A method and system for combined early warning of bus brake risk

Abstract

The application relates to the technical field of brake systems, solves the technical problems that the prior art is susceptible to environmental interference, is not high in accuracy of multi-factor analysis, and is low in early warning accuracy, in particular to a passenger and freight vehicle brake risk combined early warning method and system, the method comprises the following steps: S1, collecting original data in a passenger and freight vehicle, and obtaining a dynamic calibration value by preprocessing the original data; S2, calculating a mixed filtering value according to the dynamic calibration value, and calculating a feature enhancement value for realizing feature decoupling based on the mixed filtering value, the application can clearly and explicitly express fault logic through a causal diagram model, can accurately determine a fault position in combination with a Bayesian network model, can display a fault path, can provide clear and accurate analysis basis for maintenance personnel to trace decisions, and can vividly and visually display the type of the fault in combination with AR image display, and the maintenance efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of braking system technology, and in particular to a joint early warning method and system for braking risks in passenger and freight vehicles. Background Technology

[0002] The Braking Risk Joint Warning System is a comprehensive safety system integrating multiple sensor technologies, data processing algorithms, and communication modules. It monitors the vehicle's driving status, driver behavior, and external environmental information in real time, using advanced algorithm models to fuse and analyze this multi-source heterogeneous data, thereby predicting potential braking risks in advance. Existing technologies typically use single or a limited number of sensors to monitor vehicle and driver information, such as relying on ABS sensors to obtain vehicle speed and wheel speed data, or using steering wheel angle sensors to assist in judging driving intentions. Data processing methods often employ simple threshold-based judgment logic, such as triggering a warning when the rate of change between vehicle speed and brake pedal opening exceeds a preset value. However, this method is susceptible to environmental interference; for example, severe weather can affect sensor accuracy, and it cannot fully reflect complex driving scenarios. Furthermore, threshold algorithms lack the ability to analyze the correlation between multiple factors, resulting in low accuracy in fault analysis and consequently, low warning accuracy, making it difficult to meet the complex and ever-changing real-world driving needs. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a joint early warning method and system for braking risks in passenger and freight vehicles. It solves the technical problems of existing technologies being susceptible to environmental interference, having low accuracy in multi-factor analysis, and having a low early warning accuracy rate. This invention achieves the goal of avoiding environmental interference, improving accuracy through multi-factor analysis, and significantly improving the early warning accuracy rate.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a joint early warning method for braking risks in passenger and freight vehicles, the method comprising the following steps:

[0005] S1. Collect raw data from inside passenger and freight vehicles, and preprocess the raw data to obtain dynamic calibration values. ;

[0006] S2, Based on dynamic calibration values Calculate the hybrid filter value And based on the mixed filter value Calculate the feature enhancement value used to achieve feature decoupling ;

[0007] S3, Based on Feature Enhancement Values The input set is obtained by classifying the categories. For the input set Gas path node temperature Humidity of gas path nodes Perform calibration and obtain calibration dataset ;

[0008] S4. Based on the calibration dataset Constructing a method to obtain thermal coupling values and wet coupling value Digital twin dynamic model And based on thermal coupling value and wet coupling value Calculate the pre-feature set used to provide input data for fault diagnosis. ;

[0009] S5. Based on the pre-feature set Construct a cause-effect graph of brake air circuit failure, calculate the judgment probability based on the cause-effect graph, and obtain the primary failure result and secondary abnormal result based on the judgment probability.

[0010] S6. Calculate the pump path for judging the pump failure result based on the cause-effect graph of brake air circuit failure. and drying cylinder path An AR animation is generated based on the fault path results.

[0011] Furthermore, in step S1, the specific implementation steps are as follows:

[0012] S11. Obtain the number N of multiple sensors inside the passenger and freight vehicles, and get the raw data with dimension M and time series length t. ;

[0013] S12. Based on the highest frequency sampling rate of multiple sensors inside the passenger and freight vehicles. The dynamic adjustment bandwidth value h is calculated using the following formula:

[0014]

[0015] in, Indicates the sensor's response period;

[0016] S13. Based on the original data inside the passenger and freight vehicles. Sampling time Calculate the time difference weight matrix of the original data The calculation formula is:

[0017]

[0018] in, Representation and sampling time The corresponding synchronization time;

[0019] S14. Based on the time difference weight matrix Calculating dynamic calibration values The calculation formula is as follows:

[0020]

[0021] Wherein, represents the compensation value of the ambient pressure , and represents the compensation value of the air pump rotating speed .

[0022] Further, in step S2, the specific implementation steps are as follows:

[0023] S21, forming a data sequence with multiple dynamic calibration values , constructing a rectangular window on the data sequence, and calculating the filter weight value in the rectangular window The calculation formula is as follows:

[0024]

[0025] Wherein, represents the standard deviation of the data sequence, k represents the window offset, represents the window width parameter;

[0026] S22, calculating the hybrid filter value for filtering the data sequence according to the filter weight value The calculation formula is as follows:

[0027]

[0028] Wherein, represents the dynamic calibration value after the window moves k;

[0029] S23, calculating the four-dimensional tensor according to the hybrid filter value The calculation formula is as follows:

[0030]

[0031] Wherein, S represents the voiceprint spectrum feature corresponding to the hybrid filter value .

[0032] S24, performing tensor decomposition on the four-dimensional tensor to obtain multiple principal component values and The calculation formula is as follows:

[0033]

[0034] Wherein, and ​respectively represent principal component values in the first r component values, represents an outer product operation, represent the decomposition weight corresponding to each component value;

[0035] S25, based on the plurality of component values calculating feature enhancement values for realizing feature decoupling , the calculation formula is:

[0036]

[0037] wherein, represents a Kronecker product.

[0038] Further, in step S3, the specific implementation steps are as follows:

[0039] S31, the feature enhancement values generating an input set according to the type of original data , the expression is:

[0040]

[0041] wherein, represent the key air path node temperature of the brake system after feature enhancement, represent the key air path node humidity of the brake system after feature enhancement, represent the water content of the drying cylinder air path after feature enhancement, represent the noise spectrum of the drying cylinder after feature enhancement, represent the air pump speed after feature enhancement, represent the environmental pressure after feature enhancement;

[0042] S32, according to the input set calculating temperature parameter compensation values and humidity parameter compensation values , the calculation formula is:

[0043]

[0044]

[0045] wherein, represent the rated speed of the air pump, represent the standard atmospheric pressure of the environment, respectively represent the coupling compensation coefficients of the air pump speed and the environmental pressure on temperature and humidity;

[0046] S33, repeating step S32 to obtain all temperature parameter compensation values and humidity parameter compensation values The calculation is completed, and a compensation data set is generated wherein, ;

[0047] S34, based on the temperature parameter compensation value Calculate the temperature standard deviation value of each temperature sensor , the calculation formula is:

[0048]

[0049] wherein, S represents the temperature parameter compensation value The number of, indicates the average value of the temperature of the plurality of sensors;

[0050] S35, based on the humidity parameter compensation value Calculate the humidity standard deviation value of each humidity sensor , the calculation formula is:

[0051]

[0052] wherein, J represents the humidity parameter compensation value The number of, indicates the average value of the humidity of the plurality of sensors;

[0053] S36, according to the temperature standard deviation value Drift correction is performed on the temperature sensor;

[0054] If , the temperature sensor has data drift, and drift correction is performed on the temperature sensor;

[0055] If , the temperature sensor is normal and ends, and a corrected temperature set is obtained ;

[0056] S37, according to the humidity standard deviation value Drift correction is performed on the humidity sensor;

[0057] If , the humidity sensor has data drift, and drift correction is performed on the humidity sensor;

[0058] If , the temperature sensor is normal and ends, and a corrected humidity set is obtained ;

[0059] S38, the elements in the corrected humidity set and the water content in the drying cylinder gas circuit are fused to obtain a humidity fusion value , the calculation formula is:

[0060]

[0061]

[0062] wherein, and respectively represent fusion weight values;

[0063] S39, based on the correction temperature set and the correction humidity set get the calibration data set .

[0064] Further, in step S4, the specific implementation steps are as follows:

[0065] S41, according to the drying cylinder noise spectrum in the calibration data set calculate the time domain feature value for extracting the time domain feature , the calculation formula is:

[0066]

[0067]

[0068] wherein, the time domain feature peak value, N represents the number of drying cylinder noise spectrum ;

[0069] S42, according to the drying cylinder noise spectrum in the calibration data set calculate the frequency domain feature value for extracting the frequency domain feature , the calculation formula is:

[0070]

[0071] wherein, FFT represents fast Fourier transform;

[0072] S43, according to the humidity fusion value and the temperature parameter compensation value respectively calculate the temperature gradient value and the humidity gradient value , the calculation formula is:

[0073]

[0074]

[0075] ​​​​​wherein, and represent the first and the first independent variables, respectively;

[0076] S44, the temperature gradient value and the humidity gradient value are composed into a gradient matrix ;

[0077] S45, a digital twin dynamic model is constructed according to the temperature gradient value and the humidity gradient value , and the expression is:

[0078]

[0079] wherein, represents the net outflow of fluid heat flow density at the sensor position, D represents the humidity diffusion coefficient, and v represents the convection velocity of the fluid, represents the heat generation power of the air pump, represents the dehumidification rate of the drying cylinder, represents the thermal coupling value, represents the humidity coupling value;

[0080] S46, historical data of the sensor is obtained, and temperature residual threshold value and humidity residual threshold value of the historical data are calculated, and the calculation formula is:

[0081]

[0082]

[0083] wherein, respectively represent the residual mean of the temperature and humidity of the historical data, represents the confidence coefficient, respectively represent the residual standard deviation of the temperature and humidity of the historical data;

[0084] S47, the temperature abnormal node is marked according to the temperature residual threshold value ;

[0085] If , the node corresponding to the temperature data is a temperature abnormal node and is marked;

[0086] If , the node corresponding to the temperature data is a temperature normal node and ends;

[0087] S48, the humidity abnormal node is marked according to the humidity residual threshold value ​Mark the humidity abnormal node, and combine the temperature abnormal node and the humidity abnormal node into an abnormal node set B;

[0088] If , the node corresponding to the humidity data is a humidity abnormal node and is marked;

[0089] If , the node corresponding to the humidity data is a humidity normal node and ends;

[0090] S49, combine the voiceprint feature vector , the gradient matrix , the thermal coupling value , the humidity coupling value , and the abnormal node set B into a pre-feature set .

[0091] Further, in step S5, the following steps are specifically implemented:

[0092] S51, respectively determine the air pump node, the drying cylinder node, and the key air path position node according to the pre-feature set , construct a brake air path fault causal diagram based on the air pump node, the drying cylinder node, and the key air path position node, and construct a propagation path between the key air path position node and the air pump node and the drying cylinder node;

[0093] S52, obtain the cumulative working time of the air pump, and calculate the fault prior probability of the air pump node according to the cumulative working time , and the calculation formula is:

[0094]

[0095] wherein, and respectively represent the air pump shape parameters;

[0096] S53, obtain the time domain feature value in the voiceprint feature vector , and calculate the cylinder barrier prior probability of the drying cylinder node according to the time domain feature value , and the calculation formula is:

[0097]

[0098] wherein, represents the standard deviation of the time domain feature value , and represents the average value of the time domain feature value ;

[0099] S54, according to the fault prior probability and the cartridge prior probability calculate the judgment probability, the judgment probability including the post-pump probability of the air pump and the post-cartridge probability of the drying cartridge ;

[0100] S55, according to the post-pump probability and the post-cartridge probability judge the fault and generate the primary fault result, the primary fault result including the air pump fault information and the drying cartridge fault information;

[0101] if , the air pump fault information is generated;

[0102] if , the drying cartridge fault information is generated;

[0103] S56, according to the humidity gradient value calculate the humidity gradient threshold value , the calculation formula being:

[0104]

[0105] wherein M represents the total number of the humidity gradient value ;

[0106] S57, according to the humidity gradient threshold value judge the air path fault and generate the secondary abnormal result;

[0107] if , the air path has the risk of water accumulation, the secondary abnormal result is generated and the process ends;

[0108] if , the air path is normal and the process ends.

[0109] Further, the calculation formula of the post-pump probability and the post-cartridge probability is:

[0110]

[0111]

[0112] wherein, represents the likelihood probability of the air pump rotating speed, represents the likelihood probability of the temperature gradient, represents the joint probability of the air pump rotating speed and the temperature gradient, represents the likelihood probability of the time domain feature, represents the likelihood probability of the humidity gradient, A joint probability of a time domain feature and a humidity gradient.

[0113] Further, in step S6, the implementation steps are as follows:

[0114] S61, define the air pump node as A node, define the drying cylinder node as B node, define the key air path position node as C node, and calculate the air pump path and the drying cylinder path respectively.

[0115] S62, according to the air pump path and the drying cylinder path determine the most likely fault path and generate a fault path result, wherein the fault path result includes a pump air fault result and a cylinder air fault result;

[0116] If , the air pump path is the fault path, a pump air fault result is generated and the process ends.

[0117] If , the drying cylinder path is the fault path, a cylinder air fault result is generated and the process ends.

[0118] S63, send the primary fault result and the secondary abnormal result to the early warning center, generate an AR three-dimensional disassembly animation according to the fault path result, and highlight the fault path and the fault component in the early warning system.

[0119] Further, the calculation formulas of the air pump path and the drying cylinder path are as follows:

[0120]

[0121]

[0122] Wherein, F and G represent the number of AC paths and BC paths respectively, represents the probability that node A fails and node C also fails, represents the probability that node B fails and node C also fails.

[0123] The technical scheme also provides a system for the above-mentioned passenger and freight car brake risk joint early warning method, which comprises:

[0124] A preprocessing module for collecting original data inside the passenger and freight car and preprocessing the original data to obtain a dynamic calibration value .

[0125] A feature enhancement module for obtaining a dynamic calibration value Calculate a mixed filter value , and calculate a feature enhancement value for realizing feature decoupling based on the mixed filter value ;

[0126] A calibration module is configured to perform category division based on the feature enhancement value to obtain an input set , calibrate the air path node temperature and the air path node humidity in the input set , and obtain a calibration data set for making the data more accurate ;

[0127] A digital twin module is configured to construct a digital twin dynamic model for obtaining a thermal coupling value and a humidity coupling value based on the calibration data set , and calculate a pre-feature set for providing a reference for fault judgment based on the thermal coupling value and the humidity coupling value ; A fault causality module is configured to construct a brake air path fault causality graph based on the pre-feature set

[0128] , calculate a judgment probability based on the brake air path fault causality graph, and obtain a primary fault result and a secondary abnormal result based on the judgment probability ;

[0129] An AR generation module is configured to calculate an air pump path and a drying cylinder path for judging the pump air fault result based on the brake air path fault causality graph, and generate an AR animation according to the fault path result ;

[0130] By means of the above technical solutions, the present application provides a passenger and freight car brake risk joint early warning method and system, which has at least the following beneficial effects:

[0131] 1. The present application can solve the problem of time asynchronous of multi-source sensor data, dynamically compensate environmental interference, greatly reduce calibration error, and greatly improve signal-to-noise ratio, retain the fluctuation characteristics of the sensor while suppressing the high-frequency noise of the microwave sensor, compress the feature dimension while retaining most of the information, and provide structured input for the subsequent causality graph.

[0132] ​2、The application can cover the full-dimensional state perception of the air path, can greatly improve the detection range in practice, can greatly improve the timeliness of fault prediction through millisecond-level updating of the digital twin model, can greatly reduce the temperature and humidity measurement error through the self-calibration compensation model of the calibration data set, and improves the data accuracy.

[0133] 3、The application can clearly and accurately express the fault logic through the construction of the causal diagram model, can accurately determine the fault location, and can display the fault path, provide clear and accurate analysis basis for the maintenance personnel to trace the decision, and can display the type of fault vividly and improve the maintenance efficiency in combination with the AR image display. BRIEF DESCRIPTION OF DRAWINGS

[0134] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application, the illustrative embodiments of the present application and the description thereof are used to explain the present application, and do not constitute improper limitation on the present application. In the drawings:

[0135] Figure 1 The flow chart of the application is a passenger car brake risk joint early warning method;

[0136] Figure 2 The structural block diagram of the application is a passenger car brake risk joint early warning system.

[0137] In the figure: 1, pretreatment module; 2, feature enhancement module; 3, calibration module; 4, digital twin module; 5, fault causal module; 6, AR generation module. DETAILED DESCRIPTION

[0138] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the application will be further described in detail below in combination with the drawings and specific embodiments. The realization process of how to apply technical means to solve technical problems and achieve technical effects of the application can be fully understood and implemented.

[0139] Due to the technical problems that the prior art is susceptible to environmental interference, the accuracy of multi-factor analysis is not high, and the early warning accuracy is low, the embodiment provides a passenger car brake risk joint early warning method, as shown in Figure 1 The method can avoid environmental interference, improve accuracy through multi-factor analysis, and greatly improve early warning accuracy. The method comprises the following steps:

[0140] S1, collecting the original data in the passenger car and pre-processing the original data to obtain dynamic calibration value During the acquisition of raw data, since the sensors of passenger and freight vehicles are located in the external environment, noise and vibration during vehicle operation can easily lead to inaccurate raw data. To solve this problem, the specific methods adopted are as follows:

[0141] S11. Obtain the number N of multiple sensors inside the passenger and freight vehicles, and get the raw data with dimension M and time series length t. The raw data includes the temperatures of key air passage nodes in the braking system. and humidity Moisture content in the air path of the drying cylinder Noise spectrum of drying cylinder Air pump speed and environmental pressure ;

[0142] S12. Based on the highest frequency sampling rate of multiple sensors inside the passenger and freight vehicles. The dynamic adjustment bandwidth value h is calculated using the following formula:

[0143]

[0144] in, Indicates the sensor's response period;

[0145] S13. Based on the original data inside the passenger and freight vehicles. Sampling time Calculate the time difference weight matrix of the original data The calculation formula is:

[0146]

[0147] in, Representation and sampling time The corresponding synchronization time; the synchronization time and sampling time here will deviate, affecting the real-time performance of the data. Therefore, inaccuracies can easily occur when calculating the data. To avoid this, the time difference weight matrix is ​​calculated based on this deviation. To reduce the magnitude of the deviation.

[0148] S14. Based on the time difference weight matrix Calculate dynamic calibration values The calculation formula is:

[0149]

[0150] in, Indicates environmental pressure The compensation value, Indicates the air pump speed compensation value; the compensation value can be calculated by an offline calibration experiment to obtain a compensation function, which is a commonly used method for obtaining a compensation function, and will not be described here. Through calculation of the time difference weight matrix of the original data, the time asynchronous problem of multi-source sensor data can be solved, environmental interference can be dynamically compensated, calibration error can be greatly reduced, signal-to-noise ratio can be greatly improved, the fluctuation characteristics of the sensor can be retained while the high-frequency noise of the microwave sensor is suppressed, the feature dimension is compressed, most of the information is retained, and a structured input is provided for the subsequent causal graph.

[0151] S2, calculating a dynamic calibration value according to the dynamic calibration value calculating a mixed filter value , and based on the mixed filter value calculating a feature enhancement value for realizing feature decoupling After preprocessing the original data, the features of the data are still not obvious. In order to solve this problem, the embodiment proposes a more detailed implementation method, as follows:

[0152] S21, calculating a plurality of dynamic calibration values forming a data sequence, constructing a rectangular window on the data sequence, and calculating a filter weight value in the rectangular window , the calculation formula is:

[0153]

[0154] wherein, denotes the standard deviation of the data sequence, k denotes the window offset, denotes the window width parameter; the window width parameter is obtained by taking the integer of the fundamental frequency ratio in the data, and the standard deviation and the fundamental frequency ratio taking integer method are commonly used data processing methods, which will not be described here.

[0155] S22, calculating a mixed filter value for filtering the data sequence according to the filter weight value , the calculation formula is:

[0156]

[0157] wherein, denotes the dynamic calibration value after the window moves k;

[0158] S23, calculating a four-dimensional tensor according to the mixed filter value , the calculation formula is:

[0159]

[0160] wherein, S denotes the mixed filter value ​​The corresponding acoustic signature spectrum characteristics; acoustic signature spectrum characteristics can be obtained by analyzing the noise spectrum of the drying cylinder. Video transformation is performed to obtain the data.

[0161] S24, Using four-dimensional tensors Tensor decomposition yields multiple principal component values. and The calculation formula is:

[0162]

[0163] in, and These represent the principal component values ​​among the first r component values. This represents the outer product operation. This represents the decomposition weight corresponding to each component value; the decomposition weight can be obtained by combining alternating least squares with alternating optimization steps. Least squares and alternating optimization are common methods for obtaining decomposition weights. r can be obtained through a four-dimensional tensor. The variance contribution rate is determined to be greater than or equal to 95%. Variance contribution rate is a common data processing method, which will not be elaborated here.

[0164] S25, based on multiple component values Calculate the feature enhancement value used to achieve feature decoupling The calculation formula is:

[0165]

[0166] in, The Kronecker product, through a spatiotemporal-physical joint calibration mechanism, eliminates the bias of multi-source heterogeneous data. At the same time, through modal sensing filtering technology, it takes into account both nanometer-level sensitivity and fast response requirements. It can also achieve feature decoupling through tensor fusion architecture, improving the fault feature separability by 32% compared with traditional PCA methods.

[0167] S3, Based on Feature Enhancement Values The input set is obtained by classifying the categories. For the input set Gas path node temperature Humidity of gas path nodes Perform calibration and obtain calibration dataset Based on step S2, further data calibration is needed to ensure greater accuracy in subsequent steps. The detailed implementation steps are as follows:

[0168] S31. Enhance the feature value Generate an input set based on the type of the original data. The expression is:

[0169]

[0170] in, This indicates the temperature of key air path nodes in the enhanced braking system. This indicates the humidity of key air path nodes in the enhanced braking system. Indicates the moisture content of the drying cylinder gas path after feature enhancement. The noise spectrum of the drying cylinder after feature enhancement is represented. This indicates the pump speed after feature enhancement. This indicates the environmental pressure after feature enhancement;

[0171] S32, Based on the input set Calculate temperature parameter compensation value Humidity parameter compensation value The calculation formula is:

[0172]

[0173]

[0174] in, Indicates the rated speed of the air pump. Indicates the environmental standard atmospheric pressure. These represent the coupling compensation coefficients of the air pump speed and ambient pressure with respect to temperature and humidity, respectively. The coupling compensation coefficients can be obtained through the step response method. The step response method is a method that applies a step input to the system, measures the output response, and inversely calculates the compensation coefficients by the amplitude or phase of the coupling interference. The step response method is a commonly used method for obtaining coupling compensation coefficients, which will not be elaborated here.

[0175] S33. Repeat step S32 to compensate all temperature parameters. Humidity parameter compensation value The calculation is complete, and the compensation dataset is generated. ,in, ;

[0176] S34, Based on temperature parameter compensation value Calculate the temperature standard deviation for each temperature sensor. The calculation formula is:

[0177]

[0178] Where S represents the temperature parameter compensation value Quantity, This represents the average temperature from multiple sensors.

[0179] S35, Based on humidity parameter compensation value Calculate the humidity standard deviation value of each humidity sensor , the calculation formula is:

[0180]

[0181] Wherein, J represents the humidity parameter compensation value The number of Indicates the average value of the humidity of multiple sensors;

[0182] S36, according to the temperature standard deviation value Drift correction is performed on the temperature sensor;

[0183] If , the temperature sensor has data drift, and drift correction is performed on the temperature sensor;

[0184] If , the temperature sensor is normal and ends, and the corrected temperature set is obtained ;

[0185] S37, according to the humidity standard deviation value Drift correction is performed on the humidity sensor;

[0186] If , the humidity sensor has data drift, and drift correction is performed on the humidity sensor;

[0187] If , the temperature sensor is normal and ends, and the corrected humidity set is obtained ;

[0188] S38, the elements in the corrected humidity set And the water content in the drying cylinder gas circuit Characteristic fusion is carried out to obtain the humidity fusion value , the calculation formula is:

[0189]

[0190]

[0191] Wherein, And Respectively represent the fusion weight value; the fusion weight value is dynamically optimized by Kalman filtering, which is a common weight acquisition method, and will not be repeated here.

[0192] S39, based on the corrected temperature set And the corrected humidity set Obtain the calibration data set , wherein, ​, through the cooperation of temperature and humidity sensors, microwave humidity detection and system voiceprint analysis, it can achieve full-dimensional state perception of air path, which can greatly improve the detection range in practice, through millisecond-level update of digital twin model, it can greatly improve the timeliness of fault prediction, through self-calibration compensation model of calibration data set, the measurement error of temperature and humidity is greatly reduced, and the data accuracy is improved.

[0193] S4、according to the calibration data set constructing a digital twin dynamic model for obtaining thermal coupling values and wet coupling values , and calculating a pre-feature set based on the thermal coupling values and the wet coupling values ; due to the volatility of data and external interference, there are still some abnormal data, in order to solve this problem, the detailed implementation steps are as follows:

[0194] S41、according to the dry cylinder noise spectrum in the calibration data set calculate the time domain feature value for extracting time domain features , the calculation formula is:

[0195]

[0196]

[0197] wherein, the time domain feature peak value, N represents the number of dry cylinder noise spectrum ; and

[0198] S42、according to the dry cylinder noise spectrum in the calibration data set calculate the frequency domain feature value for extracting frequency domain features , the time domain feature value and the frequency domain feature value are synthesized into a voiceprint feature vector , the calculation formula is:

[0199]

[0200] wherein, FFT represents fast Fourier transform; fast Fourier transform is a kind of efficient algorithm, which is used for converting time domain signal into frequency domain signal to obtain its frequency domain feature value, and fast Fourier transform is a common frequency domain feature acquisition method, which will not be described here.

[0201] S43、according to the humidity fusion value and the temperature parameter compensation value​​​​​ respectively and humidity gradient values , the calculation formula is:

[0202]

[0203]

[0204] wherein, and respectively represent the first and the first independent variables;

[0205] S44, the temperature gradient value and the humidity gradient value are composed into a gradient matrix , wherein, ;

[0206] S45, according to the temperature gradient value and the humidity gradient value , a digital twin dynamic model is constructed , the expression is:

[0207]

[0208] wherein, represents the net outflow of fluid heat flux at the sensor position, D represents the humidity diffusion coefficient, v represents the convection velocity of the fluid, represents the heat generation power of the air pump, represents the drying cylinder dehumidification rate, represents the thermal coupling value, represents the wet coupling value; the humidity diffusion coefficient, the convection velocity of the fluid, the heat generation power of the air pump and the drying cylinder dehumidification rate can be obtained by the sensor, the boundary conditions are set according to the air inlet and outlet of the air path, and in actual application, the rated value of temperature and humidity at the air inlet is generally set as the boundary condition, and the boundary condition at the air outlet is that the partial derivative of the ratio of temperature and humidity to the boundary vector at the air outlet is equal to 0.

[0209] S46, the historical data of the sensor is obtained and the temperature residual threshold value and the humidity residual threshold value of the historical data are calculated, the calculation formula is:

[0210]

[0211]

[0212] wherein, respectively represent the residual mean of temperature and humidity of the historical data, confidence coefficient, respectively represent the residual standard deviation of temperature and humidity of historical data; in practice, the confidence coefficient is generally set to between the numbers, corresponding to the confidence interval of 95%-99.7%, the calculation of residual mean and residual standard deviation is a commonly used calculation method, which is not described here.

[0213] S47, according to the temperature residual threshold mark the temperature abnormal node;

[0214] If , the node corresponding to the temperature data is a temperature abnormal node and is marked;

[0215] If , the node corresponding to the temperature data is a temperature normal node and ends;

[0216] S48, according to the humidity residual threshold mark the humidity abnormal node, and combine the temperature abnormal node and the humidity abnormal node into an abnormal node set B;

[0217] If , the node corresponding to the humidity data is a humidity abnormal node and is marked;

[0218] If , the node corresponding to the humidity data is a humidity normal node and ends;

[0219] S49, combine the voiceprint feature vector , the gradient matrix , the thermal coupling value , the humidity coupling value and the abnormal node set B into a pre-feature set , wherein , through the cooperation of the temperature and humidity sensor, the microwave humidity detection and the system voiceprint analysis, the full-dimensional state perception of the air path can be achieved, which can greatly improve the detection range in practice. Through the millisecond-level update of the digital twin model, the timeliness of fault prediction can be greatly improved. Through the self-calibration compensation model of the calibration data set, the temperature and humidity measurement error is greatly reduced, and the data accuracy is improved.

[0220] S5, according to the pre-feature set , a brake air path fault causal diagram is constructed, a judgment probability is calculated based on the brake air path fault causal diagram, and a primary fault result and a secondary abnormal result are obtained based on the judgment probability, wherein the judgment probability includes a pump probability , a cylinder probability and a humidity gradient threshold ; in the pre-feature set Based on this, it is necessary to quickly identify the type and path of the fault through a cause-and-effect diagram of the brake air circuit fault. To solve this problem, this embodiment proposes a more detailed implementation method, as follows:

[0221] S51, Increase the air pump speed Heating power of air pump As the air pump node in the cause-and-effect graph of brake air circuit failure, the soundprint feature vector frequency domain eigenvalues and dehumidification rate of drying drum As the dryer node in the cause-and-effect graph of brake air circuit failure, the pre-feature set gradient matrix in Thermal coupling value and wet coupling value As key air path location nodes, a cause-effect graph of brake air path failure is constructed based on the air pump node, dryer cylinder node, and key air path location nodes, with the connections between nodes representing the propagation paths. ;

[0222] S52. Obtain the cumulative operating time of the air pump. Based on cumulative working hours Calculate the prior probability of failure of the air pump node. The calculation formula is:

[0223]

[0224] in, and These represent the shape parameters of the air pump. The shape parameters of the air pump can be obtained by the staff through direct measurement, which is a common method for obtaining shape parameters and will not be elaborated here.

[0225] S53. Obtain the voiceprint feature vector Temporal eigenvalues Based on time domain eigenvalues Calculate the prior probability of the obstacle at the dryer drum node. The calculation formula is:

[0226]

[0227] in, Representing time-domain eigenvalues standard deviation Representing time-domain eigenvalues The average value;

[0228] S54. Based on the prior probability of failure Prior probability of the tube barrier Calculate the post-pump probability of the air pump separately. and the barrel posterior probability of the drying barrel , the calculation formula is:

[0229]

[0230]

[0231] wherein, represents the likelihood probability of the air pump rotation speed, represents the likelihood probability of the temperature gradient, represents the joint probability of the air pump rotation speed and the temperature gradient, represents the likelihood probability of the time domain feature, represents the likelihood probability of the humidity gradient, represents the joint probability of the time domain feature and the humidity gradient; the likelihood probability and the joint probability are common methods for calculating the posterior probability, and will not be described here.

[0232] S55, according to the pump posterior probability and the barrel posterior probability judging the fault and generating the primary fault result, wherein the primary fault result includes air pump fault information and drying barrel fault information;

[0233] if , the air pump fault information is generated;

[0234] if , the drying barrel fault information is generated;

[0235] S56, according to the humidity gradient value calculating the humidity gradient threshold value , the calculation formula is:

[0236]

[0237] wherein, M represents the total number of humidity gradient values .

[0238] S57, according to the humidity gradient threshold value judging the air path fault and generating the secondary abnormal result;

[0239] if , the air path has the risk of water accumulation, the secondary abnormal result is generated and the process is ended;

[0240] if , the air circuit is normal and ends, through the causal diagram model construction, the fault logic can be clearly and explicitly expressed, combined with the Bayesian network model, the fault position can be accurately determined, and the fault path can be displayed, providing clear and accurate analysis basis for maintenance personnel to trace decision, and combined with AR image display, the type of fault can be vividly and visually displayed, improving the maintenance efficiency.

[0241] S6, calculate the air pump path based on the brake air circuit fault causal diagram and the drying cylinder path , according to the air pump path and the drying cylinder path get the pump fault result and generate AR three-dimensional disassembly animation; on the basis of step S5, after finding the possible equipment of the fault, the path of the fault needs to be determined and the image animation is vividly generated. In order to solve this problem, the specific implementation method of the embodiment is as follows:

[0242] S61, define the air pump node as A node, define the drying cylinder node as B node, define the key air circuit position node as C node, calculate the air pump path and the drying cylinder path , the calculation formula is:

[0243]

[0244]

[0245] Wherein, F and G represent the number of AC path and BC path respectively, represents the probability that node A fails and node C also fails, represents the probability that node B fails and node C also fails; since the path formed in the brake system is generally from the air pump to the air circuit or from the drying cylinder to the air circuit, both paths need to be known and compared in order to display AR.

[0246] S62, according to the air pump path and the drying cylinder path determine the most possible fault path and generate the fault path result, wherein the fault path result includes the pump fault result and the cylinder fault result;

[0247] If , the air pump path is the fault path, the pump fault result is generated and the process ends;

[0248] If , the drying cylinder path is the fault path, the cylinder fault result is generated and the process ends;

[0249] S63. Generate AR 3D disassembly animation based on the fault path results, highlighting the fault path and faulty components within the early warning system. Through the construction of a cause-effect graph model, the fault logic can be clearly expressed. Combined with a Bayesian network model, the fault location can be accurately determined, and the fault path can be displayed, providing clear and accurate analytical basis for maintenance personnel to trace and make decisions. Combined with AR image display, the types of faults can be vividly displayed, improving maintenance efficiency.

[0250] Due to the susceptibility of existing technologies to environmental interference, low accuracy in multi-factor analysis, and low early warning accuracy, this embodiment also proposes a joint early warning system for braking risks in passenger and freight vehicles. This system can avoid environmental interference, improve accuracy through multi-factor analysis, and significantly improve the early warning accuracy rate. Figure 2 As shown, the monitoring system includes a preprocessing module 1, a feature enhancement module 2, a calibration module 3, a digital twin module 4, a fault causation module 5, and an AR generation module 6.

[0251] Preprocessing module 1 is used to collect raw data from inside passenger and freight vehicles and preprocess the raw data to obtain dynamic calibration values. Feature enhancement module 2, used to adjust dynamic calibration values. Calculate the hybrid filter value And based on the mixed filter value Calculate the feature enhancement value used to achieve feature decoupling Calibration module 3, used for feature enhancement values The input set is obtained by classifying the categories. For the input set Gas path node temperature Humidity of gas path nodes Perform calibration and obtain a calibration dataset to make the data more accurate. Digital twin module 4, used to generate a calibration dataset. Constructing a method to obtain thermal coupling values and wet coupling value Digital twin dynamic model And based on thermal coupling value and wet coupling value Calculate the pre-feature set used to provide a reference for fault diagnosis. Fault causality module 5, used to determine the cause and effect of a fault based on a pre-feature set. A cause-effect graph of brake air circuit faults is constructed, and the judgment probability is calculated based on the cause-effect graph. The primary fault result and secondary abnormal result are obtained based on the judgment probability. AR generation module 6 is used to calculate the air pump path for judging the air pump fault result based on the cause-effect graph of brake air circuit faults. and drying cylinder path generate an AR animation according to the fault path result.

[0252] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment methods can be completed by instructing relevant hardware through programs, therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment in the form of combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer usable program codes.

[0253] Each of the embodiments in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the above embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiments.

[0254] The above embodiments have been described in detail, and the principles and implementation manners of the present application are described by applying specific examples; the above embodiment descriptions are only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges will have changes; in conclusion, the content of the specification should not be understood as the limitation of the present application.

Claims

1. A method for combined early warning of bus brake risk, characterized in that, The method comprises the following steps: S1, collect the original data of the interior of the passenger and freight car, and pretreat the original data to obtain a dynamic calibration value ; S2, based on the dynamic calibration value calculating a hybrid filter value and based on the hybrid filter value calculating a feature enhancement value for enabling feature decoupling ; S3, based on feature enhancement value Classifying to obtain input set Calibrating and obtaining calibration data set Calibrating and obtaining calibration data set Calibrating and obtaining calibration data set Calibrating and obtaining calibration data set ; S4. deriving, from the calibration data set a digital twin dynamic model for obtaining thermal coupling values and wet coupling values and calculating, based on the thermal coupling values and the wet coupling values a pre-feature set for providing input data for fault judgment ;​ S5、According to the pre-feature set The brake air path fault causal diagram is constructed, the judgment probability is calculated based on the brake air path fault causal diagram, and the primary fault result and the secondary abnormal result are obtained based on the judgment probability, and the specific implementation steps are as follows: S51, determining the pre-feature set The air pump node, the drying cylinder node and the key air path position node are determined respectively, the brake air path fault causal diagram is constructed based on the air pump node, the drying cylinder node and the key air path position node, and the key air path position node is connected with the air pump node and the drying cylinder node to construct a propagation path ; S52, acquire the accumulated working time of the air pump , according to the accumulated working time calculate the failure prior probability of the air pump node , the calculation formula is: ； wherein and respectively represent the air pump shape parameters; S53, obtain the voiceprint feature vector time domain feature value in the formula , according to the time domain feature value calculate the cylinder barrier prior probability of the drying cylinder node , the calculation formula is: ； wherein denotes the standard deviation of the time-domain feature values denotes the average value of the time-domain feature values ​​ S54, according to the failure prior probability and the drum barrier prior probability calculating the judgment probability, the judgment probability including the post-pump probability of the air pump and the post-drum probability of the drying drum ; representing the air pump rotating speed after feature enhancement, is a temperature gradient value, is a humidity gradient value; S55, determining the post-pump probability and the post-cartridge probability judging the fault and generating a primary fault result, the primary fault result including air pump fault information and drying cartridge fault information; If a gas pump failure message is generated; If a drying drum failure message is generated; S56, according to the humidity gradient value Computing the humidity gradient threshold value The formula is: ； wherein M represents a humidity gradient value the total number of the plurality of the first and second groups; S57、according to the humidity gradient threshold judging the gas path failure and generating the secondary abnormality result; If , the air path has the risk of water accumulation, a secondary abnormal result is generated, and the process ends. If then the gas circuit is normal and the process ends. The post-pump probability And the post-barrel probability The formula for calculating the post-pump probability is: ； ； wherein, represents a likelihood probability of the air pump rotation speed, represents a likelihood probability of the temperature gradient, represents a joint probability of the air pump rotation speed and the temperature gradient, represents a likelihood probability of the time domain feature, represents a likelihood probability of the humidity gradient, represents a joint probability of the time domain feature and the humidity gradient; S6、Based on the brake air path fault cause and effect diagram, the air pump path for judging the pump air fault result is calculated and the drying cylinder path , and generating an AR animation according to the fault path result.

2. The combined warning method according to claim 1, characterized in that, In step S1, the steps are implemented as follows: S11, obtain the number N of the plurality of sensors inside the passenger and freight car, to obtain the original data with the dimension M and the time sequence length t ; S12, highest frequency sampling rate of the plurality of sensors inside the passenger truck A dynamic adjustment bandwidth value h is calculated according to the following formula: ； wherein, denotes the response period of the sensor; S13, calculate the time difference weight matrix of the original data according to the sampling time of the original data of the passenger and freight car , the calculation formula is:​​ ； wherein, represents the sampling time corresponding synchronization time; S14、according to the time difference weight matrix calculating the dynamic calibration value The calculation formula is: ； wherein represents a compensation value for the ambient pressure represents a compensation value for the ambient temperature, represents a compensation value for the gas pump rotation speed represents a compensation value for the gas pump rotation speed.

3. The combined warning method of claim 1, wherein, In step S2, the steps are implemented as follows: S21, a plurality of dynamic calibration values forming a data sequence, constructing a rectangular window on the data sequence, and calculating filter weight values in the rectangular window the calculation formula is: ； wherein, denotes the standard deviation of the data sequence, k denotes a window offset, denotes a window width parameter; S22、According to the filter weight value calculating a hybrid filter value for filtering the data sequence , the calculation formula is: ； wherein, represents the dynamic calibration value after window movement k; S23、According to the mixed filter value Computing a four-dimensional tensor The formula is: ； wherein S represents a mixed filter value corresponding voiceprint spectrum features; S24, the four-dimensional tensor performing tensor decomposition to obtain a plurality of principal component values and , the calculation formula is: ； wherein, respectively represent principal component values among the first r component values, represents an outer product operation, represents a decomposition weight corresponding to each component value; S25, based on the plurality of component values and calculating a feature enhancement value for enabling feature decoupling , the calculation formula being: ； wherein denotes the Kronecker product.

4. The combined warning method of claim 1, wherein, In step S3, the steps are implemented as follows: S31, the feature enhancement value generating input sets according to the kind of original data , the expression is: ； wherein, represents the temperature of the key pneumatic node of the brake system after feature enhancement, represents the humidity of the key pneumatic node of the brake system after feature enhancement, represents the water content of the pneumatic path of the drying cylinder after feature enhancement, represents the noise spectrum of the drying cylinder after feature enhancement, represents the rotation speed of the air pump after feature enhancement, represents the ambient pressure after feature enhancement; S32、According to the input set Computing temperature parameter compensation value And humidity parameter compensation value The formula is: ； ； wherein, represents the rated rotation speed of the air pump, represents the standard atmospheric pressure of the environment, respectively represent the coupling compensation coefficients of the rotation speed of the air pump and the pressure of the environment to the temperature and humidity. S33, repeat step S32 to compensate all temperature parameters and humidity parameters calculations are completed and a compensation dataset is generated wherein ; S34, compensating the temperature parameter based on the temperature parameter compensation value calculating a temperature standard deviation value of each temperature sensor , and the calculation formula is: ； where S represents a temperature parameter compensation value the number of sensors, represents an average value of the plurality of sensor temperatures; S35、based on the humidity parameter compensation value Calculate the humidity standard deviation value of each humidity sensor The calculation formula is: ； wherein J represents a humidity parameter compensation value the number of, represents an average value of the humidity of the plurality of sensors; S36、According to the temperature standard deviation value Drift correction is performed on the temperature sensor; If a temperature sensor has data drift, drift correction is performed on the temperature sensor; If then the temperature sensor is normal and the end and the corrected temperature set is obtained ; S37、according to the humidity standard deviation value drift correction is performed on the humidity sensor; If If the humidity sensor has data drift, the humidity sensor is corrected for drift. If then the temperature sensor is normal and the process ends and the corrected humidity set is obtained ; S38, the corrected humidity set elements in the set with the moisture content of the drying cylinder gas circuit characteristics to obtain a humidity fusion value The calculation formula is: ； ； wherein, and respectively represent fusion weight values; S39, based on the corrected temperature set and the corrected humidity set obtain a calibration data set .

5. The combined warning method according to claim 4, characterized in that, In step S4, the steps are implemented as follows: S41、According to the calibration data set the drying cylinder noise spectrum in the calibration data set computing the time domain feature values for extracting the time domain features , the formula is: ； ； wherein N represents the number of dry drum noise spectrum number; S42、extracting frequency domain feature values from the time domain feature values S42、extracting frequency domain feature values from the time domain feature values S42、extracting frequency domain feature values from the time domain feature values S42、extracting frequency domain feature values from the time domain feature values S42、extracting frequency domain feature values from the time domain feature values S42、extracting frequency domain feature values from the time domain feature values S42、extracting frequency domain feature values from the time domain feature values ； Wherein, FFT represents fast Fourier transform; S43、According to the humidity fusion value and temperature parameter compensation value Respectively calculate temperature gradient value and humidity gradient value The calculation formula is: ； ； wherein, and represent the first and the second independent variables, respectively; S44, the temperature gradient value and the humidity gradient value composing the gradient matrix ; S45、 according to the temperature gradient value and the humidity gradient value constructing a digital twin dynamic model , expressed as: ； wherein, represents the net outflow of the fluid heat flux at the sensor location, D represents the humidity diffusion coefficient, and v represents the convective velocity of the fluid, represents the heat generation power of the air pump, represents the drying drum dehumidification rate, represents the thermal coupling value, represents the wet coupling value; S46, obtain historical data of the sensor and calculate a temperature residual threshold value of the historical data and a humidity residual threshold value , the calculation formula is: ； ； wherein, respectively represent the residual mean of temperature and humidity of the historical data, represents a confidence coefficient, respectively represent the residual standard deviation of temperature and humidity of the historical data; S47, according to the temperature residual threshold marking the temperature abnormal node; If , the node corresponding to the temperature data is a temperature abnormal node and is marked; If , the node corresponding to the temperature data is a normal temperature node and the process ends. S48、according to the humidity residual threshold value The humidity abnormal node is marked, and the temperature abnormal node and the humidity abnormal node are combined into an abnormal node set B; If , the node corresponding to the humidity data is a humidity abnormal node and is marked; If , the humidity data corresponds to a normal humidity node and ends; S49, the voiceprint feature vector , gradient matrix , thermal coupling value , wet coupling value and the anomaly node set B are merged into a pre-feature set .

6. The method of claim 1, wherein, In step S6, the steps are implemented as follows: S61, define the air pump node as A node, define the drying cylinder node as B node, define the key air path position node as C node, and calculate the air pump path and the drying cylinder path respectively S62, according to the air pump path and the dry cylinder path determining the most likely fault path and generating a fault path result, wherein the fault path result includes a pump air fault result and a cylinder air fault result; If , the air pump path is a fault path, generate a pump air fault result and end; If , the drying cylinder path is a fault path, a cylinder gas fault result is generated and the process ends; S63, send the primary fault result and the secondary abnormal result to the early warning center, generate an AR three-dimensional disassembly animation according to the fault path result, and highlight the fault path and the fault component in the early warning system.

7. The combined warning method according to claim 6, characterized in that, The air pump path And the drying cylinder path The calculation formula is: ； ； Wherein, F and G represent the number of AC path and BC path respectively, represents the probability of node C failure when node A fails, represents the probability of node C failure when node B fails; The propagation path is constructed for the key gas path position node respectively with the gas pump node and the drying cylinder node.

8. A system for applying the combined pre-warning method of the risk of braking of a truck or bus according to any one of claims 1 to 7, characterized in that, The system comprises: The pre-processing module (1) is used for collecting original data in the interior of the passenger and freight vehicle and pre-processing the original data to obtain a dynamic calibration value ; a feature enhancement module (2) for calculating a feature enhancement value for enabling feature decoupling based on the mixed filter value computing a mixed filter value and based on the mixed filter value computing a feature enhancement value for enabling feature decoupling ; a calibration module (3) for calibrating based on feature enhancement values performing a class division to obtain an input set , calibrating the air path node temperature and the air path node humidity in the input set and obtaining a calibration data set for making the data more accurate ; a digital twin module (4) for deriving a thermal coupling value from the calibration data set and a wet coupling value from the calibration data set and calculating a pre-feature set for providing a reference for fault judgment based on the thermal coupling value and the wet coupling value ​ a fault cause module (5) for determining a pre-feature set constructing a brake air path fault cause graph, calculating a judgment probability based on the brake air path fault cause graph, and obtaining a primary fault result and a secondary abnormal result based on the judgment probability; AR generation module (6) for generating an AR animation based on the brake air path fault cause and effect diagram and the pump air path fault cause and effect diagram and the drying drum path based on the fault path results.

Images