A method and system for predicting the remaining life of a wheel-rail tread

Abstract

The present application provides a method and system for predicting the remaining life of a wheel set tread, relating to the technical field of rail transit fault prediction, and particularly to a method for predicting the remaining life of a wheel set tread, including: obtaining the current health assessment result of the wheel set tread to be measured, wherein the health assessment result is at least divided into a first state for characterizing the normal operation of the wheel set tread to be measured, and a second state for characterizing the occurrence of the tread ovality fault feature of the wheel set tread to be measured; determining the remaining life prediction model of the wheel set tread to be measured according to the health assessment result, wherein the remaining life prediction model is one of a first life prediction model corresponding to the first state and a second life prediction model corresponding to the second state; using the remaining life prediction model to predict the remaining life of the wheel set tread to be measured, and obtaining the remaining life prediction result. The present application can improve the reliability of the wheel set tread life prediction.

Description

Technical Field

[0001] The present application relates to the technical field of rail transit fault prediction, and particularly to a method and system for predicting the remaining life of a wheel tread. Background Art

[0002] At present, the turning and maintenance of components in the rail transit field mostly adopt a planned turning method. Although this traditional mode guarantees the operation safety of equipment to a certain extent, there are also many drawbacks. For example, the phenomenon of "over-maintenance" occurs frequently, resulting in over-maintenance of equipment, waste of resources, and a significant increase in maintenance costs. With the continuous development of PHM (Prognostics and Health Management) technology, the demand for preventive maintenance in various industries is becoming increasingly urgent. It is hoped that through precise health management, the service life of components can be extended and the maintenance strategy can be optimized, so as to achieve cost reduction and efficiency improvement. As one of the core technologies of PHM, the remaining life prediction technology has developed rapidly in recent years driven by the wide application of intelligent device monitoring systems. The remaining life prediction technology of components based on real-time monitoring data provides strong support for realizing precise preventive maintenance.

[0003] The wheel tread is one of the key components of the running gear of urban rail vehicles, and its performance and state are directly related to the running safety, smoothness, and overall operation efficiency of the vehicle. However, at present, the methods for predicting the remaining life of the wheel tread mostly rely on empirical thresholds to divide the degradation stages of components, and accordingly select the corresponding degradation models. Since the setting of empirical thresholds is mostly based on subjective judgment and is easily affected by personal experience and thus varies, it will lead to deviations in the division of degradation stages, not conforming to the actual situation of the application data, reducing the reliability of life prediction, and making it difficult to meet the precise maintenance requirements of the wheel tread.

[0004] Therefore, how to improve the reliability of wheel tread life prediction is a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0005] To solve the above technical problems, the present application provides a method for predicting the remaining life of a wheel tread, which can improve the reliability of wheel tread life prediction. The present application also provides a system for predicting the remaining life of a wheel tread, which has the same technical effect.

[0006] The first object of the present application is to provide a method for predicting the remaining life of a wheel tread.

[0007] The above object one of the present application is achieved through the following technical solutions:

[0008] A method for predicting the remaining life of a wheel tread includes:

[0009] Obtain the current health assessment result of the tread of the wheel set to be measured, where the health assessment result is at least divided into a first state for characterizing the normal operation of the tread of the wheel set to be measured, and a second state for characterizing the occurrence of tread roundness loss fault characteristics of the tread of the wheel set to be measured;

[0010] According to the health assessment result, determine the remaining life prediction model of the tread of the wheel set to be measured, where the remaining life prediction model is one of a first life prediction model corresponding to the first state and a second life prediction model corresponding to the second state;

[0011] Use the remaining life prediction model to predict the remaining life of the tread of the wheel set to be measured, and obtain the remaining life prediction result.

[0012] Preferably, in the above remaining life prediction method of the wheel set tread, the health assessment result is at least divided into a normal state, a sub - healthy state and a fault state, where the normal state is used as the first state, and the sub - healthy state and the fault state are used as the second state.

[0013] Preferably, in the above remaining life prediction method of the wheel set tread, the step of using the remaining life prediction model to predict the remaining life of the tread of the wheel set to be measured and obtain the remaining life prediction result includes:

[0014] Obtain the real - time monitoring data and the operating mileage of the tread of the wheel set to be measured, where the real - time monitoring data includes the measured data of out - of - round runout and the measured data of vibration dB value;

[0015] Use the remaining life prediction model to predict the remaining life according to the real - time monitoring data and the operating mileage, and obtain the remaining life prediction result.

[0016] Preferably, in the above remaining life prediction method of the wheel set tread, the step of using the remaining life prediction model to predict the remaining life according to the real - time monitoring data and the operating mileage and obtain the remaining life prediction result includes:

[0017] Perform data pre - processing on the measured data of out - of - round runout, the measured data of vibration dB value and the operating mileage respectively, to obtain out - of - round runout data, vibration dB value data and operating mileage data;

[0018] Perform feature extraction on the out - of - round runout data, the vibration dB value data and the operating mileage data respectively, to obtain out - of - round runout characteristics, vibration effective value characteristics and operating mileage characteristics;

[0019] Perform feature fusion on the out - of - round runout characteristics and the vibration effective value characteristics to obtain fused degradation characteristics;

[0020] Using the remaining life prediction model, based on the fused degradation characteristics and the operating mileage characteristics, perform remaining life prediction to obtain the remaining life prediction result.

[0021] Preferably, in the above remaining life prediction method for the wheel set tread surface, the feature extraction of the out-of-round runout data to obtain the out-of-round runout feature includes:

[0022] Calculate the adaptive first-order density weight-quantile according to the out-of-round runout data to obtain the out-of-round runout feature, and the specific calculation formula is as follows:

[0023] ;

[0024] ;

[0025] In the formula, represents the th feature extraction interval, , is a positive integer, represents the data density within the th feature extraction interval, represents the maximum value of the data density within all feature extraction intervals, represents the first-order weighted density of the th feature extraction interval, , represents the th measurement value within the feature extraction interval, is a positive integer, represents taking the th quantile of represents the out-of-round runout feature.

[0026] Preferably, in the above remaining life prediction method for the wheel set tread surface, the feature extraction of the vibration dB value data to obtain the vibration effective value feature includes:

[0027] Calculate the root mean square value within a preset period according to the vibration dB value data to obtain the vibration effective value feature, and the specific calculation formula is as follows:

[0028] ;

[0029] In the formula, represents the th vibration dB value within the preset period, , is the number of vibration dB values corresponding to one preset period, Represents the average value of the vibration dB value data, Represents the vibration effective value feature.

[0030] Preferably, in the above method for predicting the remaining life of the wheel set tread surface, the feature fusion of the out-of-round runout amount feature and the vibration effective value feature to obtain a fused degradation feature includes:

[0031] Perform normalization processing on the out-of-round runout amount feature and the vibration effective value feature to obtain the normalized out-of-round runout amount feature and the normalized vibration effective value feature;

[0032] Adopt the method of fuzzy weighting to perform feature fusion on the normalized out-of-round runout amount feature and the normalized vibration effective value feature to obtain the fused degradation feature. The specific calculation formula is as follows:

[0033] ;

[0034] In the formula, Represents the normalized out-of-round runout amount feature The corresponding fuzzy weight, Represents the normalized vibration effective value feature The corresponding fuzzy weight, Represents the fused degradation feature.

[0035] Preferably, in the above method for predicting the remaining life of the wheel set tread surface, the first life prediction model adopts a double exponential model, and the second life prediction model adopts a single exponential model.

[0036] Preferably, in the above method for predicting the remaining life of the wheel set tread surface, the use of the remaining life prediction model to perform remaining life prediction according to the fused degradation feature and the operating mileage feature to obtain a remaining life prediction result includes:

[0037] Obtain the turning prediction failure threshold of the wheel set tread surface, where the turning prediction failure threshold is calculated based on multiple historical case data of the wheel set tread surface, and each historical case data includes runout value data, out-of-round runout historical data, and vibration dB value historical data;

[0038] Use the remaining life prediction model to perform remaining life prediction according to the turning prediction failure threshold, the fused degradation feature, and the operating mileage feature to obtain a remaining life prediction result.

[0039] Preferably, in the above method for predicting the remaining life of the wheel set tread surface, the obtaining of the turning prediction failure threshold of the wheel set tread surface includes:

[0040] Obtain multiple pieces of the historical case data of the wheel-rail tread;

[0041] Based on the multiple pieces of the historical case data, calculate to obtain a fused degradation eigenvalue sequence, a runout value sequence, and a data serial number sequence corresponding to the runout value sequence and the fused degradation eigenvalue sequence;

[0042] Sort the fused degradation eigenvalue sequence in ascending order to obtain a sorted first sequence, and based on the first sequence, sort the data serial number sequence to obtain a sorted second sequence;

[0043] Segment the first sequence according to a preset interval length to obtain multiple segmented sequences;

[0044] According to the segmented sequences, respectively count the data serial numbers of the second sequence corresponding to each segmented sequence;

[0045] According to the data serial numbers of the second sequence corresponding to each segmented sequence, count the number of runout values of the corresponding runout value sequence, so as to obtain a runout value number sequence corresponding to the segmented sequence;

[0046] According to the runout value number sequence, the maximum and minimum values of the first sequence, and the number of segmented sequences, draw a frequency histogram of the runout value number sequence;

[0047] According to the frequency histogram, perform data fitting using a statistical distribution to obtain a fitting result;

[0048] According to the fitting result and a preset confidence interval, determine a fused degradation eigenvalue coverage interval, and calculate the mean value of the fused degradation eigenvalues within the fused degradation eigenvalue coverage interval as the turning prediction failure threshold.

[0049] Preferably, in the above-mentioned remaining life prediction method for the wheel-rail tread, when the remaining life prediction model is a first life prediction model corresponding to the first state, using the remaining life prediction model, according to the turning prediction failure threshold, the fused degradation feature, and the in-service mileage feature, perform remaining life prediction to obtain a remaining life prediction result, including:

[0050] Obtain multiple historical case fused degradation features of the wheel-rail tread, where each historical case fused degradation feature is extracted based on a corresponding piece of the historical case data;

[0051] Calculate the similarity between the fused degradation feature and each historical case fused degradation feature respectively, and select the historical case fused degradation feature with the largest similarity as the first historical case fused degradation feature;

[0052] Obtain the pre-constructed double exponential model corresponding to the fused degradation feature of the first historical case as the first remaining life prediction model, where the double exponential model is obtained by fitting based on the fused degradation feature of the first historical case;

[0053] Utilize the first remaining life prediction model to perform remaining life prediction according to the in-service mileage feature and the turning prediction failure threshold, and obtain the point estimate value and interval estimate value of the remaining service life of the tread of the to-be-tested wheel set.

[0054] Preferably, in the above-mentioned method for predicting the remaining life of the wheel set tread, when the remaining life prediction model is the second remaining life prediction model corresponding to the second state, the utilization of the remaining life prediction model to perform remaining life prediction according to the turning prediction failure threshold, the fused degradation feature, and the in-service mileage feature to obtain the remaining life prediction result includes:

[0055] Obtain the pre-constructed single exponential model, and utilize the fused degradation feature and the in-service mileage feature to fit the single exponential model to obtain the second remaining life prediction model;

[0056] Utilize the second remaining life prediction model to perform remaining life prediction according to the in-service mileage feature and the turning prediction failure threshold, and obtain the point estimate value and interval estimate value of the remaining service life of the tread of the to-be-tested wheel set.

[0057] Preferably, in the above-mentioned method for predicting the remaining life of the wheel set tread, the utilization of the fused degradation feature and the in-service mileage feature to fit the single exponential model to obtain the second remaining life prediction model includes:

[0058] Utilize the fused degradation feature and the in-service mileage feature to fit the single exponential model by using the least squares method to obtain the second remaining life prediction model.

[0059] Preferably, in the above-mentioned method for predicting the remaining life of the wheel set tread, the utilization of the second remaining life prediction model to perform remaining life prediction according to the in-service mileage feature and the turning prediction failure threshold, and obtain the point estimate value and interval estimate value of the remaining service life of the tread of the to-be-tested wheel set includes:

[0060] Utilize the second remaining life prediction model to calculate the predicted in-service mileage corresponding to the turning prediction failure threshold;

[0061] According to the predicted operating mileage and the characteristics of the operated mileage, a point estimate of the remaining service life of the tread of the wheel set to be measured is calculated, and an interval estimate of the remaining life of the tread of the wheel set to be measured is calculated by using the Monte Carlo method.

[0062] The second object of the present application is to provide a remaining life prediction system for a wheel set tread.

[0063] The above-mentioned second object of the present application is achieved by the following technical solutions:

[0064] A remaining life prediction system for a wheel set tread includes:

[0065] An acquisition unit, configured to acquire the current health assessment result of the tread of the wheel set to be measured, where the health assessment result is at least divided into a first state for characterizing the normal operation of the tread of the wheel set to be measured, and a second state for characterizing the tread ovality fault characteristics that have occurred on the tread of the wheel set to be measured;

[0066] A determination unit, configured to determine the remaining life prediction model of the tread of the wheel set to be measured according to the health assessment result, where the remaining life prediction model is one of a first life prediction model corresponding to the first state and a second life prediction model corresponding to the second state;

[0067] A prediction unit, configured to use the remaining life prediction model to predict the remaining life of the tread of the wheel set to be measured, and obtain a remaining life prediction result.

[0068] In the above technical solution, starting from the perspective of the development of component failures, the optimal segmentation point of the segmentation model is selected. Specifically, based on the current health assessment result of the tread of the wheel set to be measured, the remaining life prediction model of the tread of the wheel set to be measured is determined. The health assessment result is at least divided into a first state for characterizing the normal operation of the tread of the wheel set to be measured, and a second state for characterizing the tread ovality fault characteristics that have occurred on the tread of the wheel set to be measured. The remaining life prediction model is one of a first life prediction model corresponding to the first state and a second life prediction model corresponding to the second state. Among them, the second state of the health assessment result is used as the segmentation point of the model. When the current health assessment result of the tread of the wheel set to be measured is the second state, the tread of the wheel set to be measured has appeared the tread ovality fault characteristics, and the degradation characteristics show a stable trend and a large amount of data. Compared with the existing solution that relies on empirical thresholds to divide the degradation stage of components and selects the corresponding degradation model accordingly, the above technical solution effectively avoids the problem that the division of the degradation stage caused by subjective judgment is deviated and does not conform to the actual situation of the application data, and can improve the reliability of the wheel set tread life prediction.

[0069] In addition, for the above technical solution, starting from the failure mode, analyzing the physical meaning of the indicators and the physical characteristics of the faults reflected by the component monitoring index data from the mechanism of the faults, the vibration dB value and the out-of-round runout are selected as the monitoring data for remaining life prediction. Considering the number of characteristic dimensions, the complexity and efficiency of engineering applications, a fusion degradation feature extraction method based on the vibration dB value and the out-of-round runout is proposed, which can reflect the overall degradation characteristics of the component in multiple dimensions and effectively improve the accuracy and efficiency of the remaining life prediction of the wheel tread.

[0070] In addition, for the above technical solution, starting from the data characteristics at different stages of the fault, a segmented life prediction model is established. The set theoretical double-exponential model degradation model is adopted in the first stage, and the single-exponential dynamic prediction model is adopted in the second stage, effectively solving the problem that the parameter distribution fitted by relying on a small number of case samples in the existing life prediction methods deviates greatly from the parameter distribution of the actual application data. In addition, according to the accumulated component fault cases, the degradation characteristics of the entire life cycle of the cases are calculated. Taking the first turning point / replacement point as the end point of a cycle, distribution fitting is performed based on all the accumulated cases, and the value when the confidence level is the set value is taken as the turning prediction failure threshold, which can fuse the information of the accumulated component fault case data, formulate the optimal turning point / replacement point, make the remaining life prediction result more in line with the actual application situation, and effectively solve the problem that the failure threshold is set relying on subjective experience in the existing life prediction methods and there is a large deviation from the actual situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0072] Figure 1 It is a schematic flow chart of a method for predicting the remaining life of a wheel tread in an embodiment of the present application;

[0073] Figure 2 It is a frequency histogram provided in an embodiment of the present application;

[0074] Figure 3 It is a multi-peak distribution fitting distribution diagram provided in an embodiment of the present application;

[0075] Figure 4 It is a schematic structural diagram of a system for predicting the remaining life of a wheel tread in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0076] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0077] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described below are only illustrative. For example, the division of units and modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or modules can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the couplings, direct couplings, or communication connections between the components shown or discussed with each other can be through some interfaces, indirect couplings or communication connections of devices or modules, and can be electrical, mechanical or other forms.

[0078] In addition, in each embodiment of this application, each functional unit can be fully integrated in a processor, or each unit can be separately used as a device, or two or more units can be integrated in a device; each functional unit in each embodiment of this application can be implemented in the form of hardware, or in the form of a combination of hardware and software functional units.

[0079] Those of ordinary skill in the art can understand that all or part of the steps of implementing the following method embodiments can be completed through program instructions and related hardware. The foregoing program instructions can be stored in a computer-readable storage medium. When the program instructions are executed, the steps of the following method embodiments are executed; and the foregoing storage medium includes: various media that can store program codes such as removable storage devices, read-only memory (ROM), magnetic disks, or optical discs.

[0080] It should be understood that in this application, if the terms "system", "device", "unit" and / or "module" are used, they are only a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the term can be replaced by other expressions.

[0081] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of this application, the meaning of "a plurality of" or "several" is two or more, unless otherwise specifically defined.

[0082] If a flowchart is used in this application, the flowchart is used to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the previous or subsequent operations are not necessarily executed precisely in sequence. On the contrary, the steps can be processed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several steps can be removed from these processes.

[0083] It should also be noted that in this text, terms such as "comprising", "including" or any other variants thereof are intended to cover non-exclusive inclusion, so that an article or device comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed, or elements inherent to such article or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the existence of another identical element in the article or device comprising the above elements.

[0084] The embodiments of this application are written in a progressive manner.

[0085] As Figure 1 shown, the embodiments of this application provide a method for predicting the remaining life of a wheel-rail tread, including:

[0086] S101. Obtain the current health assessment result of the wheel-rail tread to be measured;

[0087] In S101, specifically, for the wheel-rail tread to be measured in the working state, generally, a health monitoring system is used to monitor its health status in real time and output the corresponding health assessment result in real time. Therefore, each time the remaining life of the wheel-rail tread is predicted, the health assessment result of the wheel-rail tread to be measured at the current moment can be obtained immediately from the health monitoring system of the wheel-rail tread to be measured, such as the health status at the current prediction moment, etc. Taking the wheel-rail tread of an urban rail vehicle as an example, the wheel-rail tread is one of the key components of the running gear of an urban rail vehicle. An intelligent operation and maintenance system for urban rail vehicles is installed in the depot for its condition monitoring and health assessment, realizing dynamic monitoring and health assessment of the wheel-rail tread. Therefore, each time the remaining life of the wheel-rail tread is predicted, the current health assessment result of the wheel-rail tread to be measured can be obtained immediately from the intelligent operation and maintenance system of the urban rail vehicle. This application is not limited thereto.

[0088] Among them, the health assessment results are at least divided into a first state for characterizing the normal operation of the tread of the wheel set to be measured, and a second state for characterizing the occurrence of tread roundness loss fault characteristics on the tread of the wheel set to be measured. According to the application scenario of the wheel set tread, analyze the data characteristics of a tread roundness loss fault cycle. When a wheel set tread starts to operate normally, the amount of degradation characteristic data monitored is small, the degradation trend is uncertain, and all are normal data, and it is not easy to fix the model. Correspondingly, the health assessment results at least include a first state for characterizing the normal operation of the tread of the wheel set to be measured; when early tread roundness loss characteristics appear on the wheel set tread, there is a basic trend in the degradation characteristics and a certain amount of data, and the model can be basically fixed. When the tread roundness loss fault characteristics appear, the degradation characteristics show a stable trend and a large amount of data, and the model can be determined. Correspondingly, the health assessment results at least include a second state for characterizing the occurrence of tread roundness loss fault characteristics on the tread of the wheel set to be measured.

[0089] The health assessment results can also be divided into more levels of states. In some embodiments, the health assessment results are at least divided into a normal state, a sub-healthy state, and a fault state. Among them, the normal state is used as the first state, and the sub-healthy state and the fault state are used as the second state; in other embodiments, the health assessment results are at least divided into a normal state, a sub-healthy state, a minor fault state, a medium fault state, and a severe fault state. Among them, the normal state is used as the first state, and the sub-healthy state, the minor fault state, the medium fault state, and the severe fault state are used as the second state; this application is not limited thereto.

[0090] Among them, the definition of sub-health is the appearance of abnormal signs, and the performance of the component begins to decay, but it can still perform its design function normally. The sub-healthy state can be expressed as the appearance of tread roundness loss fault characteristics, and the degradation characteristics show a stable trend and a large amount of data. Therefore, the second state in the health assessment results, that is, the sub-healthy state, can be used as the segmentation point of the model.

[0091] S102. According to the health assessment results, determine the remaining life prediction model of the tread of the wheel set to be measured. Among them, the remaining life prediction model is one of a first life prediction model corresponding to the first state and a second life prediction model corresponding to the second state;

[0092] In S102, specifically, according to different degradation stages, corresponding degradation models are selected; when the health assessment result is in the first state, that is, when the tread of the wheel set under test is operating normally, the first life prediction model corresponding to the first state is selected; when the health assessment result is in the second state, that is, when the tread of the wheel set under test has shown the fault characteristics of tread ovality, the second life prediction model corresponding to the second state is selected. Among them, the first life prediction model and the second life prediction model can be selected based on actual application requirements. For example, the first life prediction model adopts a double-exponential model or other types of degradation models, and the second life prediction model adopts a single-exponential model or other types of degradation models. This application is not limited to this.

[0093] S103. Use the remaining life prediction model to predict the remaining life of the tread of the wheel set under test, and obtain the remaining life prediction result.

[0094] In S103, specifically, after determining the remaining life prediction model based on the current health assessment result of the tread of the wheel set under test, the remaining life prediction model can be used to predict the remaining life based on the real-time monitoring data and the operating mileage of the tread of the wheel set under test, and obtain the remaining life prediction result.

[0095] Currently, the methods for predicting the remaining life of the wheel set tread mostly rely on empirical thresholds to divide the degradation stages of components, and accordingly select the corresponding degradation models. Since the setting of empirical thresholds is mostly based on subjective judgment, it is easily affected by personal experience and there are differences, which will lead to deviations in the division of degradation stages, not conforming to the actual situation of application data, reducing the reliability of life prediction, and it is difficult to meet the accurate maintenance requirements of the wheel set tread.

[0096] In the above embodiments, starting from the perspective of the development of component failures, the optimal segmentation point of the segmented model is selected. Specifically, based on the current health assessment result of the tread of the wheel set under test, the remaining life prediction model of the tread of the wheel set under test is determined. The health assessment result is at least divided into a first state for characterizing the normal operation of the tread of the wheel set under test, and a second state for characterizing the occurrence of the fault characteristics of tread ovality of the tread of the wheel set under test. The remaining life prediction model is one of the first life prediction model corresponding to the first state and the second life prediction model corresponding to the second state; among them, the second state of the health assessment result is used as the segmentation point of the model. When the current health assessment result of the tread of the wheel set under test is in the second state, the tread of the wheel set under test has shown the fault characteristics of tread ovality, and the degradation characteristics show a stable trend and a large amount of data; compared with the existing solution that relies on empirical thresholds to divide the degradation stages of components and accordingly select the corresponding degradation models, the above embodiments effectively avoid the problem that the division of degradation stages caused by subjective judgment deviates from the actual situation of application data, and can improve the reliability of the life prediction of the wheel set tread.

[0097] Currently, for the remaining life prediction methods of wheelset treads, feature extraction methods such as average and minimum values are often used for key tread indicators. These methods only consider features of a single dimension, do not analyze data features from the perspective of failure mode mechanisms, do not consider multi-dimensional fusion features, and are unable to reflect the overall degradation characteristics of components.

[0098] To solve the above technical problems, in other embodiments of the present application, a remaining life prediction model is used to predict the remaining life of the tread of the wheel pair to be measured, and one implementation method of the steps of obtaining the remaining life prediction result specifically includes:

[0099] S201 obtains real-time monitoring data of the wheelset tread to be measured and the operating mileage;

[0100] In S201, specifically, taking the wheelset tread of an urban rail vehicle as an example, each time the remaining life of the wheelset tread is predicted, real-time monitoring data and operating mileage of the wheelset tread to be tested can be obtained from the urban rail vehicle intelligent operation and maintenance system. The monitoring data includes vibration and impact data, specifically trend data such as radial runout, out-of-round runout, vibration effective value, vibration dB value, impact waveform duty cycle, and impact dB value. Detailed descriptions of each physical quantity are shown in the following table:

[0101] Table 1 Description of physical quantities

[0102]

[0103] The data characteristics and physical meanings of the monitoring indicators are analyzed in combination with the failure mode of the wheelset tread. Taking the wheelset tread out-of-round failure mode as an example, the analysis of the physical quantities of the monitoring indicators is as follows:

[0104] (1) Runout: This is a physical value measured during wheel turning and repair, which can directly reflect the degree of tread out-of-roundness in the wheel set tread out-of-round fault mode. However, this indicator is an offline measurement value and only has a value when the tread is turned. The number is small and discontinuous, making it unsuitable for life prediction scenarios.

[0105] (2) Out-of-round runout: This is an online monitoring quantity that reflects the tread out-of-roundness and is strongly correlated with the tread out-of-round failure mode. In addition, there is monitoring data during normal vehicle operation, which is suitable for life prediction scenarios.

[0106] (3) Impact waveform duty cycle: It can reflect the loss of roundness caused by continuous peeling and fragmentation, and is strongly correlated with the tread loss of roundness failure mode. However, due to the limitations of the data storage mechanism, the data records are incomplete and it is impossible to obtain an accurate impact waveform duty cycle. In addition, expert experience shows that if the train runs in a serpentine manner during operation, the impact waveform duty cycle calculation is inaccurate, so it is not suitable for life prediction scenarios.

[0107] (4)Impact dB value: It reflects the impact information generated after the peeling of the tread surface and cannot reflect the tread state in the case of tread ovality failure mode. Therefore, it is not applicable to the life prediction scenario;

[0108] (5)Vibration effective value: It reflects the vibration energy of the wheel set tread and is strongly correlated with the tread ovality failure mode. In addition, monitoring data is available during normal vehicle operation, which is applicable to the life prediction scenario;

[0109] (6)Vibration dB value: It reflects the vibration intensity of the wheel set tread and is strongly correlated with the tread ovality failure mode. In addition, monitoring data is available during normal vehicle operation, which is applicable to the life prediction scenario.

[0110] Based on the above analysis, the indicators that can be used for the life prediction of tread ovality failure are ovality runout, vibration effective value, and vibration dB value. Among them, the calculation of the vibration dB value uses the vibration effective value, and the vibration dB value covers the energy of the vibration effective value. Therefore, the indicators for the life prediction of tread ovality failure are the vibration dB value and ovality runout. Correspondingly, the real-time monitoring data includes the measured data of ovality runout and the measured data of vibration dB value.

[0111] S202. Use the remaining life prediction model to predict the remaining life based on the real-time monitoring data and the operating mileage, and obtain the remaining life prediction result.

[0112] In S202, specifically, after determining the remaining life prediction model based on the current health assessment result of the tread of the wheel set to be measured, the remaining life prediction model can be used to predict the remaining life based on the measured data of ovality runout, the measured data of vibration dB value, and the operating mileage of the tread of the wheel set to be measured, and obtain the remaining life prediction result.

[0113] In some embodiments, one implementation manner of this step specifically includes:

[0114] S2021. Perform data preprocessing on the measured data of ovality runout, the measured data of vibration dB value, and the operating mileage respectively to obtain ovality runout data, vibration dB value data, and operating mileage data;

[0115] In S2021, specifically, perform outlier processing on the measured data of ovality runout and the measured data of vibration dB value respectively to obtain ovality runout data and vibration dB value data. For example, for abnormal data where the ovality runout and vibration dB value exceed the specified range, such as the ovality runout is greater than or less than , the vibration dB value is greater than or less than , directly adopt the elimination method to avoid the influence of abnormal data on the accuracy of subsequent feature calculation.

[0116] Perform mileage correction processing on the operated mileage to obtain the operated mileage data. Specifically, due to operations such as software replacement and new machine installation, the operated mileage data recorded by the urban rail vehicle online monitoring system has problems such as downward jumps, upward mutations, and zero values. Combining with the daily operation mileage experience value of the vehicle, perform point-by-point correction on the operated mileage to ensure the monotonically increasing property of the operated mileage data.

[0117] S2022. For the out-of-round runout data, vibration dB value data, and operated mileage data, perform feature extraction respectively to obtain the out-of-round runout feature, vibration effective value feature, and operated mileage feature.

[0118] In S2022, specifically, based on the data characteristics, slide and extract the effective features of the out-of-round runout data, vibration dB value data, and operated mileage data respectively to obtain the out-of-round runout feature, vibration effective value feature, and operated mileage feature.

[0119] Among them, considering the non-uniform density characteristic of the out-of-round runout data, one implementation method of the step of performing feature extraction on the out-of-round runout data to obtain the out-of-round runout feature specifically includes: calculating the adaptive first-order density weight-quantile according to the out-of-round runout data to obtain the out-of-round runout feature; the adaptive first-order density weight-quantile is insensitive and relatively stable to the fault-free stage, while being more sensitive to the fault development stage, and can more accurately describe the fault evolution process of the tread. The specific calculation formula is as follows:

[0120] ;

[0121] ;

[0122] In the formula, represents the th feature extraction interval, , is a positive integer, represents the out-of-round runout data density within the th feature extraction interval, represents the maximum value of the out-of-round runout data density within all feature extraction intervals, represents the first-order weighted density of the th feature extraction interval, , represents the out-of-round runout measurement value within the th feature extraction interval, is a positive integer, represents taking the th quantile of , and the operating mileage corresponding to each interval can be Represents the out-of-round runout characteristic.

[0123] Among them, considering that the effective vibration value reflects the vibration intensity of the wheel set tread, the root mean square value within the calculation period reflects the energy of its vibration intensity. Correspondingly, for the vibration dB value data, one implementation method of the steps for feature extraction to obtain the effective vibration value feature specifically includes: calculating the root mean square value within a preset period based on the vibration dB value data to obtain the effective vibration value feature. The specific calculation formula is as follows:

[0124] ;

[0125] In the formula, represents the th vibration dB value within the preset period, , is the number of vibration dB values corresponding to a preset period, represents the average value of the vibration dB value data. The operating mileage corresponding to a preset period can be , represents the effective vibration value feature.

[0126] Among them, one implementation method of the steps for feature extraction from the operated mileage data to obtain the operated mileage feature specifically includes: sliding and calculating the mean value of the operated mileage data to obtain the operated mileage feature. Specifically, the sliding interval length can be 500 km, and the fixed interval length can be 2000 km. This application is not limited to this.

[0127] S2023. Perform feature fusion on the out-of-round runout characteristic and the effective vibration value feature to obtain the fused degradation feature;

[0128] In S2023, specifically, perform normalization processing on the extracted out-of-round runout characteristic and effective vibration value feature. At the same time, considering the number of feature dimensions, the complexity and efficiency of engineering applications, and combining with manual analysis experience, use the method of fuzzy weighting for fused feature extraction to obtain the fused degradation feature. One implementation method of this step specifically includes:

[0129] Perform normalization processing on the out-of-round runout characteristic and the effective vibration value feature to obtain the normalized out-of-round runout characteristic and the normalized effective vibration value feature ; Use the method of fuzzy weighting to perform feature fusion on the normalized out-of-round runout characteristic and the normalized effective vibration value feature to obtain the fused degradation feature. The specific calculation formula is as follows:

[0130] ;

[0131] In the formula, represents the out-of-round runout characteristic after normalization corresponding fuzzy weight, represents the effective vibration value characteristic after normalization corresponding fuzzy weight, represents the fused degradation characteristic.

[0132] S2024. Using the remaining life prediction model, based on the fused degradation characteristic and the in-service mileage characteristic, perform the remaining life prediction to obtain the remaining life prediction result.

[0133] Specifically in S2024, after determining the remaining life prediction model based on the current health assessment result of the tread of the wheel set to be measured, the remaining life prediction model can be used to perform the remaining life prediction based on the fused degradation characteristic and the in-service mileage characteristic of the tread of the wheel set to be measured, so as to obtain the remaining life prediction result.

[0134] In the above embodiment, starting from the failure mode, analyzing the physical meaning of the indicators and the physical characteristics of the failures reflected by the component monitoring index data from the mechanism of the failures, selecting the vibration dB value and the out-of-round runout as the monitoring data, performing the remaining life prediction, and considering the number of characteristic dimensions, the complexity and efficiency of engineering applications, a method for extracting the fused degradation characteristic based on the vibration dB value and the out-of-round runout is proposed, which can reflect the overall degradation characteristics of the component in a multi-dimensional fusion manner, and effectively improve the accuracy and efficiency of the life prediction of the wheel set tread.

[0135] Currently, for the remaining life prediction method of the wheel set tread, the Bayesian principle and the Monte Carlo method are often used to estimate the parameters of the tread degradation model to obtain the predicted degradation trajectory. However, the component may not fail until it has run for a certain period of time, and at the beginning of the component operation, the amount of its monitoring data is small. In the case of a small amount of data, the parameter solution method of the Bayesian theory depends on the set initial parameters and is prone to falling into a local optimum, and the initial parameters are the parameter distributions fitted from a small number of case samples, which deviate greatly from the parameter distributions of the actual application data.

[0136] To solve the above technical problems, in other embodiments of the present application, corresponding degradation models are selected according to the two degradation stages divided by the first state and the second state. Specifically:

[0137] When the health assessment result is in the first state, that is, when the tread of the wheel set under test is operating normally, the first life prediction model corresponding to the first state is selected, and the first life prediction model adopts a double-exponential model; specifically, when the tread of the wheel set under test is operating normally and the degradation trend indicates insufficient data accumulation, based on the historical data of tread ovality failure cases, an initial prediction model is obtained by fitting with the double-exponential model, and then a set double-exponential integration model is formed; among them, the expression of the double-exponential model is as follows:

[0138] ;

[0139] In the formula, represents the degradation characteristic variable, represents the mileage variable, 、 、 、 represent the model parameters.

[0140] When the health assessment result is in the second state, that is, when the tread of the wheel set under test has shown the characteristics of tread ovality failure, the second life prediction model corresponding to the second state is selected, and the second life prediction model adopts a single-exponential model. Specifically, when the tread of the wheel set under test has shown the characteristics of tread ovality failure, the degradation characteristic shows a stable trend and the data has a certain amount, and a single-exponential model is established as the remaining life prediction model, and the expression of the model is as follows:

[0141] ;

[0142] In the formula, represents the degradation characteristic variable, represents the mileage variable, and represent the model parameters.

[0143] Based on the above embodiments and the construction of the above segmented life prediction model, one implementation manner of the step of using the remaining life prediction model to predict the remaining life according to the fused degradation characteristics and the characteristics of the operating mileage is specifically as follows:

[0144] S301. Obtain the turning prediction failure threshold of the wheel set tread;

[0145] In S301, specifically, for the segmented life prediction model, when performing dynamic life prediction, it is necessary to first determine the turning prediction failure threshold corresponding to the optimal turning point. Among them, the turning prediction failure threshold can be calculated based on multiple historical case data of the wheel set tread, and each historical case data includes runout value data, historical data of ovality runout, and historical data of vibration dB value.

[0146] In some embodiments, one implementation of this step specifically includes:

[0147] S3011. Obtain multiple pieces of historical case data of the wheel set tread surface;

[0148] In S3011, specifically, multiple pieces of historical case data of the wheel set tread surface can be obtained from an existing expert knowledge base or case base. The historical case data can specifically be historical case data of tread ovality faults. Each piece of historical case data includes at least one set of data on the runout value during a turning cycle, historical data on ovality runout, and historical data on vibration dB values. The historical case data can also be obtained through other reasonable means, and this application is not limited thereto.

[0149] S3012. Calculate a fused degradation eigenvalue sequence, a runout value sequence, and a data serial number sequence corresponding to the runout value sequence and the fused degradation eigenvalue sequence based on the multiple pieces of historical case data;

[0150] In S3012, specifically, based on the historical data on ovality runout and vibration dB values during one turning cycle in each piece of historical case data, with reference to the specific implementation details in steps S2021 - S2023, calculate the fused degradation eigenvalue corresponding to each piece of historical case data. For example, perform data preprocessing on the historical data on ovality runout and vibration dB values respectively to obtain the preprocessed ovality runout data and the preprocessed vibration dB value data; perform feature extraction on the preprocessed ovality runout data and the preprocessed vibration dB value data respectively to obtain the historical feature of ovality runout and the historical feature of vibration effective value; perform feature fusion on the historical feature of ovality runout and the historical feature of vibration effective value to obtain the fused degradation eigenvalue. Further, based on the fused degradation eigenvalues corresponding to each piece of historical case data, construct a fused degradation eigenvalue sequence, denoted as ; construct a runout value sequence based on the runout value data during one turning cycle in each piece of historical case data, denoted as ; based on the runout value sequence and the fused degradation eigenvalue sequence , construct a data serial number sequence corresponding to the two sequences, denoted as .

[0151] In a specific embodiment, taking 26 pieces of historical case data collected based on an expert knowledge base as an example, extract the fused degradation eigenvalues during one turning cycle of each piece of historical case data, and count the fused degradation eigenvalue sequence =[0.59, 0.82, 0.83, 0.67, 0.76, 0.91, 0.58, 0.89, 0.86, 0.67, 0.74, 1.17, 0.92, 0.75, 0.76, 0.53, 0.38, 0.42, 0.26, 0.41, 0.34, 0.55, 0.35, 0.44, 0.35, 0.85], and statistically obtain the runout value sequence during its grinding =[0.53, 1.05, 1.11, 1.07, 1.11, 1.28, 1.04, 0.8, 1.17, 1.26, 0.86, 0.96, 1.20, 1.40, 1.10, 0.52, 0.53, 0.86, 0.68, 0.76, 0.62, 0.82, 0.91, 0.66, 0.71, 0.93], and the two sequences and and the corresponding data serial number sequences =[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26].

[0152] S3013. Sort the fused degradation eigenvalue sequence in ascending order to obtain the sorted first sequence, and based on the first sequence, sort the data serial number sequence to obtain the sorted second sequence;

[0153] Specifically, in S3013, sort the fused degradation eigenvalue in ascending order to obtain the sorted first sequence, denoted as ; and based on the first sequence , sort the data serial number sequence to obtain the sorted second sequence, denoted as .

[0154] Taking the above specific embodiment as an example, sort the fused degradation eigenvalue in ascending order to obtain the first sequence =[0.26, 0.34, 0.35, 0.35, 0.38, 0.41, 0.42, 0.44, 0.53, 0.55, 0.58, 0.59, 0.67, 0.67, 0.74, 0.75, 0.76, 0.76, 0.82, 0.83, 0.85, 0.86, 0.89, 0.91, 0.92, 1.17], and based on the first sequence , sort the data serial number sequence to obtain the sorted second sequence =[19, 21, 23, 25, 17, 20, 18, 24, 16, 22, 7, 1, 4, 10, 11, 14, 5, 15, 2, 3, 26, 9, 8, 6, 13, 12]. It should be noted that for the same numerical values that appear in the first sequence such as two 0.35, two 0.67, and two 0.76 in the first sequence when sorting the data serial number sequence according to the first sequence the data serial number sequence can be sorted in ascending order according to the data serial numbers corresponding to these numerical values.

[0155] S3014. Segment the first sequence according to the preset interval length to obtain multiple segmented sequences;

[0156] In S3014, specifically, the preset interval length can be set based on actual requirements. Preferably, the preset interval length can be set to 0.1; segment the first sequence according to the preset interval length to obtain multiple segmented sequences after segmentation.

[0157] Taking the above specific embodiment as an example, the maximum and minimum values of the first sequence are 1.17 and 0.26 respectively. Segment the first sequence according to the interval length of 0.1, and the following 10 segmented sequences can be obtained: The segmented sequence in the interval [0.2, 0.3) = [0.26], the segmented sequence in the interval [0.3, 0.4) = [0.34, 0.35, 0.35, 0.38], the segmented sequence in the interval [0.4, 0.5) = [0.41, 0.42, 0.44], the segmented sequence in the interval [0.5, 0.6) = [0.53, 0.55, 0.58, 0.59], the segmented sequence in the interval [0.6, 0.7) = [0.67, 0.67], the segmented sequence in the interval [0.7, 0.8) = [0.74, 0.75, 0.76, 0.76], the segmented sequence in the interval [0.8, 0.9) = [0.82, 0.83, 0.85, 0.86, 0.89], the segmented sequence in the interval [0.9, 1.0) = [0.91, 0.92], the segmented sequence in the interval [1.0, 1.1) is empty, the segmented sequence in the interval [1.1, 1.2) = [1.17].

[0158] S3015. According to the segmented sequences, respectively count the data serial numbers of the second sequence corresponding to each segmented sequence;

[0159] In S3015, specifically, according to the segmented sequences, respectively count the data serial numbers of the second sequence within the range of each segmented sequence.

[0160] Taking the above specific embodiment as an example, statistically segment the data serial numbers of the second sequence within each segment range of the above 10 segmented sequences, as shown in the following table:

[0161] Table 2 Statistical table of segmented sequences and the second sequence

[0162]

[0163] S3016. According to the data serial numbers of the second sequence corresponding to each segmented sequence, statistically count the number of runout values of the corresponding runout value sequence, so as to obtain the runout value number sequence corresponding to the segmented sequence;

[0164] Specifically, in S3016, based on the data serial numbers of the second sequence within the range of each segmented sequence, statistically count the number of data in the corresponding range of the runout value sequence, that is, the number of runout values, and finally obtain the runout value number sequence of the segmented runout value sequence denoted as .

[0165] Taking the above specific embodiment as an example, based on the data serial numbers of the second sequence within the range of each of the above 10 segmented sequences, statistically count the number of data in the corresponding range of the runout value sequence and finally obtain the runout value number sequence of the segmented runout value sequence as shown in the following table:

[0166] Table 3 Statistical table of segmented sequences and runout value sequences

[0167]

[0168] As can be seen from the above table, the runout value number sequence = [1, 4, 3, 4, 2, 4, 5, 2, 0, 1].

[0169] S3017. According to the runout value number sequence, the maximum and minimum values of the first sequence, and the number of segmented sequences, draw a frequency histogram of the runout value number sequence;

[0170] Specifically, in S3017, screen out the maximum and minimum values of the first sequence and, based on the runout value number sequence , the maximum and minimum values of the first sequence , and the number of segmented sequences, draw the runout value number sequence ​​​​​Frequency histogram. Taking the above specific embodiment as an example, based on the above-mentioned number sequence of jump values , the first sequence with the maximum value of 1.17 and the minimum value of 0.26, and the number of segmented sequences being 10, a frequency histogram of the number sequence of radial jump values is plotted (i.e., the fused degradation eigenvalue dot plot during maintenance), as shown in Figure 2 .

[0171] S3018. According to the frequency histogram, use statistical distribution for data fitting to obtain the fitting result;

[0172] In S3018, specifically, according to the frequency histogram, use statistical distribution for data fitting to obtain the fitting result. Among them, the method of statistical distribution can be selected based on actual needs. Preferably, a multi-peak distribution is adopted. Taking the above specific embodiment as an example, the multi-peak distribution fitting diagram based on multi-peak distribution fitting is shown in Figure 3 .

[0173] S3019. According to the fitting result and the preset confidence interval, determine the coverage interval of the fused degradation eigenvalue, and calculate the mean value of the fused degradation eigenvalues within the coverage interval of the fused degradation eigenvalue as the turning prediction failure threshold.

[0174] In S3019, specifically, the preset confidence interval can be set based on actual needs. Preferably, the 95% confidence interval of the multi-peak distribution can be taken as the preset confidence interval; according to the multi-peak distribution fitting result and the preset confidence interval, determine the coverage interval of the fused degradation eigenvalue corresponding to the preset confidence interval, and calculate the mean value of the fused degradation eigenvalues within the coverage interval of the fused degradation eigenvalue as the turning prediction failure threshold.

[0175] Taking the above specific embodiment as an example, taking the 95% confidence interval of the multi-peak distribution, the coverage interval of the fused degradation eigenvalue during maintenance is [0.4, 1]. It can be determined that the fused degradation eigenvalues within the coverage interval [0.4, 1] of the fused degradation eigenvalue are specifically [0.41, 0.42, 0.44, 0.53, 0.55, 0.58, 0.59, 0.67, 0.67, 0.74, 0.75, 0.76, 0.76, 0.82, 0.83, 0.85, 0.86, 0.89, 0.91, 0.92]. Calculate the average value of all values in this sequence, which is 0.6975, as the optimal turning prediction failure threshold.

[0176] In some other embodiments, according to application experience, when the radial jump value is greater than 0.5, tread turning is required. According to the sequence of radial jump values corresponding to the coverage interval [0.4, 1] of the fused degradation eigenvalue obtained during maintenance Values are subject to a screening process where values greater than 0.5 are selected, and runout values less than or equal to 0.5 are excluded. The resulting sequence of runout values after screening is [0.76, 0.86, 0.66, 0.52, 0.82, 1.04, 0.53, 1.07, 1.26, 0.86, 1.40, 1.11, 1.10, 1.05, 1.11, 0.93, 1.17, 0.80, 1.28, 1.20], along with their corresponding data sequence numbers [20, 18, 24, 16, 22, 7, 1, 4, 10, 11, 14, 5, 15, 2, 3, 26, 9, 8, 6, 13], and the corresponding sequence of fusion degradation eigenvalues for calculating the optimal turning repair prediction failure threshold [0.41, 0.42, 0.44, 0.53, 0.55, 0.58, 0.59, 0.67, 0.67, 0.74, 0.75, 0.76, 0.76, 0.82, 0.83, 0.85, 0.86, 0.89, 0.91, 0.92].

[0177] Through steps S3011 - S3019, the determination of the optimal prediction failure point of the component is achieved. It can integrate the accumulated component failure case data information, with the end point of each cycle being a turning repair point / replacement point, and formulate the optimal turning repair point / replacement point, making the remaining life prediction result more in line with the actual application situation.

[0178] S302. Using the remaining life prediction model, based on the turning repair prediction failure threshold, fusion degradation characteristics, and the characteristics of the operating mileage, perform the remaining life prediction to obtain the remaining life prediction result.

[0179] In S302, specifically, after determining the remaining life prediction model based on the current health assessment result of the tread surface of the wheel set to be measured, the remaining life prediction model can be used to estimate the current remaining life prediction point and calculate the interval estimate value based on the fusion degradation characteristics, the characteristics of the operating mileage, and the turning repair prediction failure threshold of the tread surface of the wheel set to be measured, as the remaining life prediction result.

[0180] In some embodiments, when the health assessment result is in the first state, that is, when the tread surface of the wheel set to be measured is operating normally, select the first life prediction model corresponding to the first state. Correspondingly, one implementation manner of this step specifically includes:

[0181] S3021. Obtain the fusion degradation characteristics of multiple historical cases of the tread surface of the wheel set;

[0182] In S3021, specifically, each fusion degradation characteristic of the historical case is extracted based on the corresponding historical case data of the tread surface of the wheel set. Each historical case data includes historical data of out - of - round runout and historical data of vibration dB values.

[0183] In some embodiments, one implementation of this step specifically includes: obtaining multiple historical case data of the wheel-rail tread; and calculating the corresponding historical case fusion degradation features respectively according to each piece of historical case data.

[0184] Among them, multiple historical case data of the wheel-rail tread can be obtained from the existing expert knowledge base or case base. The historical case data can specifically be historical case data of tread ovality faults. Each piece of historical case data includes historical data of ovality runout and historical data of vibration dB value. The historical case data can also be obtained through other reasonable means, and this application is not limited thereto. According to the historical data of ovality runout and the historical data of vibration dB value in each piece of historical case data, referring to the specific implementation details in steps S2021 - S2023, calculate the historical case fusion degradation features corresponding to each piece of historical case data; for example, perform data preprocessing on the historical data of ovality runout and the historical data of vibration dB value respectively to obtain the preprocessed ovality runout data and the preprocessed vibration dB value data; perform feature extraction on the preprocessed ovality runout data and the preprocessed vibration dB value data respectively to obtain the historical features of ovality runout and the historical features of vibration effective value; perform feature fusion on the historical features of ovality runout and the historical features of vibration effective value to obtain the historical case fusion degradation features.

[0185] S3022. Calculate the similarity between the fusion degradation feature and each historical case fusion degradation feature respectively, and select the historical case fusion degradation feature with the maximum similarity as the first historical case fusion degradation feature;

[0186] In S3022, specifically, an existing similarity calculation method can be used to calculate the similarity between the fusion degradation feature and each historical case fusion degradation feature, and select the historical case fusion degradation feature corresponding to the maximum similarity value as the first historical case fusion degradation feature.

[0187] S3023. Obtain the pre-constructed double-exponential model corresponding to the first historical case fusion degradation feature as the first life prediction model;

[0188] In S3023, specifically, the pre-constructed double-exponential model corresponding to the first historical case fusion degradation feature is obtained by fitting based on the first historical case fusion degradation feature; in some embodiments, a double-exponential model can be used to fit each historical case fusion degradation feature respectively to obtain a double-exponential integration model, where the double-exponential integration model includes a double-exponential model corresponding to each historical case fusion degradation feature; after determining the first historical case fusion degradation feature, the pre-constructed double-exponential model corresponding to the first historical case fusion degradation feature can be obtained from the double-exponential integration model as the first life prediction model.

[0189] It should be noted that in the process of pre - constructing the double - exponential model, a parameter solution method based on Bayesian theory or an alternating iterative parameter solution method in the same round can be adopted; however, in the segmented model, in the first stage, the theoretical initial prediction model obtained by directly using the historical case fusion degradation features extracted from the historical case data in the expert knowledge base or case base for fitting the double - exponential function does not require a parameter update method.

[0190] S3024. Using the first life prediction model, according to the operating mileage characteristics and the turning prediction failure threshold, conduct remaining life prediction to obtain the point estimate and interval estimate of the remaining service life of the tread of the wheel set to be measured.

[0191] In S3024, specifically, using the first life prediction model, based on the operating mileage characteristics and the turning prediction failure threshold, calculate the point estimate and interval estimate of the remaining service life of the tread of the wheel set to be measured at the current moment as the remaining life prediction result.

[0192] In some other embodiments, when the health assessment combines into the second state, that is, when the tread of the wheel set to be measured has shown the fault characteristics of tread ovality, select the second life prediction model corresponding to the second state. Correspondingly, one implementation manner of this step specifically includes:

[0193] S3025. Obtain the pre - constructed single - exponential model, and use the fusion degradation characteristics and the operating mileage characteristics to fit the single - exponential model to obtain the second life prediction model;

[0194] In S3025, specifically, use the fusion degradation characteristics and the operating mileage characteristics to estimate the parameters of the single - exponential model at the current moment to obtain the second life prediction model; preferably, the fusion degradation characteristics and the operating mileage characteristics can be used to fit the single - exponential model by using the least - squares method to obtain the second life prediction model. Among them, the least - squares method has accurate parameter solution in the case of a large amount of data and is often used for models with a small number of parameters and simple maximum likelihood.

[0195] S3026. Using the second life prediction model, according to the operating mileage characteristics and the turning prediction failure threshold, conduct remaining life prediction to obtain the point estimate and interval estimate of the remaining service life of the tread of the wheel set to be measured.

[0196] In S3026, specifically, using the second life prediction model, calculate the predicted operating mileage corresponding to the turning prediction failure threshold; according to the predicted operating mileage and the characteristics of the already-operated mileage, calculate the point estimate of the remaining service life of the tread of the wheel set to be measured, and use the Monte Carlo method to calculate the interval estimate of the remaining life of the tread of the wheel set to be measured. Among them, the point estimate of the remaining service life of the tread of the wheel set to be measured at the current moment can be obtained by subtracting the characteristics of the already-operated mileage from the predicted operating mileage; the Monte Carlo method is a numerical calculation method based on random sampling. Through the Monte Carlo method, a large number of possible degradation trajectories can be generated, so as to estimate the future state of the component and obtain the interval estimate of the remaining service life of the tread of the wheel set to be measured at the current moment.

[0197] In the above embodiment, starting from the data characteristics in different stages of the fault, a segmented life prediction model is established. In the first stage, the set theoretical double-exponential model degradation model is adopted, and in the second stage, the single-exponential dynamic prediction model is adopted, effectively solving the problem that the parameter distribution fitted by relying on a small number of case samples in the existing life prediction method deviates greatly from the parameter distribution of the actual application data.

[0198] In addition, in the above embodiment, according to the accumulated component fault cases, the degradation characteristics of the entire life cycle of the cases are calculated. Taking the first turning point / replacement point as the end point of a cycle, distribution fitting is performed based on all the accumulated cases, and the value when the confidence level is the set value is taken as the turning prediction failure threshold, which can integrate the accumulated component fault case data information, formulate the optimal turning point / replacement point, make the remaining life prediction result more in line with the actual application situation, and effectively solve the problem that the failure threshold is set relying on subjective experience in the existing life prediction method and there is a large deviation from the actual situation.

[0199] As Figure 4 shown, in another embodiment of the present application, a remaining life prediction system for the tread of a wheel set is further provided, including:

[0200] An acquisition unit 10, configured to acquire the current health assessment result of the tread of the wheel set to be measured, where the health assessment result is at least divided into a first state for characterizing the normal operation of the tread of the wheel set to be measured, and a second state for characterizing the tread ovality fault characteristics that have occurred on the tread of the wheel set to be measured;

[0201] A determination unit 11, configured to determine the remaining life prediction model of the tread of the wheel set to be measured according to the health assessment result, where the remaining life prediction model is one of a first life prediction model corresponding to the first state and a second life prediction model corresponding to the second state;

[0202] A prediction unit 12 is configured to use a remaining life prediction model to predict the remaining life of the tread of the wheel set to be measured, and obtain a remaining life prediction result.

[0203] In other embodiments of the present application, in the above remaining life prediction system for the wheel set tread, the health assessment result is at least divided into a normal state, a sub-healthy state, and a failure state. Among them, the normal state is used as the first state, and the sub-healthy state and the failure state are used as the second state.

[0204] Preferably, in the above remaining life prediction method for the wheel set tread, when the prediction unit 12 executes using the remaining life prediction model to predict the remaining life of the tread of the wheel set to be measured and obtain a remaining life prediction result, it is specifically configured to:

[0205] Obtain the real-time monitoring data and the operating mileage of the tread of the wheel set to be measured. Among them, the real-time monitoring data includes the measured data of the out-of-round runout and the measured data of the vibration dB value;

[0206] Use the remaining life prediction model to perform remaining life prediction based on the real-time monitoring data and the operating mileage, and obtain a remaining life prediction result.

[0207] In other embodiments of the present application, in the above remaining life prediction system for the wheel set tread, when the prediction unit 12 executes using the remaining life prediction model to perform remaining life prediction based on the real-time monitoring data and the operating mileage and obtain a remaining life prediction result, it is specifically configured to:

[0208] Perform data preprocessing on the measured data of the out-of-round runout, the measured data of the vibration dB value, and the operating mileage respectively to obtain out-of-round runout data, vibration dB value data, and operating mileage data;

[0209] Perform feature extraction on the out-of-round runout data, the vibration dB value data, and the operating mileage data respectively to obtain out-of-round runout features, vibration effective value features, and operating mileage features;

[0210] Perform feature fusion on the out-of-round runout features and the vibration effective value features to obtain fused degradation features;

[0211] Use the remaining life prediction model to perform remaining life prediction based on the fused degradation features and the operating mileage features, and obtain a remaining life prediction result.

[0212] In other embodiments of the present application, in the above remaining life prediction system for the wheel set tread, when the prediction unit 12 executes performing feature extraction on the out-of-round runout data to obtain out-of-round runout features, it is specifically configured to:

[0213] According to the data of out-of-round runout, calculate the adaptive first-order density weight - quantile to obtain the out-of-round runout characteristics. The specific calculation formula is as follows:

[0214] ;

[0215] ;

[0216] In the formula, represents the th feature extraction interval, , is a positive integer, represents the data density within the th feature extraction interval, represents the maximum value of the data density within all feature extraction intervals, represents the first-order weighted density of the th feature extraction interval, , represents the measured value of the th feature extraction interval, is a positive integer, represents taking the th quantile of represents the out-of-round runout characteristic.

[0217] In other embodiments of the present application, in the above remaining life prediction system for the wheel - set tread surface, when the prediction unit 12 performs feature extraction on the vibration dB value data to obtain the vibration effective value characteristic, it specifically is used for:

[0218] According to the vibration dB value data, calculate the root mean square value within a preset period to obtain the vibration effective value characteristic. The specific calculation formula is as follows:

[0219] ;

[0220] In the formula, represents the th vibration dB value within the preset period, , is the number of vibration dB values corresponding within a preset period, represents the average value of the vibration dB value data, represents the vibration effective value characteristic.

[0221] In other embodiments of the present application, in the above remaining life prediction system for the wheel - set tread surface, when the prediction unit 12 performs feature fusion on the out-of-round runout characteristic and the vibration effective value characteristic to obtain the fusion degradation characteristic, it specifically is used for:

[0222] Normalize the out-of-round runout feature and the effective vibration value feature to obtain the normalized out-of-round runout feature and the normalized effective vibration value feature;

[0223] Adopt the method of fuzzy weighting to perform feature fusion on the normalized out-of-round runout feature and the normalized effective vibration value feature to obtain the fused degradation feature. The specific calculation formula is as follows:

[0224] ;

[0225] In the formula, represents the fuzzy weight corresponding to the normalized out-of-round runout feature ; represents the fuzzy weight corresponding to the normalized effective vibration value feature ; represents the fused degradation feature.

[0226] In other embodiments of the present application, in the above remaining life prediction system for the wheel set tread surface, the first life prediction model adopts a double exponential model, and the second life prediction model adopts a single exponential model.

[0227] In other embodiments of the present application, in the above remaining life prediction system for the wheel set tread surface, when the prediction unit 12 executes the remaining life prediction using the remaining life prediction model according to the fused degradation feature and the operating mileage feature to obtain the remaining life prediction result, it is specifically used for:

[0228] Obtain the turning prediction failure threshold of the wheel set tread surface, where the turning prediction failure threshold is calculated based on multiple historical case data of the wheel set tread surface, and each historical case data includes radial runout value data, out-of-round runout historical data, and vibration dB value historical data;

[0229] Use the remaining life prediction model to perform remaining life prediction according to the turning prediction failure threshold, the fused degradation feature, and the operating mileage feature to obtain the remaining life prediction result.

[0230] In other embodiments of the present application, in the above remaining life prediction system for the wheel set tread surface, when the prediction unit 12 executes to obtain the turning prediction failure threshold of the wheel set tread surface, it is specifically used for:

[0231] Obtain multiple historical case data of the wheel set tread surface;

[0232] According to the multiple historical case data, calculate the fused degradation feature value sequence, the radial runout value sequence, and the data serial number sequence corresponding to the radial runout value sequence and the fused degradation feature value sequence;

[0233] Sort the sequence of fused degradation eigenvalues in ascending order to obtain the first sorted sequence, and based on the first sequence, sort the data serial number sequence to obtain the second sorted sequence;

[0234] Segment the first sequence according to a preset interval length to obtain multiple segmented sequences;

[0235] According to the segmented sequences, respectively count the data serial numbers of the second sequence corresponding to each segmented sequence;

[0236] According to the data serial numbers of the second sequence corresponding to each segmented sequence, count the number of runout values of the corresponding runout value sequence, so as to obtain a runout value number sequence corresponding to the segmented sequence;

[0237] According to the runout value number sequence, the maximum and minimum values of the first sequence, and the number of segmented sequences, draw a frequency histogram of the runout value number sequence;

[0238] According to the frequency histogram, perform data fitting using a statistical distribution to obtain a fitting result;

[0239] According to the fitting result and a preset confidence interval, determine the coverage interval of the fused degradation eigenvalue, and calculate the mean value of the fused degradation eigenvalues within the coverage interval of the fused degradation eigenvalue as the turning repair prediction failure threshold.

[0240] In other embodiments of the present application, in the above remaining life prediction system for the wheel tread, when the remaining life prediction model is the first life prediction model corresponding to the first state, when the prediction unit 12 executes using the remaining life prediction model to perform remaining life prediction according to the turning repair prediction failure threshold, the fused degradation feature, and the in-service mileage feature to obtain a remaining life prediction result, it is specifically used for:

[0241] Obtain multiple historical case fused degradation features of the wheel tread, where each historical case fused degradation feature is extracted based on a corresponding piece of historical case data;

[0242] Calculate the similarity between the fused degradation feature and each historical case fused degradation feature respectively, and select the historical case fused degradation feature with the largest similarity as the first historical case fused degradation feature;

[0243] Obtain the pre-constructed double exponential model corresponding to the first historical case fused degradation feature as the first life prediction model, where the double exponential model is fitted based on the first historical case fused degradation feature;

[0244] Use the first life prediction model to perform remaining life prediction according to the in-service mileage feature and the turning repair prediction failure threshold to obtain a point estimate value and an interval estimate value of the remaining service life of the wheel tread to be measured.

[0245] In other embodiments of the present application, in the above remaining life prediction system for the wheel-rail tread, when the remaining life prediction model is the second life prediction model corresponding to the second state, when the prediction unit 12 executes using the remaining life prediction model to predict the remaining life according to the turning repair prediction failure threshold, the fused degradation feature, and the operating mileage feature to obtain the remaining life prediction result, it is specifically used for:

[0246] Obtain a pre-constructed single exponential model, and use the fused degradation feature and the operating mileage feature to fit the single exponential model to obtain the second life prediction model;

[0247] Use the second life prediction model to predict the remaining life according to the operating mileage feature and the turning repair prediction failure threshold to obtain the point estimate value and the interval estimate value of the remaining service life of the wheel-rail tread to be measured.

[0248] In other embodiments of the present application, in the above remaining life prediction system for the wheel-rail tread, when the prediction unit 12 executes using the fused degradation feature and the operating mileage feature to fit the single exponential model to obtain the second life prediction model, it is specifically used for:

[0249] Use the fused degradation feature and the operating mileage feature, and adopt the least squares method to fit the single exponential model to obtain the second life prediction model.

[0250] In other embodiments of the present application, in the above remaining life prediction system for the wheel-rail tread, when the prediction unit 12 executes using the second life prediction model to predict the remaining life according to the operating mileage feature and the turning repair prediction failure threshold to obtain the point estimate value and the interval estimate value of the remaining service life of the wheel-rail tread to be measured, it is specifically used for:

[0251] Use the second life prediction model to calculate the predicted operating mileage corresponding to the turning repair prediction failure threshold;

[0252] According to the predicted operating mileage and the operating mileage feature, calculate the point estimate value of the remaining service life of the wheel-rail tread to be measured, and use the Monte Carlo method to calculate the interval estimate value of the remaining life of the wheel-rail tread to be measured.

[0253] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the remaining life of a wheel-rail tread, characterized in that, Including: Obtain the current health assessment result of the tread of the wheel set to be measured, where the health assessment result is at least divided into a first state for characterizing the normal operation of the tread of the wheel set to be measured, and a second state for characterizing the occurrence of tread ovality fault characteristics in the tread of the wheel set to be measured; Determine the remaining life prediction model of the tread of the wheel set to be measured according to the health assessment result, where the remaining life prediction model is one of a first life prediction model corresponding to the first state and a second life prediction model corresponding to the second state; Use the remaining life prediction model to predict the remaining life of the tread of the wheel set to be measured to obtain a remaining life prediction result; The using the remaining life prediction model to predict the remaining life of the tread of the wheel set to be measured to obtain a remaining life prediction result includes: Obtain the real-time monitoring data and the operating mileage of the tread of the wheel set to be measured, where the real-time monitoring data includes the measured data of the ovality jump amount and the measured data of the vibration dB value; Use the remaining life prediction model to perform remaining life prediction according to the real-time monitoring data and the operating mileage to obtain a remaining life prediction result; The using the remaining life prediction model to perform remaining life prediction according to the real-time monitoring data and the operating mileage to obtain a remaining life prediction result includes: Perform data preprocessing on the measured data of the ovality jump amount, the measured data of the vibration dB value, and the operating mileage respectively to obtain ovality jump amount data, vibration dB value data, and operating mileage data; Perform feature extraction on the ovality jump amount data, the vibration dB value data, and the operating mileage data respectively to obtain an ovality jump amount feature, a vibration effective value feature, and an operating mileage feature; Perform feature fusion on the ovality jump amount feature and the vibration effective value feature to obtain a fused degradation feature; Use the remaining life prediction model to perform remaining life prediction according to the fused degradation feature and the operating mileage feature to obtain a remaining life prediction result.

2. The method according to claim 1, characterized in that, The health assessment result is at least divided into a normal state, a sub-healthy state, and a fault state, where the normal state is used as the first state, and the sub-healthy state and the fault state are used as the second state.

3. The method according to claim 1, characterized in that, The performing feature extraction on the ovality jump amount data to obtain an ovality jump amount feature includes: Calculate the adaptive first-order density weight-quantile according to the ovality jump amount data to obtain the ovality jump amount feature, and the specific calculation formula is as follows: ； ； In the formula, represents the th feature extraction interval, , where is a positive integer, represents the data density within the th feature extraction interval, represents the maximum value of the data density within all feature extraction intervals, represents the first-order weighted density of the th feature extraction interval, represents the measurement value of the th feature extraction interval, is a positive integer, represents taking the quantile of , represents the out-of-round runout amount feature.

4. The method according to claim 3, wherein The performing feature extraction on the vibration dB value data to obtain a vibration effective value feature includes: Calculate the root mean square value within a preset period according to the vibration dB value data to obtain the vibration effective value feature, and the specific calculation formula is as follows: ； Wherein, represents the -th vibration dB value within the preset period, , is the number of corresponding vibration dB values within one preset period, represents the average value of the vibration dB value data, represents the vibration effective value feature.

5. The method according to claim 4, characterized in that The performing feature fusion on the ovality jump amount feature and the vibration effective value feature to obtain a fused degradation feature includes: Perform normalization processing on the ovality jump amount feature and the vibration effective value feature to obtain a normalized ovality jump amount feature and a normalized vibration effective value feature; Using the method of fuzzy weighting, the out-of-round runout amount feature after normalization and the effective vibration value feature after normalization are subjected to feature fusion to obtain the fused degradation feature. The specific calculation formula is as follows: ； In the formula, represents the out-of-round runout amount feature after normalization corresponding fuzzy weight, represents the effective vibration value feature after normalization corresponding fuzzy weight, represents the fused degradation feature.

6. The method according to claim 1, wherein The first remaining life prediction model adopts a double exponential model, and the second remaining life prediction model adopts a single exponential model.

7. The method according to claim 6, characterized in that, Using the remaining life prediction model, based on the fused degradation feature and the operating mileage feature, the remaining life is predicted to obtain the remaining life prediction result, including: Obtain the turning prediction failure threshold of the wheel set tread surface, where the turning prediction failure threshold is calculated based on multiple historical case data of the wheel set tread surface, and each piece of historical case data includes runout value data, out-of-round runout amount historical data, and vibration dB value historical data; Using the remaining life prediction model, based on the turning prediction failure threshold, the fused degradation feature, and the operating mileage feature, the remaining life is predicted to obtain the remaining life prediction result.

8. The method according to claim 7, wherein The obtaining of the turning prediction failure threshold of the wheel set tread surface includes: Obtain multiple pieces of the historical case data of the wheel set tread surface; Based on multiple pieces of the historical case data, calculate the fused degradation feature value sequence, the runout value sequence, and the data serial number sequence corresponding to the runout value sequence and the fused degradation feature value sequence; Sort the fused degradation feature value sequence in ascending order to obtain the sorted first sequence, and based on the first sequence, sort the data serial number sequence to obtain the sorted second sequence; Segment the first sequence according to a preset interval length to obtain multiple segmented sequences; According to the segmented sequences, respectively count the data serial numbers of the second sequence corresponding to each segmented sequence; According to the data serial numbers of the second sequence corresponding to each segmented sequence, count the number of runout values of the corresponding runout value sequence to obtain the runout value number sequence corresponding to the segmented sequence; According to the runout value number sequence, the maximum and minimum values of the first sequence, and the number of segmented sequences, draw the frequency histogram of the runout value number sequence; According to the frequency histogram, perform data fitting using statistical distribution to obtain the fitting result; According to the fitting result and the preset confidence interval, determine the fused degradation feature value coverage interval, and calculate the mean value of the fused degradation feature values within the fused degradation feature value coverage interval as the turning prediction failure threshold.

9. The method according to claim 7, wherein When the remaining life prediction model is the first remaining life prediction model corresponding to the first state, the using of the remaining life prediction model, based on the turning prediction failure threshold, the fused degradation feature, and the operating mileage feature, to predict the remaining life to obtain the remaining life prediction result, includes: Obtain multiple historical case fused degradation features of the wheel set tread surface, where each historical case fused degradation feature is extracted based on a corresponding piece of the historical case data; Calculate the similarity between the fused degradation feature and each of the historical case fused degradation features respectively, and select the historical case fused degradation feature with the maximum similarity as the first historical case fused degradation feature; Obtain the pre-constructed double exponential model corresponding to the first historical case fused degradation feature as the first remaining life prediction model, where the double exponential model is obtained by fitting based on the first historical case fused degradation feature; Use the first remaining life prediction model to perform remaining life prediction according to the in-service mileage feature and the turning prediction failure threshold, and obtain the point estimate value and the interval estimate value of the remaining service life of the tread of the to-be-tested wheel set.

10. The method according to claim 7, characterized in that, When the remaining life prediction model is the second remaining life prediction model corresponding to the second state, using the remaining life prediction model, according to the turning prediction failure threshold, the fused degradation feature and the in-service mileage feature, perform remaining life prediction, and obtain the remaining life prediction result, including: Obtain the pre-constructed single exponential model, and use the fused degradation feature and the in-service mileage feature to fit the single exponential model to obtain the second remaining life prediction model; Use the second remaining life prediction model to perform remaining life prediction according to the in-service mileage feature and the turning prediction failure threshold, and obtain the point estimate value and the interval estimate value of the remaining service life of the tread of the to-be-tested wheel set.

11. The method according to claim 10, wherein The using the fused degradation feature and the in-service mileage feature to fit the single exponential model to obtain the second remaining life prediction model includes: Use the fused degradation feature and the in-service mileage feature to fit the single exponential model by the least squares method to obtain the second remaining life prediction model.

12. The method according to claim 10, wherein The using the second remaining life prediction model to perform remaining life prediction according to the in-service mileage feature and the turning prediction failure threshold, and obtain the point estimate value and the interval estimate value of the remaining service life of the tread of the to-be-tested wheel set includes: Use the second remaining life prediction model to calculate the predicted in-service mileage corresponding to the turning prediction failure threshold; According to the predicted in-service mileage and the in-service mileage feature, calculate the point estimate value of the remaining service life of the tread of the to-be-tested wheel set, and use the Monte Carlo method to calculate the interval estimate value of the remaining life of the tread of the to-be-tested wheel set.

13. A remaining life prediction system for a wheel tread, characterized in that, including: An acquisition unit for acquiring the current health assessment result of the tread of the to-be-tested wheel set, where the health assessment result is at least divided into a first state for characterizing the normal operation of the tread of the to-be-tested wheel set and a second state for characterizing the tread ovality fault feature that has occurred on the tread of the to-be-tested wheel set; A determination unit for determining the remaining life prediction model of the tread of the to-be-tested wheel set according to the health assessment result, where the remaining life prediction model is one of a first remaining life prediction model corresponding to the first state and a second remaining life prediction model corresponding to the second state; A prediction unit for using the remaining life prediction model to perform remaining life prediction on the tread of the to-be-tested wheel set to obtain a remaining life prediction result; When the prediction unit performs the remaining life prediction on the tread of the wheel pair to be measured by using the remaining life prediction model and obtains the remaining life prediction result, it is specifically used for: Obtain the real-time monitoring data and the operating mileage of the tread of the wheel pair to be measured, wherein the real-time monitoring data includes the measured data of the out-of-round runout and the measured data of the vibration dB value; Use the remaining life prediction model to perform remaining life prediction according to the real-time monitoring data and the operating mileage, and obtain the remaining life prediction result; When the prediction unit performs the remaining life prediction according to the real-time monitoring data and the operating mileage by using the remaining life prediction model and obtains the remaining life prediction result, it is specifically used for: Perform data preprocessing on the measured data of the out-of-round runout, the measured data of the vibration dB value and the operating mileage respectively to obtain the out-of-round runout data, the vibration dB value data and the operating mileage data; Perform feature extraction on the out-of-round runout data, the vibration dB value data and the operating mileage data respectively to obtain the out-of-round runout feature, the vibration effective value feature and the operating mileage feature; Perform feature fusion on the out-of-round runout feature and the vibration effective value feature to obtain the fused degradation feature; Use the remaining life prediction model to perform remaining life prediction according to the fused degradation feature and the operating mileage feature, and obtain the remaining life prediction result.

Images