effects of the influencing factors on the performance, and then form guidance on operation and maintenance to improve the safety performance of the brake valve. Because there is a correlation between the uncertainty of the brake valve operation site environment and the influencing factors, and the correlation factor data can be extracted or manually maintained through the railway vehicle operation related information system. Summary of the invention

[0003] In order to solve the technical problem that the influence of marginal test data within the above qualified range on the effect of use after installation and the lack of basic fault prediction means will bring hidden dangers to vehicle driving safety, the present invention provides a CNN locomotive vehicle brake valve detection method, which takes the brake valve as a carrier and various factor data as neurons to construct a neural network system of the brake valve body, and forms a predictive diagnosis of the effect of the brake valve installation and use through correlation analysis after importing the data of each neuron, and judges whether to install and use or re-inspect according to the diagnosis result, so as to finally achieve the effect of reducing the failure of the locomotive vehicle brake system and improving the transportation safety of railway vehicles.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0005] A locomotive vehicle brake valve detection method based on CNN, characterized in that it includes the following steps:

[0006] S1, classify and organize the brake valve test data and construct it into a two-dimensional convolutional neural network input layer;

[0007] S2. Construct the convolution kernel of test value factors, site environment factors, key component measurement data factors, operation interval environment factors, and personnel skill level factors. Define the correlation function through the correlation of test data and influencing factors, so as to dynamically determine the convolution kernel parameters according to the specific values.

[0008] S3, after the numerical value of the convolution kernel two-dimensional matrix is ​​determined, convolution feature extraction is performed on the input layer of the two-dimensional convolutional neural network;

[0009] S4. Classify and sort out the brake valve faults, other similar brake valve applications and off-site fault cause analysis conclusions through the transfer learning model for parameter adjustment of weight calculation incentive function, design the influencing factor weight calculation incentive function and substitute the parameters generated by the transfer model into the incentive function to calculate the weight of the influencing factor;

[0010] S5. Construct a weight feature pooling layer to extract the influencing effect features and form corresponding numerical analysis weights;

[0011] S6. The overall performance of the brake valve is used to form a judgment on the brake valve maintenance quality and a diagnosis of application prediction, completing the comprehensive inspection of the brake valve.

[0012] The method for constructing the two-dimensional convolutional neural network input layer in S1 is as follows: in the two-dimensional matrix, the row definitions are classified according to the brake valve detection data, namely, leakage amount Boolean type, leakage numerical type, leakage numerical type, pressure numerical type, and time numerical type; the column definitions are test items, namely, brake valve inflation position, emergency brake position, brake and relief sensitivity, local reduction valve, stability, emergency boost, full relief position, and balanced sensitivity; a 4X8 two-dimensional matrix is ​​constructed, and the data in the two-dimensional matrix are the values ​​converted from the original test records, and the items not involved are filled with 0, and the Boolean type is filled with 0 or 1 in the test record;

[0013] The method for constructing the test numerical factor convolution kernel in S2 is: according to the maintenance standard, the leakage deviation coefficient m and the time deviation coefficient n are set, and the convolution kernel sets the numerical interval feature to extract a 2X2 two-dimensional matrix, which respectively stores the maximum and minimum values ​​of the leakage standard and the maximum and minimum values ​​of the time standard; by multiplying the deviation coefficient corresponding to the input two-dimensional matrix data type, the interval exceeding the dynamic convolution kernel range is counted, wherein the time quantity uses the range median technology, and the deviation coefficient less than the median takes the negative absolute value to participate in the operation, and is extracted into a 2x2 two-dimensional matrix characterized by a counter, which is recorded as the leakage exceeding limit count and the time quantity exceeding limit count respectively.

[0014] The method for constructing the convolution kernel of the station environment factors in S2 is: according to the influence of the temperature and air source pressure of the brake valve test station on the test data results, the dynamic convolution kernel is set with the total air source of the station and the standard test air source pressure value; since the temperature and pressure mainly affect the time value, the extraction feature is a 2x2 two-dimensional matrix of the trend feature of the time performance, which is extracted as a 2x2 two-dimensional matrix characterized by a counter, and the deviation trend counts are recorded respectively.

[0015] The method for constructing the convolution kernel of the key component measurement data factor in S2 is: according to the deviation between the measurement data of the components of the brake valve and the standard value, a 2x2 two-dimensional matrix of the deviation coefficient affecting the feature extraction is set, and the values ​​of the two-dimensional matrix are the brake valve spring test deviation coefficient, the rubber template deviation coefficient, the sliding valve roughness deviation coefficient, and the sliding valve stop valve wear deviation coefficient. The 2x2 two-dimensional matrix of the feature extraction of the influence of the component test data deviation on the brake valve data is extracted, and the influence degree values ​​are recorded respectively.

[0016] The method for constructing the convolution kernel of the environmental factors in the application interval in S2 is: according to the temperature and humidity difference in the application interval and the application time point, a 2x3 two-dimensional matrix is ​​set to extract features, corresponding to the highest and lowest values ​​of the temperature and humidity difference and the estimated temperature and humidity values ​​at the operating time point, and extracted as a 2x3 two-dimensional matrix with a counter as the feature, and the deviation trend counts are recorded respectively.

[0017] The method for constructing the convolution kernel of the personnel skill factor in S2 is: setting a feature extraction 2x2 two-dimensional matrix according to the personnel skill level, rework rate, key maintenance failure rate, and working years, extracting the feature extraction 2x2 two-dimensional matrix of the influence of the personnel skill factor on the brake valve data, and recording the influence degree values ​​respectively.

[0018] The method for calculating the weight of the influencing factor in S4 by designing the influencing factor weight calculation activation function and substituting the migration model generation parameters into the activation function is as follows:

[0019] S4.1. Classify vehicle control valve faults into single vehicle faults, centralized wind test faults, and driving operation faults; then classify the main causes of the faults into a secondary classification; then classify the secondary classification abnormal causes again to form a third-level classification of component factors, such as templates, springs, slide valves, stop valves, mounting seats, and apertures; extract the fault probability m corresponding to the proportion of faults caused by the three-level classification components, and dynamically increase the probability value of the same type of brake valves with similar working principles based on the probability n, and take the median (mn) / 2 of the increased part as the fault pre-judgment;

[0020] S4.2. Classify the operational failures caused by temperature and humidity, with the number of events n and the total number of failures m, and calculate the probability n / m of failures caused by temperature and humidity. According to the open parabola relationship between temperature and humidity in the working process of the brake valve, calculate the value based on the temperature difference and humidity difference at the application site, and then multiply it by the probability to calculate the prediction probability factor.

[0021] S4.3. Classify the operational failures caused by personnel skills, the number of events is n, the total number of failures is m, and the probability of failure caused by personnel functional level is calculated as n / m. Based on the negative correlation between brake valve failure and personnel experience, the repair rate and working years are calculated and multiplied by the probability to calculate the prediction probability factor.

[0022] The method for constructing the weighted feature pooling layer in S5 is: the numerical sum of the test data convolution operation result S=∑i=1m∑j=1naij result and the deviation calculation f(n)=(a ij -(an 2 +bn-c)) 2 , find the derivative of f(n) and get a and b, and use the numerical weight activation function f(x) = ax 2 + bx + c, and the weight value A is obtained; the sum of the convolution operation values ​​of the station environmental factors S = ∑i = 1m∑j = 1naij and the deviation between the result and the expected numerical parameter n is calculated f(n) = (a ij -(an 2 +bn-c)) 2 , find the derivative of f(n) and get a and b, and use the numerical weight activation function f(x) = ax 2 + bx + c, and the weight value B is obtained; the sum of the convolution operation values ​​in the running interval S = ∑i = 1m∑j = 1naij and the deviation between the result and the expected numerical parameter n is calculated f(n) = (a ij -(an 2 +bn-c)) 2 , find the derivative of f(n) and get a and b, and use the numerical weight activation function f(x) = ax 2 +bx+c, and the weight value C is obtained; after the key component factor is calculated, the weight value D is obtained through the incentive function f(K)=K*∑i=1m∑j=1naij; the personnel skill factor is calculated through the incentive function f(K)=K*∑i=1m∑j=1naij, and the weight value E is obtained.

[0023] The method for forming a judgment on the brake valve maintenance quality and a diagnosis of application prediction for the overall performance of the brake valve in S6 is: data normalization is performed according to the weight influencing factors to form a comprehensive weight M=A+B*a+C*b+D*c+E, wherein the a coefficient is mainly the influence of the off-site wind source pressure and temperature difference on the failure of the concentrated part of the wind test, the b coefficient is the influence of the temperature and humidity in the application range on the application failure, and c is the influence of human factors on the failure of a single vehicle, off-site, and application range. The corresponding fault diagnosis is classified according to the transfer learning model, and finally the overall performance of the brake valve is formed to form a judgment on the brake valve maintenance quality and a diagnosis of application prediction to complete the comprehensive detection of the brake valve.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention solves the problem of potential safety hazards in railway driving caused by the traditional railway locomotive vehicle maintenance that simply relies on the one-time test results to install and use the vehicle without considering the influence of comprehensive factors such as the use environment factors, component replacement and running-in, and personnel maintenance level on the performance of the brake valve. By constructing a convolutional neural network for railway locomotive vehicle brake valve detection, the test data factors, station environment factors, vehicle use interval factors, component maintenance and replacement factors, and maintenance personnel skill factors are used as neurons in the neural network, and the faults and cause analysis that occur during the use of railway locomotive vehicles are summarized and classified to form a fault diagnosis library, and the fault phenomena and causes of brake valves with similar functions are classified and added to the brake valve fault diagnosis library through a transfer learning model, and the influence of the interrelated effects of neuron factors on the use performance of the brake valve in the actual process is found to design an activation function. After the convolutional neural network inputs the one-time test data of the test bench and is formatted into a two-dimensional array and brought into the neural network, the convolution kernel of each neuron completes the parameter configuration according to the actual value of different factors, and then the convolution operation mapping is performed and the factor influencing factor is obtained through the activation function. After the influencing factor is fully linked, it is normalized to the comprehensive diagnosis conclusion of the brake valve. By using this method to detect brake valves, the impact of the test environment, the operating environment and maintenance factors on the performance of the brake valve can be fully considered, ensuring that the brake valve is put into use in the best condition after maintenance, thereby reducing the safety hazards of railway locomotives and vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0027] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0028] Figure 1 It is a schematic diagram of the detection method of the present invention.

[0029] Among them: 1 is the test data input, 2 is the test numerical factor convolution kernel, 3 is the site environment factor convolution kernel, 4 is the operation interval environment factor convolution kernel, 5 is the key component factor convolution kernel, 6 is the personnel skill factor convolution kernel, 7 is the numerical factor feature extraction, 8 is the site factor feature extraction, 9 is the interval environment factor feature extraction, 10 is the component factor feature extraction, 11 is the skill factor feature extraction, 12 is the comprehensive weight, and 13 is the brake valve pre-diagnosis result. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. These descriptions are only to further illustrate the features and advantages of the present invention, rather than to limit the claims of the present invention. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0031] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0032] In this embodiment, if Figure 1As shown, it consists of a convolutional neural network input layer, an influencing factor convolution layer and its influencing factor convolution kernels, a correlation function, a transfer learning model, a weight calculation excitation function, a feature extraction pooling, and a comprehensive weight function. The test data input 1 is the original data input layer of the neural network, which is converted from the test data results into a numerical matrix through data sorting, and is processed with the corresponding convolution kernel to extract the corresponding features. The test numerical factor convolution kernel 2 and the station environment factor convolution kernel 3, the application interval environment factor convolution kernel 4, the key component factor convolution kernel 5, and the personnel skill factor convolution kernel 6 act as network neurons to affect the numerical factor feature 7. The station environment factor convolution kernel 3 acts as a network neuron to affect the numerical factor feature 7, the station factor feature extraction 8, and the component factor feature extraction 10. The application interval environment factor convolution kernel 4 acts as a network neuron to affect the numerical factor feature 7, the interval environment factor feature extraction 9, and the component factor feature extraction 10. The key component factor convolution kernel 5 acts as a network neuron to affect the numerical factor feature 7 and the component factor feature extraction 10. The personnel skill factor convolution kernel 6 acts as a network neuron to influence the numerical factor feature 7 and the skill factor feature extraction 11. The comprehensive weight value is formed by the numerical factor feature 7, the station factor feature extraction 8, the interval environment factor feature extraction 9, the component factor feature extraction 10, and the skill factor feature extraction 11, and then the comprehensive weight of each extracted feature is calculated. The brake valve pre-diagnosis result 13 is calculated, and the diagnosis conclusion, maintenance guidance and installation and use precautions are given according to the diagnosis result.

[0033] A locomotive vehicle brake valve detection method based on CNN, characterized in that it includes the following steps:

[0034] Step 1: Classify and organize the brake valve test data to construct a two-dimensional convolutional neural network input layer, where the rows in the two-dimensional matrix are defined according to the brake valve detection data classification, namely leakage Boolean type, leakage numerical type, leakage numerical type, pressure numerical type, and time numerical type. The columns are defined as test items, namely brake valve inflation position, emergency brake position, brake and relief sensitivity, local reduction valve, stability, emergency boost, full relief position, and balanced sensitivity. A 4X8 two-dimensional matrix is ​​constructed. The data in the two-dimensional matrix are the values ​​converted from the original test records. Items not involved are filled with 0, and the Boolean type is filled with 0 or 1 in the test record.

[0035] Step 2: Construct the convolution kernel of test numerical factors, site environment factors, key component measurement data factors, operation interval environment factors, and personnel skill level factors. Define the correlation function through the correlation of test data and influencing factors, and dynamically determine the convolution kernel parameters according to the specific numerical values.

[0036] Test numerical factor convolution kernel: according to the maintenance standard, set the leakage deviation coefficient m and time deviation coefficient n. Set the numerical interval feature of the convolution kernel to extract a 2X2 two-dimensional matrix to store the maximum and minimum values ​​of the leakage standard and the maximum and minimum values ​​of the time standard. By multiplying the deviation coefficient corresponding to the input two-dimensional matrix data type, count the intervals beyond the dynamic convolution kernel range. The time quantity uses the range median technology, and the deviation coefficient less than the median takes the negative absolute value to participate in the operation, and is extracted into a 2x2 two-dimensional matrix characterized by a counter, which is recorded as leakage exceeding limit count and time exceeding limit count respectively.

[0037] Convolution kernel of site environmental factors: According to the influence of the temperature and air source pressure of the brake valve test site on the test data results, the dynamic convolution kernel is set with the total air source of the site and the standard test air source pressure value. Since the temperature and pressure mainly affect the time value, the extraction feature is a 2x2 two-dimensional matrix with trend features of time performance, which is extracted as a 2x2 two-dimensional matrix with counter features, and the deviation trend counts are recorded respectively.

[0038] Convolution kernel of environmental factors in the application interval: According to the temperature and humidity difference in the application interval and the application time point, a 2x3 two-dimensional matrix is ​​set to extract features, corresponding to the highest and lowest values ​​of the temperature and humidity difference and the estimated temperature and humidity values ​​at the operating time point, and extracted as a 2x3 two-dimensional matrix with a counter as the feature, which records the deviation trend counts respectively.

[0039] Convolution kernel of key component factors: According to the deviation between the measured data of the components of the brake valve and the standard value, a 2x2 two-dimensional matrix of deviation coefficient affecting feature extraction is set. The values ​​of the two-dimensional matrix are the brake valve spring test deviation coefficient, rubber template deviation coefficient, sliding valve roughness deviation coefficient, and sliding valve stop valve wear deviation coefficient. The 2x2 two-dimensional matrix of feature extraction based on the influence of component test data deviation on brake valve data is extracted, and the influence degree values ​​are recorded respectively.

[0040] Personnel skill factor convolution kernel: set the feature extraction 2x2 two-dimensional matrix according to the personnel skill level, rework rate, key repair failure rate, and working years, extract the feature extraction 2x2 two-dimensional matrix of the influence of personnel skill factors on brake valve data, and record the influence degree values ​​respectively.

[0041] Step 3: After the numerical value of the convolution kernel two-dimensional matrix is ​​determined, convolution feature extraction is performed on the input layer of the two-dimensional convolutional neural network.

[0042] Step 4: Classify and organize the brake valve failure itself, other similar brake valve applications and off-site fault cause analysis conclusions through the transfer learning model for parameter adjustment of the weight calculation incentive function, design the influencing factor weight calculation incentive function, and substitute the parameters generated by the transfer model into the incentive function to calculate the influencing factor weight.

[0043] 1) Classify vehicle control valve faults into single vehicle faults, centralized wind test faults, and driving operation faults. Then classify the main causes of the faults into secondary classifications, such as external leakage, relief abnormality, poor braking, and emergency start. Then classify the secondary classification abnormality causes again to form a three-level classification of component factors, such as templates, springs, sliding valves, stop valves, mounting seats, and apertures. The fault occurrence probability m is extracted from the proportion of faults caused by the three-level classification components. The probability value of the brake valve with similar working principles is dynamically increased based on the probability n of the same type of occurrence. The median (mn) / 2 of the increased part is taken as the fault pre-occurrence judgment.

[0044] 2) Classify the operational failures caused by temperature and humidity, the number of events is n, the total number of failures is m, and the probability of failure caused by temperature and humidity is calculated as n / m. According to the open parabola correspondence between the temperature and humidity in the working process of the brake valve, the numerical value is calculated based on the temperature difference and humidity difference at the application site, and then multiplied by the probability to calculate the prediction probability factor.

[0045] 3) Classify the operational failures caused by personnel skills, the number of events n, the total number of failures m, and calculate the probability n / m of failures caused by personnel functional level. Based on the negative correlation between brake valve failure and personnel experience, the repair rate and working years are calculated and multiplied by the probability to calculate the prediction probability factor.

[0046] Step 5: Construct a weighted feature pooling layer to extract the influencing effect features and form the corresponding numerical analysis weights. The numerical sum of the experimental data convolution operation results S = ∑i = 1m∑j = 1naij results and the deviation calculation of the expected numerical parameter n f(n) = (a ij -(an 2 +bn-c)) 2 , find the derivative of f(n) and get a and b, and use the numerical weight activation function f(x) = ax 2 + bx + c, and the weight value A is obtained. The sum of the convolution operation values ​​of the station environmental factors S = ∑i = 1m∑j = 1naij and the deviation between the result and the expected numerical parameter n is calculated f(n) = (a ij -(an 2 +bn-c)) 2 , find the derivative of f(n) and get a and b, and use the numerical weight activation function f(x) = ax 2 +

[0047] bx+c, and the weight value B is obtained. The sum of the convolution operation values ​​in the running interval S=∑i=1m∑j=1naij and the deviation between the result and the expected value parameter n are calculated f(n)=(a ij -(an 2 +bn-c))2 , find the derivative of f(n) and get a and b, and use the numerical weight activation function f(x) = ax 2 +bx+c, and the weight value C is obtained. After the key component factor is calculated, the weight value D is obtained through the incentive function f(K)=K*∑i=1m∑j=1naij. The personnel skill factor is calculated through the incentive function f(K)=K*∑i=1m∑j=1naij, and the weight value E is obtained.

[0048] Step six. Finally, the data is normalized according to the weight influencing factors to form a comprehensive weight M=A+B*a+C*b+D*c+E, where the a coefficient is mainly the influence of the off-site wind source pressure and temperature difference on the failure of the concentrated part of the wind test, the b coefficient is the influence of the temperature and humidity in the operation range on the operation failure, and c is the influence of human factors on the failure of a single vehicle, off-site, and the operation range. The corresponding fault diagnosis is classified according to the transfer learning model, and finally the overall performance of the brake valve is formed to form a judgment on the brake valve maintenance quality and a diagnosis of application prediction, completing the comprehensive detection of the brake valve.

[0049] Only the preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention, and various changes should be included in the protection scope of the present invention.

Claims

1. A CNN-based vehicle brake valve detection method, characterized in that: The following steps are involved: S1, classify and organize the brake valve test data and construct it into a two-dimensional convolutional neural network input layer; S2. Construct the convolution kernel of test value factors, site environment factors, key component measurement data factors, operation interval environment factors, and personnel skill level factors. Define the correlation function through the correlation of test data and influencing factors, so as to dynamically determine the convolution kernel parameters according to the specific values. S3, after the numerical value of the convolution kernel two-dimensional matrix is ​​determined, convolution feature extraction is performed on the input layer of the two-dimensional convolutional neural network; S4. Classify and sort out the brake valve faults, other similar brake valve applications and off-site fault cause analysis conclusions through the transfer learning model for parameter adjustment of weight calculation incentive function, design the influencing factor weight calculation incentive function and substitute the parameters generated by the transfer model into the incentive function to calculate the weight of the influencing factor; S5. Construct a weight feature pooling layer to extract the influencing effect features and form corresponding numerical analysis weights; S6. The overall performance of the brake valve is used to form a judgment on the brake valve maintenance quality and a diagnosis of application prediction, completing the comprehensive inspection of the brake valve.

2. The CNN-based vehicle brake valve detection method according to claim 1, characterized in that: The method for constructing the two-dimensional convolutional neural network input layer in S1 is as follows: in the two-dimensional matrix, the row definitions are classified according to the brake valve detection data, namely leakage amount Boolean type, leakage numerical type, leakage numerical type, pressure numerical type, and time numerical type; the columns are defined as test items, namely brake valve inflation position, emergency brake position, brake and relief sensitivity, local reduction valve, stability, emergency boost, full relief position, and balanced sensitivity; a 4X8 two-dimensional matrix is ​​constructed, and the data in the two-dimensional matrix are the values ​​converted from the original test records, and the items not involved are filled with 0, and the Boolean type is filled with 0 or 1 in the test record.

3. The CNN-based vehicle brake valve detection method according to claim 1, characterized in that: The method for constructing the test numerical factor convolution kernel in S2 is: according to the maintenance standard, the leakage deviation coefficient m and the time deviation coefficient n are set, and the convolution kernel sets the numerical interval feature to extract a 2X2 two-dimensional matrix, which respectively stores the maximum and minimum values ​​of the leakage standard and the maximum and minimum values ​​of the time standard; by multiplying the deviation coefficient corresponding to the input two-dimensional matrix data type, the interval exceeding the dynamic convolution kernel range is counted, wherein the time quantity uses the range median technology, and the deviation coefficient less than the median takes the negative absolute value to participate in the operation, and is extracted into a 2x2 two-dimensional matrix characterized by a counter, which is recorded as the leakage exceeding limit count and the time quantity exceeding limit count respectively.

4. The CNN-based vehicle brake valve detection method according to claim 1, characterized in that: The method for constructing the convolution kernel of the station environment factors in S2 is: according to the influence of the temperature and air source pressure of the brake valve test station on the test data results, the dynamic convolution kernel is set with the total air source of the station and the standard test air source pressure value; since the temperature and pressure mainly affect the time value, the extraction feature is a 2x2 two-dimensional matrix of the trend feature of the time performance, which is extracted as a 2x2 two-dimensional matrix characterized by a counter, and the deviation trend counts are recorded respectively.

5. The CNN-based vehicle brake valve detection method according to claim 1, characterized in that: The method for constructing the convolution kernel of the key component measurement data factor in S2 is: according to the deviation between the measurement data of the components of the brake valve and the standard value, a 2x2 two-dimensional matrix of the deviation coefficient affecting the feature extraction is set, and the values ​​of the two-dimensional matrix are the brake valve spring test deviation coefficient, the rubber template deviation coefficient, the sliding valve roughness deviation coefficient, and the sliding valve stop valve wear deviation coefficient. The 2x2 two-dimensional matrix of the feature extraction of the influence of the component test data deviation on the brake valve data is extracted, and the influence degree values ​​are recorded respectively.

6. The CNN-based vehicle brake valve detection method according to claim 1, characterized in that: The method for constructing the convolution kernel of the environmental factors in the application interval in S2 is: according to the temperature and humidity difference in the application interval and the application time point, a 2x3 two-dimensional matrix is ​​set to extract features, corresponding to the highest and lowest values ​​of the temperature and humidity difference and the estimated temperature and humidity values ​​at the operating time point, and extracted as a 2x3 two-dimensional matrix with a counter as the feature, and the deviation trend counts are recorded respectively.

7. The CNN-based vehicle brake valve detection method according to claim 1, characterized in that: The method for constructing the convolution kernel of the personnel skill factor in S2 is: setting a feature extraction 2x2 two-dimensional matrix according to the personnel skill level, rework rate, key maintenance failure rate, and working years, extracting the feature extraction 2x2 two-dimensional matrix of the influence of the personnel skill factor on the brake valve data, and recording the influence degree values ​​respectively.

8. The CNN-based vehicle brake valve detection method according to claim 1, characterized in that: The method for calculating the weight of the influencing factor in S4 by designing the influencing factor weight calculation activation function and substituting the migration model generation parameters into the activation function is as follows: S4.

1. Classify vehicle control valve faults into single vehicle faults, centralized wind test faults, and driving operation faults; then classify the main causes of the faults into a secondary classification; then classify the secondary classification abnormal causes again to form a third-level classification of component factors, such as templates, springs, slide valves, stop valves, mounting seats, and apertures; extract the fault probability m corresponding to the proportion of faults caused by the three-level classification components, and dynamically increase the probability value of the same type of brake valves with similar working principles based on the probability n, and take the median (mn) / 2 of the increased part as the fault pre-judgment; S4.

2. Classify the operational failures caused by temperature and humidity, with the number of events n and the total number of failures m, and calculate the probability n / m of failures caused by temperature and humidity. According to the open parabola relationship between temperature and humidity in the working process of the brake valve, calculate the value based on the temperature difference and humidity difference at the application site, and then multiply it by the probability to calculate the prediction probability factor. S4.

3. Classify the operational failures caused by personnel skills, the number of events is n, the total number of failures is m, and the probability of failure caused by personnel functional level is calculated as n / m. Based on the negative correlation between brake valve failure and personnel experience, the repair rate and working years are calculated and multiplied by the probability to calculate the prediction probability factor.

9. The CNN-based vehicle brake valve detection method according to claim 1, characterized in that: The method for constructing the weighted feature pooling layer in S5 is: the numerical sum of the test data convolution operation result S=∑i=1m∑j=1naij result and the deviation calculation f(n)=(a ij -(an 2 +bn-c)) 2 , find the derivative of f(n) and get a and b, and use the numerical weight activation function f(x) = ax 2 + bx + c, and the weight value A is obtained; the sum of the convolution operation values ​​of the station environmental factors S = ∑i = 1m∑j = 1naij and the deviation between the result and the expected numerical parameter n is calculated f(n) = (a ij -(an 2 +bn-c)) 2 , find the derivative of f(n) and get a and b, and use the numerical weight activation function f(x) = ax 2 + bx + c, and the weight value B is obtained; the sum of the convolution operation values ​​in the running interval S = ∑i = 1m∑j = 1naij and the deviation between the result and the expected numerical parameter n is calculated f(n) = (a ij -(an 2 +bn-c)) 2 , find the derivative of f(n) and get a and b, and use the numerical weight activation function f(x) = ax 2 +bx+c, and the weight value C is obtained; after the key component factors are calculated, the weight value D is obtained through the activation function f(K)=K*∑i=1m∑j=1naij; The personnel skill factor is obtained through the incentive function f(K)=K*∑i=1m∑j=1naij to obtain the weight value E.

10. The CNN-based vehicle brake valve detection method according to claim 1, characterized in that: The method for forming a judgment on the brake valve maintenance quality and a diagnosis of application prediction for the overall performance of the brake valve in S6 is: data normalization is performed according to the weight influencing factors to form a comprehensive weight M=A+B*a+C*b+D*c+E, wherein the a coefficient is mainly the influence of the off-site wind source pressure and temperature difference on the failure of the concentrated part of the wind test, the b coefficient is the influence of the temperature and humidity in the application range on the application failure, and c is the influence of human factors on the failure of a single vehicle, off-site, and application range. The corresponding fault diagnosis is classified according to the transfer learning model, and finally the overall performance of the brake valve is formed to form a judgment on the brake valve maintenance quality and a diagnosis of application prediction to complete the comprehensive detection of the brake valve.

Images