Method for generating performance degradation data of train air braking system

Abstract

The invention provides a method for generating performance degradation data of a train air braking system. The method comprises the following steps: on the basis of a brake system of a preset model, analyzing three stages of pressure rise, maximum pressure and pressure stability of pressure change of a brake cylinder, and extracting a plurality of typical fault modes; based on an extended random process model with non-monotonicity and independent incremental property, three stages of pressure change of a brake cylinder and multiple typical fault modes are combined, and linear, nonlinear or stepped performance degradation process modeling of the train air brake system is carried out; overlapping normal distribution random noise on a performance degradation track generated by modeling, simulating a control precision error of an actual braking system, and optimizing a performance degradation process model of the train air braking system; and utilizing the train air braking system performance degradation process model to generate train air braking system performance degradation data in batches.

Description

Technical Field

[0001] This invention relates to the field of rail transit braking technology, and more specifically, to a method for generating performance degradation data of a train air braking system. Background Technology

[0002] The air braking system of a train is a critical component ensuring the safety of rail transit operations, and its performance degradation directly affects the reliability and safety of train operation. With the development of rail transit technology, data-driven fault diagnosis and life prediction technologies have become important means to ensure the stable operation of braking systems. The application of these technologies highly relies on a large amount of performance degradation data that closely reflects actual operating conditions. Currently, the acquisition of train braking system degradation data mainly relies on two methods: field data collection and traditional simulation. Field data collection requires long-term tracking of train operation, which suffers from problems such as long cycles, high costs, and incomplete coverage of operating conditions, and it is difficult to obtain degradation data under extreme operating conditions. Traditional simulation methods are mostly based on simplified mathematical models and do not fully consider the multi-stage characteristics of brake cylinder pressure changes and the coupling effect of typical fault modes. Brake cylinder pressure changes include three stages: pressure rise to maximum pressure and pressure stabilization. These methods only simulate actual errors by simply superimposing noise, resulting in data that deviates significantly from the true degradation characteristics and has low fidelity.

[0003] Furthermore, existing technologies lack a unified framework for modeling different degradation types, such as linear, nonlinear, and step-type degradation. In particular, the criteria for dividing the stages of step-type degradation and the parameter constraints for nonlinear degradation are not clear enough, which further reduces the practicality of the data. Summary of the Invention

[0004] This invention provides a method for generating performance degradation data of a train air brake system, comprising: based on a preset model of brake system, analyzing the three stages of brake cylinder pressure change—pressure rise, maximum pressure, and pressure stabilization—and extracting multiple typical fault modes; based on an extended stochastic process model with non-monotonicity and independent increment, combining the three stages of brake cylinder pressure change and the multiple typical fault modes, modeling the linear, nonlinear, or step-like performance degradation process of the train air brake system; superimposing normally distributed random noise on the modeled performance degradation trajectory to simulate the control accuracy error of the actual brake system, thereby optimizing the performance degradation process model of the train air brake system; and using the train air brake system performance degradation process model to generate train air brake system performance degradation data in batches.

[0005] According to one embodiment of the present invention, the typical failure modes include at least three of the following: under-pressure, over-pressure, excessively rapid inflation, stepwise changes in the steady-state phase, and pressure oscillations in the steady-state phase.

[0006] According to one embodiment of the present invention, the extended stochastic process model is an extended Wiener process; the modeling formula for the linear performance degradation process is: in, The drift coefficient, The diffusion coefficient is... is the diffusion function.

[0007] According to one embodiment of the present invention, the nonlinear performance degradation process is achieved by combining a time-scale transformation function with the linear performance degradation process; the time-scale transformation function is: in, t For time, b It is a power exponent, when b >0 and b When <1, it is convex degenerate; when b When the value is greater than 1, it is a concave degeneration.

[0008] According to one embodiment of the present invention, the modeling method of the stepped performance degradation process is as follows: the degradation process is divided into stages based on the pressure rise stage, maximum pressure stage, pressure stabilization stage of the brake cylinder pressure change, and the typical failure modes corresponding to each pressure stage; within the same stage, a linear, convex, or concave performance degradation trajectory is used for modeling; different stages are switched by adjusting the drift coefficient and the diffusion coefficient, and then the trajectories of each stage are spliced ​​together to form a composite degradation trajectory.

[0009] According to one embodiment of the present invention, the mean of the normally distributed random noise is a preset mean, and the variance is a preset variance; when the trajectory is generated, the diffusion coefficient is controlled within a preset range so that the control accuracy of the brake cylinder pressure meets the preset pressure accuracy value.

[0010] According to one embodiment of the present invention, the power exponent b corresponding to the concave degradation is greater than 1 and close to 1; the power exponent corresponding to the convex degradation... b The value is less than 1 and close to 1.

[0011] According to one embodiment of the present invention, the maximum and minimum values ​​of the drift coefficient μ are determined by: using the steady-state pressure offset of the brake cylinder as a constraint, a fault state is determined when the offset exceeds a first preset offset, and the maximum offset does not exceed a second preset offset; combined with the identification capability of the degradation detection model, ensuring... μ The corresponding degradation rate can cover the entire degradation process from normal to fault, and can be effectively captured by the degradation detection model.

[0012] According to one embodiment of the present invention, the performance degradation trajectory and degradation pressure value of the train air braking system are generated as follows: the degree of degradation of the train air braking system is characterized by the offset between the steady-state pressure value of the brake cylinder and the standard pressure value; the larger the absolute value of the offset, the higher the degree of degradation; wherein the standard pressure value is the rated steady-state pressure value of a preset model brake system; the degradation trajectory of the train air braking system is composed of the offset data of the corresponding steady-state pressure values ​​during multiple air braking processes arranged in order of time or number of braking cycles; the degradation pressure value is obtained by adding the generated offset data to the standard pressure value.

[0013] According to one embodiment of the present invention, the batch generation of train air brake system performance degradation data includes: based on the modeling method and parameter settings of the extended stochastic process model, combined with actual braking process data, and superimposed with the normal distribution random noise, generating brake cylinder pressure data that conforms to the natural degradation characteristics; the change curve of the brake cylinder steady-state pressure value of a single air braking process is a single cycle data, and all the single cycle data are arranged in order of time or number of braking to form complete degradation trajectory data.

[0014] This invention effectively overcomes the shortcomings of existing methods for acquiring degradation data of train braking systems. It eliminates the need for long-term on-site tracking of trains, significantly shortening the data acquisition cycle and reducing costs. It also flexibly covers various operating conditions, including extreme conditions, completely solving the problems of incomplete coverage and difficulty in obtaining extreme condition data during on-site acquisition. This invention fully integrates the multi-stage characteristics of brake cylinder pressure changes with the coupled effects of typical fault modes. Brake cylinder pressure changes include three stages: pressure rise, maximum pressure, and pressure stabilization. The superimposed normally distributed random noise accurately simulates the control accuracy error of the actual braking system, rather than simply adding noise. This ensures that the generated data highly matches the actual degradation characteristics of the braking system, significantly improving fidelity.

[0015] Meanwhile, this invention constructs a unified modeling framework that is adaptable to different degradation types, including linear, nonlinear, and stepped degradation. It clarifies the stage division criteria for stepped degradation and the parameter constraints for nonlinear degradation, effectively solving the problems of inconsistent modeling frameworks and unclear parameters and division criteria in existing technologies, further enhancing the practicality of the generated data. Ultimately, the high-fidelity degradation data generated by this invention can provide reliable data support for subsequent data-driven fault diagnosis and life prediction technologies, helping to improve the reliability and safety of train braking systems, promoting the further development of rail transit braking technology, and possessing significant engineering application value. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the method for generating performance degradation data of a train air braking system provided by the present invention.

[0018] Figure 2 This is a flowchart of the modeling method for the stepped performance degradation process provided by the present invention.

[0019] Figure 3 This is a flowchart illustrating the method for determining the maximum and minimum values ​​of the drift coefficient μ provided by the present invention.

[0020] Figure 4 This is a flowchart illustrating the generation method of the performance degradation trajectory and degradation pressure value of the train air braking system provided by the present invention.

[0021] Figure 5 This is a flowchart provided by the present invention for batch generating performance degradation data of train air braking systems. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] The following is combined Figures 1-5 The present invention describes a method for generating performance degradation data of a train air braking system. Figure 1 This is a flowchart of the method for generating performance degradation data of a train air braking system provided by the present invention. Figure 1As shown, this invention provides a method for generating performance degradation data of a train air brake system, comprising: in step S100, based on a preset model brake system, analyzing the three stages of brake cylinder pressure change—pressure rise, maximum pressure, and pressure stabilization—and extracting multiple typical fault modes; in step S200, based on an extended stochastic process model with non-monotonicity and independent incrementality, combining the three stages of brake cylinder pressure change and multiple typical fault modes, modeling the linear, nonlinear, or step-like performance degradation process of the train air brake system; in step S300, superimposing normally distributed random noise on the modeled performance degradation trajectory to simulate the control accuracy error of the actual brake system and optimizing the train air brake system performance degradation process model; and in step S400, using the train air brake system performance degradation process model to generate train air brake system performance degradation data in batches.

[0024] Specifically, step S100 is the foundational preparation stage of the entire data generation method, accurately acquiring the basic characteristics and fault correlation patterns of the preset model braking system. First, the specific model of the target train's braking system needs to be determined, and the factory technical parameters and design documents of this model system need to be retrieved to clarify basic information such as the rated working pressure and normal operating sequence of its brake cylinders. Then, by building a test platform or collecting historical operating data of this model braking system, the pressure timing data of the brake cylinders in a complete braking cycle is obtained. Using data acquisition equipment, complete pressure curves are recorded for the pressure rise stage from the initial value to the peak value, the maximum pressure stage where the pressure remains at the peak value, and the pressure stabilization stage where the pressure drops from the peak value to a stable value. Based on the collected pressure curves, key features such as the pressure change rate, peak pressure fluctuation range, and stabilization stage pressure deviation are analyzed using feature extraction algorithms. Simultaneously, combined with common fault cases and maintenance records of this model braking system, typical fault modes such as under-inflation, over-inflation, excessively rapid inflation, stepwise changes in the stabilization stage, and pressure oscillation in the stabilization stage are extracted. The abnormal pressure characteristics corresponding to each fault mode in each pressure stage are clarified, providing a basis for fault correlation in subsequent modeling.

[0025] Step S200 is the modeling phase, focusing on achieving accurate modeling of different degradation types. The modeling process deeply integrates the stress stage characteristics and failure modes obtained in step S100. The extended stochastic process model selected in this step is specifically the extended Wiener process. This model possesses non-monotonicity and independent increments, enabling it to accurately fit the stochastic fluctuation characteristics of the braking system degradation process. For linear degradation processes, the extended Wiener process is directly used to construct the basic model, and its modeling formula is: in, This represents the initial steady-state pressure value of the brake cylinder. The drift coefficient characterizes the degradation rate. The diffusion coefficient is used to characterize the degree of random fluctuation. This is the standard Brownian diffusion function. For the nonlinear degradation process, a time-scale transformation function is incorporated into the above linear model. This is achieved by adjusting the power exponent. b The value of distinguishes between convex and concave degradation, when b A value greater than 0 and less than 1 indicates convex degradation, corresponding to scenarios where the degradation rate gradually slows down over time; when... b A value greater than 1 indicates concave degradation, corresponding to scenarios where the degradation rate gradually accelerates over time, and aligns with natural degradation characteristics. b The value should be close to 1. For the stepped degradation process, based on the three stages—pressure rise, maximum pressure, and pressure stabilization—determined in step S100 and their corresponding typical failure modes, the entire degradation process is divided into multiple continuous stages. Within the same stage, linear, convex, or concave trajectories are selected for modeling according to the degradation characteristics of that stage. The drift coefficient is adjusted between different stages. μ With diffusion coefficient δ By switching parameters, the trajectories of each stage are seamlessly spliced ​​together to form a complete composite degradation trajectory.

[0026] Step S300 aims to improve the fidelity of the generated data by optimizing the model through noise superposition. First, based on the pressure control accuracy requirements of the preset braking system model, the parameters of the normally distributed random noise are determined. The noise mean is set to a preset mean, typically 0, to ensure the unbiased influence of noise on the degradation trend. The variance is set to a preset variance. Simultaneously, the diffusion coefficient... δ The pressure is controlled within a preset range to ensure that the control accuracy of the brake cylinder pressure meets the preset pressure accuracy value. Then, the normal distribution random noise of the above parameters is superimposed on the degradation trajectory generated in step S200. This noise can accurately simulate the control accuracy error of the actual braking system caused by factors such as changes in sealing performance, pipeline pressure loss, and control element errors. This avoids the distortion problem caused by simply superimposing noise in traditional simulations, making the optimized model more consistent with the actual operating state.

[0027] Step S400 generates degradation data in batches, outputting degradation trajectory data that meets the requirements of actual applications. First, based on the model optimized in step S300, and combined with the actual braking cycle parameters of a preset braking system, different initial conditions and operating parameters are input to generate brake cylinder pressure data in batches. During this process, the change curve of the brake cylinder steady-state pressure value during a single air braking cycle is used as a single cycle data point. By repeatedly generating pressure data for multiple braking cycles, multiple sets of offset data between steady-state pressure values ​​and standard pressure values ​​are obtained. The standard pressure value is the rated steady-state pressure value of the preset braking system. Finally, all cycle data are arranged in chronological order or braking frequency order to form complete degradation trajectory data. This data can be directly used for subsequent fault diagnosis and life prediction algorithm training, model verification, and other scenarios.

[0028] According to an embodiment of the present invention, typical failure modes include at least three of the following: under-pressure, over-pressure, excessively rapid inflation, stepwise changes in the steady-state phase, and pressure oscillations in the steady-state phase.

[0029] Specifically, each typical failure mode is clearly associated with three stages of brake cylinder pressure change, and its characteristics and extraction logic are based on the actual operating rules and failure manifestations of the preset model brake system.

[0030] The characteristics of under-pressure are manifested in dual anomalies during the pressure rise and steady-state phases. During the pressure rise phase, the rate of pressure increase in the brake cylinder is significantly lower than the design threshold for this system model, and the pressure curve slope is gentle. After entering the maximum pressure phase, the peak pressure fails to reach the rated maximum pressure standard. Even after transitioning to the steady-state phase, the pressure value remains consistently lower than the rated steady-state pressure, and the deviation exceeds the allowable range. The identification of this failure mode requires comparing the pressure rise slope under normal operating conditions, the rated maximum pressure value, and the steady-state pressure benchmark to determine the threshold for under-pressure. Its causes are often related to minor leaks in the brake lines, insufficient opening of the air inlet valve, or low air supply pressure, and are a common progressive degradation manifestation of braking systems.

[0031] Overcharging and undercharging exhibit opposite characteristics. Overcharging primarily manifests as an excessively rapid pressure rise, quickly exceeding the rated maximum pressure and maintaining an overpressure state during the maximum pressure phase, with the overpressure exceeding the design tolerance. After entering the stabilization phase, although the pressure may slightly decrease, it remains above the rated steady-state pressure. When identifying this fault mode, it is crucial to focus on three key parameters: the peak overpressure value, the duration of overpressure, and the pressure deviation during the stabilization phase. The corresponding degradation scenarios often include stuck air charging valves and malfunctioning pressure control modules, representing typical manifestations of decreased braking system control accuracy.

[0032] The fault characteristics of excessively rapid inflation are concentrated in the pressure rise phase. The criteria for judgment are that the duration of the pressure rise phase is significantly shorter than the normal timing range of the preset system model, and the pressure increment per unit time far exceeds the design standard. Even if the final maximum pressure and stable pressure of the brake cylinder may meet the rated requirements under this fault mode, the excessively rapid pressure rise process will cause braking shock, and in the long run, may lead to accelerated wear of brake components, which is an important type of dynamic performance degradation of the braking system. During the refinement process, it is necessary to set a threshold for judging excessively rapid inflation by statistically analyzing the pressure rise time range under normal operating conditions, and to further confirm this by combining it with the steepness of the pressure rise rate curve.

[0033] The fault characteristic of stepped changes in the stable phase only appears during the pressure stabilization phase. This manifests as the brake cylinder pressure reaching its initial stable value but not remaining constant; instead, it exhibits phased, jump-like changes. After each pressure jump, a brief period of stability occurs, followed by another jump, with each jump's pressure amplitude exceeding the normal fluctuation range. Identifying this fault mode requires capturing the timing of pressure abrupt changes, the pressure amplitude of each jump, and the duration of each stable phase. The corresponding degradation scenarios are often pressure leakage fluctuations caused by wear of the brake cylinder seals, and regulation lag caused by increased clearance in the pressure regulating mechanism—typical manifestations of phased degradation.

[0034] The characteristics of pressure oscillation during the steady-state phase are the absence of a distinct constant pressure range. The brake cylinder pressure fluctuates frequently around the rated steady-state pressure, with high fluctuation frequency and amplitude exceeding the allowable range. The pressure curve exhibits an irregular sawtooth or wavy shape. When identifying this fault mode, it is necessary to establish judgment criteria by calculating parameters such as pressure fluctuation frequency, the difference between peak and trough values, and fluctuation period. Its causes are often related to pressure sensor signal interference, abnormal control loop feedback adjustment, or aging and deformation of internal brake cylinder seals. It is a typical fault form of brake system control accuracy degradation.

[0035] In actual implementation, it is necessary to select at least three of the above five modes for modeling, taking into account the operating environment, protection requirements, and common fault types of the pre-designed braking system. For example, for braking systems commonly used in urban rail transit, three high-frequency fault modes can be prioritized: under-pressure, step change in the stable phase, and pressure oscillation in the stable phase. For long-distance train braking systems, three fault modes related to long-distance operation losses—over-pressure, excessively rapid inflation, and under-pressure—can be included to ensure that the modeling basis is highly consistent with the actual application scenario.

[0036] According to an embodiment of the present invention, the extended stochastic process model is an extended Wiener process; the modeling formula for the linear performance degradation process is: in, The drift coefficient, The diffusion coefficient is... is the diffusion function.

[0037] Specifically, this invention selects the extended Wiener process as the extended stochastic process model because its non-monotonicity and independent increment can accurately match the essential characteristics of the degradation process of the train air braking system. It has both predictable gradual performance degradation and random fluctuations caused by environmental fluctuations, minor wear of components, etc. Compared with traditional deterministic models or simple stochastic models, it can better restore the real degradation law of the braking system.

[0038] Modeling formula for linear performance degradation process In this system, each parameter corresponds to the actual physical meaning and engineering characteristics of the braking system, and the values ​​are all based on the basic parameters and operating rules of the preset model braking system.

[0039] The initial steady-state pressure value of the brake cylinder is directly derived from the factory technical parameters of the preset model brake system. That is, the rated pressure reference value of the stable stage after the system completes braking in a brand-new state and under normal working conditions. It is usually determined by retrieving the design manual or factory test report of the brake system and is the starting reference of the entire degradation trajectory.

[0040] The drift coefficient characterizes the average rate of linear degradation of the braking system. Its physical meaning is the average deviation of the steady-state pressure of the brake cylinder per unit time (or per unit number of braking cycles). The value must strictly adhere to the following constraint logic: based on the steady-state pressure offset of the brake cylinder, when the offset exceeds the first preset offset, i.e., the fault judgment threshold, the system determines it to be in a fault state, while the maximum offset must not exceed the second preset offset, i.e., the ultimate safety threshold; simultaneously, it must be combined with the recognition capability of the degradation detection model to ensure... The corresponding degradation rate can cover the entire cycle from normal state to fault state, for example from The offset to the fault threshold needs to be... It matches the degradation cycle and can be effectively captured by the detection model, avoiding detection omissions due to excessively fast degradation rates or data redundancy due to excessively slow degradation rates. In practical applications, This can be obtained through statistical analysis of historical degradation data of the braking system of that model, for example, a certain model of braking system. The value can be 0.2~0.8 kPa / thousandth braking cycles, and the specific value can be adjusted according to the wear characteristics and failure patterns of the system.

[0041] As a time variable, its measurement dimension can be selected according to the actual application scenario; it can be either continuous running time (unit: hours) or cumulative braking count (unit: times). It needs to be consistent with... The units should be kept consistent to ensure uniformity in degradation rate calculation.

[0042] The diffusion coefficient characterizes the intensity of random fluctuations during the degradation process of the braking system. Its physical meaning is a quantitative indicator of the random fluctuation amplitude of the steady-state pressure offset in the brake cylinder. The unit must match the measurement dimension of the time variable to ensure dimensional consistency in the modeling formula. Since the dimension of the diffusion function in the modeling formula is the square root of the time measurement dimension, The unit needs to be set as the ratio of the pressure unit to the square root of the time measurement dimension, for example, when time is measured in thousands of braking cycles. The unit is When time is expressed in hours, The unit is .

[0043] The value must match the control accuracy requirements of the brake cylinder pressure, and be controlled within a preset range to ensure that the control accuracy of the brake cylinder pressure meets the preset pressure accuracy value. For example, if the preset pressure accuracy value is ±20 kPa, and the time is measured in thousands of braking cycles, Values ​​can range from 1 to 3 When time is expressed in hours, Values ​​can range from 1 to 3 Such a value can avoid random fluctuations exceeding the control accuracy range of the actual braking system, and can also accurately simulate the effects of random disturbances such as brake line pressure loss and control element response errors, making the generated degradation trajectory more consistent with actual working conditions.

[0044] The standard Brownian motion, or diffusion function, is a normally distributed random process with a mean of 0 and a variance of t. It is used to quantify unpredictable random factors during the degradation of braking systems, such as fluctuations in ambient temperature, instantaneous changes in air source pressure, and minute errors in sensors. The introduction of this feature means that the linear degradation trajectory is not an absolute straight line, but rather exhibits a pattern of recursion. The random fluctuations of the trend line perfectly match the dual characteristics of deterministic trends plus random disturbances in the actual braking system degradation process.

[0045] The advantage of the entire linear modeling formula is that by combining deterministic degradation trends with random fluctuations, it not only ensures the regularity of the degradation trajectory but also restores the randomness of actual working conditions. It can accurately simulate the linear degradation process of the braking system under progressive failure modes such as slight wear and slow leakage, such as the initial stage of pressure underfilling failure and slight pressure deviation in the stable stage. This provides a unified basic framework for subsequent nonlinear and step-type degradation modeling.

[0046] According to an embodiment of the present invention, the nonlinear performance degradation process is achieved by combining a time-scale transformation function with a linear performance degradation process; the time-scale transformation function is: in, t For time, b It is a power exponent, when b >0 and b When <1, it is convex degenerate; when b When the value is greater than 1, it is a concave degeneration.

[0047] Specifically, the modeling logic for the nonlinear performance degradation process is based on the extended Wiener process linear model, using a time-scale transformation function. By applying a nonlinear mapping to the time variable, the actual characteristics of the braking system's degradation rate dynamically changing over time can be accurately fitted. Compared to the uniform degradation law described by the linear model, the nonlinear model can better reproduce the real degradation process of the braking system under complex scenarios such as accelerated component wear, deterioration of sealing performance, and decrease in the precision of control elements, thus adapting to the actual needs of gradually changing rates during fault evolution.

[0048] Time scale transformation function Its function is through the power exponent b Its flexible adjustment alters the stretching or compression effect of the time dimension, thereby adjusting the curvature of the degradation trajectory and achieving precise characterization of different nonlinear trends. Its integration with linear models is simple and unified; it only requires changing the time variable in the linear modeling formula. t Replace with The formula for nonlinear degradation modeling can then be obtained. This combination not only fully preserves the advantages of the extended Wiener process in terms of deterministic trends and random fluctuations, but also endows the model with nonlinear expressive capabilities through time scale transformation, ensuring the unity and coherence of the linear and nonlinear modeling frameworks.

[0049] Power index b When the value is greater than 0 and less than 1, a convex degradation mode is formed. In this case, the time-scale transformation function exhibits a time-stretching effect, increasing with time... t The increase, The growth rate gradually slows down, which is reflected in the degradation trajectory as a gradual decrease in the degradation rate of the braking system over time. This degradation pattern corresponds to the actual scenario where the braking system degrades rapidly in the early stages and then gradually flattens out. For example, during the break-in period of a new braking system, there is initial minor wear on components such as brake cylinder seals and pipe interfaces, leading to a rapid increase in pressure deviation. However, as the fit between components improves, the wear rate gradually slows down, and the degradation trend gradually stabilizes. Or, in the middle stage of a pressure underfill fault, the pipe leakage no longer continues to increase, and the pressure deviation rate gradually converges. The power exponent corresponding to convex degradation b The value should be less than 1 and close to 1, such as 0.7, 0.8, 0.9, etc., to avoid [further issues]. b If the value is too small, the degradation rate will decrease too quickly, deviating from the natural degradation pattern of the braking system. The actual value needs to be determined based on the historical operating data and wear characteristics of the preset model braking system to ensure a smooth transition of the trajectory and conform to the degradation logic of mechanical components after they have been broken in and stabilized.

[0050] Power index b When the value is greater than 1, a concave degradation mode is formed. At this point, the time-scale transformation function exhibits a time compression effect, increasing with time... t The increase, The growth rate gradually accelerates, reflected in the degradation trajectory as a gradual increase in the degradation rate of the braking system over time. This degradation pattern corresponds to the actual scenario of slow degradation in the early stage and rapid failure in the later stage of the braking system. For example, in the later stage of aging of brake cylinder seals, the sealing performance drops sharply, leading to an accelerated increase in pressure leakage; or in the worsening stage of pressure overcharging failure, the degree of sticking of the air valve increases, and the pressure overcharging amplitude expands rapidly with the increase of braking frequency; or in the process of intensified pressure oscillation in the stable stage, the fluctuation amplitude and frequency increase significantly with the degradation of control loop performance. The power exponent corresponding to concave degradation b The value must be greater than 1 and close to 1, such as 1.1, 1.2, 1.3, etc., to avoid [further issues]. b Taking excessively large values ​​leads to a surge in the degradation rate, which does not conform to the actual characteristics of the gradual degradation of the braking system. The specific value needs to be determined in conjunction with the failure evolution law to ensure that the degradation rate increases gradually and naturally, conforming to the gradual evolution logic of the component from minor faults to serious failures.

[0051] The advantage of the entire nonlinear modeling process lies in its ability to comprehensively cover two typical nonlinear degradation modes through flexible adjustment of a single timescale transformation function and power exponent, while sharing the basic framework of the extended Wiener process with linear models, ensuring the uniformity of the modeling system. Simultaneously, bSetting the value close to 1 further ensures that the generated nonlinear degradation trajectory conforms to the natural degradation characteristics of the braking system, avoids extreme nonlinear distortion, and can accurately simulate the complex degradation process under different fault scenarios, providing reliable support for the subsequent generation of high-fidelity degradation data.

[0052] Figure 2 This is a flowchart illustrating the modeling method for the stepped performance degradation process provided by this invention. For example... Figure 2 As shown, according to an embodiment of the present invention, the modeling method of the stepped performance degradation process is as follows: In step S210, the degradation process is divided into stages based on the pressure rise stage, maximum pressure stage, pressure stabilization stage of the brake cylinder pressure change, and the typical fault modes corresponding to each pressure stage; in step S220, linear, convex, or concave performance degradation trajectories are used to model the same stage; in step S230, different stages are switched by adjusting the drift coefficient and diffusion coefficient, and then the trajectories of each stage are spliced ​​together to form a composite degradation trajectory.

[0053] Specifically, the modeling of the stepped performance degradation process is achieved by using the logic of stage division, single-stage modeling, and cross-stage splicing to accurately restore the discontinuous degradation law of the braking system under the superposition of multiple fault modes and the characteristic changes of different pressure stages. This corresponds to the flowchart in Figure 2.

[0054] Step S210's stage division needs to deeply integrate the natural temporal sequence of brake cylinder pressure changes with the correlation characteristics of fault modes to ensure that the division is based on actual degradation logic. The pressure rise, maximum pressure, and stable phases of the brake cylinder pressure are the natural temporal boundaries of the degradation process, and each typical fault mode corresponds to an abnormal performance at a specific pressure phase. The combination of these two constitutes a dual basis for stage division. For example, if a degradation process initially only exhibits excessively rapid inflation during the pressure rise phase, followed by a step-like change in the stable phase in the middle stage, and pressure oscillations during the stable phase in the later stage, then it can be divided into three consecutive stages. If the fault mode is concentrated only on overcharging during the maximum pressure phase, and the degree of overcharging gradually changes over time, then the process can be divided into three stages: mild overcharging, moderate overcharging, and severe overcharging. During the division, it is necessary to ensure that only one type of fault mode or a set of associated fault characteristics exists within each stage to avoid cross-interference between different fault types, laying the foundation for accurate single-stage modeling in the future.

[0055] In step S220, trajectory modeling within the same stage requires selecting an adaptation type from linear, convex, or concave patterns based on the degradation rate characteristics of that stage, and the modeling process remains based on the unified framework of the extended Wiener process. If the degradation rate is uniform and stable within the stage, such as in the mild under-pressure stage where the brake cylinder pressure offset increases uniformly with time, then linear degradation trajectory modeling is used, following the established pattern. Formula; If the degradation rate gradually slows down within a stage, such as the slight over-inflation during the break-in period of a new braking system, and the abnormality in the inflation rate gradually converges as the fit of the components improves, then a convex degradation trajectory model is adopted, using a time-scale transformation function. (0 < b <1) Adjusting the time variable; if the degradation rate gradually intensifies within a stage, such as the worsening stage of an overcharge fault, where the air valve jamming worsens and the overcharge amplitude rapidly increases with the number of braking cycles, then a concave degradation trajectory model is adopted, in which case the power exponent is used. b The value should be greater than 1 and close to 1. When modeling a single stage, parameter consistency must be maintained to ensure the regularity and coherence of the degradation trajectory within that stage, while also using the diffusion coefficient... The appropriate value should be selected to preserve the random fluctuation characteristics within this stage.

[0056] Step S230 achieves a smooth transition and complete splicing between different stages, the key being the use of the drift coefficient. With diffusion coefficient The dynamic adjustment matches the differences in degradation characteristics at each stage. Drift coefficient Characterizing the degradation rate, different failure modes occur at different stages, and therefore the degradation rate will inevitably differ. For example, in the stage of overinflation... The value needs to match the degree of anomalousness in the rate of pressure rise, while the stepwise change during the steady phase... The value needs to be adapted to the amplitude and frequency of the pressure jump, by adjusting... The value allows for phase switching of the degradation rate. Diffusion coefficient. Characterizing the intensity of random fluctuations, different stages of the fault mode correspond to different random disturbance sources. For example, the fluctuation intensity in the pressure oscillation stage is much greater than that in the mild pressure underfill stage, and the intensity needs to be increased accordingly. The values ​​should closely match reality. After parameter adjustment, the trajectories of each stage need to be seamlessly spliced ​​in chronological order. During splicing, it is essential to ensure that the endpoint pressure offset of the previous stage remains continuous with the starting offset of the next stage, avoiding any abrupt changes without physical meaning, ultimately forming a complete composite degradation trajectory. For example, in a three-stage degradation process, the endpoint offset of the first stage (linear degradation) is 5 kPa. The second stage (convex degradation) should start at 5 kPa, and the third stage (concave degradation) should then inherit the endpoint offset of the second stage. Precise parameter matching achieves a natural connection of the trajectories.

[0057] The advantage of the entire stepped modeling process is that it retains the unified framework of the extended Wiener process, and through stage division and parameter switching, it adapts to the complex degradation scenarios of multiple fault modes and multiple stages of braking systems. It can accurately simulate the real degradation process in actual operation where one fault dominates, multiple faults overlap, and fault type switching occurs. The generated trajectory data has better scene coverage and fidelity, providing training data that is closer to actual working conditions for subsequent fault diagnosis technology.

[0058] According to an embodiment of the present invention, the mean of the normally distributed random noise is a preset mean, and the variance is a preset variance; when the trajectory is generated, the diffusion coefficient is controlled within a preset range so that the control accuracy of the brake cylinder pressure meets the preset pressure accuracy value.

[0059] Specifically, the introduction of normally distributed random noise is to accurately simulate the unavoidable control precision errors during the actual operation of the braking system. The preset mean and variance must strictly adhere to the actual operating characteristics of the braking system to ensure higher fidelity of the data after noise superposition. The preset mean is chosen based on the principle of unbiased interference, typically set to 0. This avoids the noise from causing a systematic shift in the deterministic trend of the degradation trajectory, ensuring that the characteristics of the degradation law are not distorted by noise interference. The preset variance needs to be determined based on statistical analysis of historical operating data of a preset braking system model. By extracting the dispersion information of pressure fluctuations during actual braking, the variance is set to a value that matches the actual fluctuation level. For example, if the random pressure fluctuation dispersion of a certain braking system model is 3 kPa... 2 The corresponding noise variance can be preset to 3 kPa. 2 This ensures that the noise accurately reflects the impact intensity of random interference factors such as pipeline pressure loss, control element response delay, and minor fluctuations in gas source pressure.

[0060] The range control of the diffusion coefficient during trajectory generation, together with the parameter setting of the normally distributed random noise, forms a synergistic constraint to ensure that the control accuracy of the brake cylinder pressure meets the preset requirements. The diffusion coefficient, as a key parameter characterizing the intensity of random fluctuations in the degradation process, needs to have its value range determined by reverse derivation from the preset pressure accuracy value. The preset pressure accuracy value is a performance indicator of the braking system, clearly defined by the design standards of the preset model braking system; for example, ±20 kPa, meaning that the deviation between the actual and theoretical values ​​of the brake cylinder pressure must not exceed this range. To meet this requirement, the maximum value of the diffusion coefficient needs to be controlled within a range that allows the random fluctuation portion ( The total fluctuation amplitude after superimposing the normalized random noise on the pressure accuracy threshold does not exceed the preset pressure accuracy threshold. For example, when the preset pressure accuracy value is ±20 kPa, the standard deviation of the normalized random noise is 1.73 kPa and the corresponding variance is 3 kPa. 2 At this time, the diffusion coefficient needs to be controlled between 1 and 3. Between braking, ensure that the total fluctuation amplitude after the two are superimposed does not exceed ±20 kPa in the 95% confidence interval, so as to avoid the generated data deviating from the control capability boundary of the actual braking system due to excessive random fluctuations.

[0061] Meanwhile, the range of the diffusion coefficient needs to be dynamically adapted to the different operating stages and fault modes of the braking system. For example, in the fault mode of pressure oscillation during the stable phase, the intensity of random fluctuations in the actual system is relatively large, so the upper limit of the diffusion coefficient can be appropriately widened; while in the initial stage of mild pressure underfill, the system fluctuations are relatively gentle, so the range of the diffusion coefficient needs to be tightened to avoid excessive amplification of random fluctuations. This dynamically adjusted range control method not only ensures that the pressure control accuracy meets the preset requirements, but also allows the generated degradation data to conform to the actual fluctuation characteristics under different fault scenarios, further improving the engineering applicability of the data.

[0062] The coordinated design of the entire noise parameter preset and diffusion coefficient control breaks the limitation of simply superimposing fixed intensity noise in traditional simulation. By setting precise parameters and range constraints based on the actual system characteristics, the generated degradation trajectory contains real random disturbances without deviating from the control accuracy boundary of the braking system, providing high-fidelity data support that is closer to the actual working conditions for subsequent fault diagnosis and life prediction technologies.

[0063] According to an embodiment of the present invention, the power exponent corresponding to concave degradation The value is greater than 1 and close to 1; the power exponent corresponding to convex degeneration. The value is less than 1 and close to 1.

[0064] Specifically, the value of the power exponent b is limited to a range close to 1, which is in line with the gradual characteristics of the natural degradation of the train's air braking system. This avoids distortion of the degradation trajectory due to excessive deviation from 1, and ensures that the generated data conforms to the actual evolution laws such as wear of mechanical parts and deterioration of sealing performance.

[0065] The power exponent corresponding to convex degeneration b Values ​​less than 1 and close to 1 are typically set between 0.7 and 0.9, such as 0.85 or 0.9. The design logic behind this range is that convex degradation is a gradual slowdown in the degradation rate, but this slowdown is a smooth transition, not a sudden halt. If... bValues ​​that are too small, such as 0.3 or 0.5, will cause the growth rate of the time-scale transformation function to slow down too quickly, and the degradation trajectory will stabilize in a short period of time, which does not conform to the evolutionary patterns of actual scenarios such as the break-in of braking system components and the convergence of minor faults. For example, during the break-in of the seals of a new braking system, the pressure offset increases rapidly in the early stage, but as the fit of the components improves, the wear rate gradually slows down. This slowdown is gradual and will not stop abruptly. On the other hand, values ​​of b close to 1, such as 0.9, allow the growth rate of the time-scale transformation function to decrease slowly. The degradation trajectory shows the characteristics of rapid growth in the early stage, slowing down in the later stage, but still continuing to degrade. This accurately reproduces the real degradation state of such scenarios, reflecting both the characteristics of rate decay and ensuring the continuity of the degradation process.

[0066] Power exponent corresponding to concave degeneration b Values ​​greater than and close to 1 are typically set between 1.1 and 1.3, such as 1.15 or 1.2. The design logic behind this value is that the degradation rate gradually increases, but this increase is gradual, not abrupt. If the value of b is too large, such as 2.0 or 3.0, it will cause the time-scale transformation function to grow too rapidly, and the degradation trajectory to deteriorate sharply in a short period of time, which does not conform to the actual laws of aging and fault deterioration in braking systems. For example, in the later stages of aging of brake cylinder seals, the decrease in sealing performance leads to an increase in pressure leakage, but the deterioration process is gradual and will not suddenly change from slight leakage to severe leakage. b Values ​​close to 1, such as 1.2, can make the growth rate of the time scale transformation function increase slowly, and the degradation trajectory shows the characteristics of initial smoothness and gradual aggravation in the later stage. This accurately restores the degradation process of scenarios such as aging of seals and aggravation of air valve jamming. It not only reflects the characteristics of increasing rate, but also avoids abrupt changes without physical meaning, ensuring that the degradation logic is consistent with the actual failure law of the components.

[0067] This near-1 value constraint, in conjunction with the extended Wiener process modeling framework, ensures that the nonlinear degradation trajectory is distinct from linear uniform degradation while remaining true to the physical characteristics of the braking system. Whether it's the gradual decrease of convex degradation or the gradual increase of concave degradation, the degradation process maintains a smooth transition. This results in performance degradation data that possesses both nonlinear characteristics and closely matches the gradual degradation patterns observed in actual operation, further enhancing data fidelity and engineering applicability, and providing a more reliable training basis for subsequent fault diagnosis and lifespan prediction.

[0068] Figure 3 The drift coefficient provided by this invention μ The flowchart shows how the maximum and minimum values ​​are determined. (See attached flowchart.) Figure 3 As shown, according to an embodiment of the present invention, the drift coefficient The maximum and minimum values ​​are determined as follows: In step S310, the steady-state pressure offset of the brake cylinder is used as a constraint; if the offset exceeds the first preset offset, it is determined to be a fault state, and the maximum offset does not exceed the second preset offset; In step S320, the recognition capability of the degradation detection model is combined to ensure... The corresponding degradation rate can cover the entire degradation process from normal to fault, and can be effectively captured by the degradation detection model.

[0069] Specifically, the drift coefficient μ Determining the maximum and minimum values ​​requires a dual logic combining physical constraints and technological adaptation. Steps S310 and S320 form a progressive constraint relationship to ensure... The value of conforms to the physical operating laws of the braking system and meets the technical requirements of subsequent fault diagnosis. Figure 3 The flowchart corresponds to this.

[0070] Step S310 uses the steady-state pressure offset of the brake cylinder as a physical constraint to clarify... The first and second preset offsets are derived from the design standards and safety specifications of the preset braking system models and are key thresholds for defining the system state. The first preset offset is a fault determination threshold, determined by the performance indicators of the braking system model. For example, if the first preset offset of a certain system model is set to 15 kPa, it means that when the deviation between the steady-state pressure and the standard pressure of the brake cylinder exceeds 15 kPa, the system is judged to be in a fault state. The second preset offset is a limit safety threshold, which is the bottom line for ensuring the safe operation of the train. For example, if it is set to 30 kPa, it means that the offset must not exceed this value under any circumstances, otherwise it will cause safety risks. Based on these two thresholds, The value must satisfy the full-cycle degradation logic: from the initial state (offset is 0) to the fault state (offset reaches the first preset offset), and then to the extreme safety state (offset does not exceed the second preset offset). The corresponding degradation rate must match the degradation cycle. For example, if a certain model of system is designed for a degradation cycle of 100,000 braking cycles, with a first preset offset of 15 kPa and a second preset offset of 30 kPa, then... The minimum value must ensure that it can reach 15 kPa from 0 deflection within 100,000 braking cycles (i.e. ≥15kPa / 100,000 braking cycles = 0.00015kPa / cycle), the maximum value must ensure that the offset within 100,000 braking cycles does not exceed 30kPa (i.e. ≤30kPa / 100,000 braking cycles = 0.0003kPa / cycle), which is clearly derived through this reverse derivation. The physical boundary.

[0071] Step S320, based on physical constraints, further calibrates the degradation detection model by combining its technical characteristics. The scope of degradation detection models is limited; they cannot capture degradation processes that are too slow or too fast. Too small, and the degradation rate is too slow, for example, in a certain model of system. =0.00005 kPa / cycle, reaching the fault threshold of 15 kPa requires 300,000 braking cycles, far exceeding the actual degradation cycle, which will lead to data redundancy and untimely fault warnings; if If it is too large, the degradation rate will be too fast, for example =0.0005 kPa / cycle, requiring only 30,000 braking cycles to exceed the fault threshold, which is inconsistent with the natural wear pattern of braking system components, and may cause the detection model to miss key fault characteristics due to the excessively rapid degradation process. Therefore, The value of needs to be adapted to the sampling frequency and recognition sensitivity of the detection model. For example, if the sampling period of the detection model is 1000 braking cycles and the minimum degradation rate of recognition is 0.0001 kPa / cycle, then It needs to be set between 0.0001 kPa / sample and 0.0003 kPa / sample to ensure that each sampling can capture identifiable degradation increments, covering the complete process from normal to fault, and can be accurately captured by the detection model to avoid detection blind spots or invalid data.

[0072] entire The process of determining the value involves defining a reasonable range through physical constraints in step S310, followed by precise range calibration through technical adaptation in step S320, ultimately forming... The maximum and minimum values ​​not only conform to the natural degradation patterns of braking system components such as wear and tear and deterioration of sealing performance, but also provide effective support for subsequent data-driven fault diagnosis techniques. For example, after a certain model of braking system undergoes dual constraints, The value range is determined to be from 0.00015 kPa / cycle to 0.00025 kPa / cycle. This range ensures that the full cycle degradation from normal to fault is completed within 100,000 braking cycles, and also allows the detection model to capture degradation increments of 0.15 kPa to 0.25 kPa in every 1,000 braking cycles. This perfectly matches the model's recognition capability and ensures that the generated degradation data has engineering applicability.

[0073] Figure 4 This is a flowchart illustrating the generation method of the performance degradation trajectory and degradation pressure value of the train air braking system provided by this invention. Figure 4As shown, according to an embodiment of the present invention, the performance degradation trajectory and degradation pressure value of the train air braking system are generated as follows: In step S410, the degree of degradation of the train air braking system is characterized by the offset between the steady-state pressure value of the brake cylinder and the standard pressure value. The larger the absolute value of the offset, the higher the degree of degradation. The standard pressure value is the rated steady-state pressure value of a preset model brake system. In step S420, the degradation trajectory of the train air braking system is formed by arranging the offset data of the steady-state pressure values ​​corresponding to multiple air braking processes in order of time or number of braking cycles. In step S430, the degradation pressure value is obtained by adding the generated offset data to the standard pressure value.

[0074] Specifically, the generation of the performance degradation trajectory and degradation pressure value of the train air braking system is achieved through a progressive logic of degradation degree quantification, trajectory construction, and pressure value calculation. Steps S410, S420, and S430 are closely linked to ensure that the generated results not only conform to the physical operating laws of the braking system but also meet the application requirements of subsequent fault diagnosis and life prediction. Figure 4 The flowchart corresponds to this.

[0075] Step S410 establishes a direct correlation between the degree of degradation and the pressure offset, providing a quantitative basis for subsequent data generation. Determining the standard pressure value is a prerequisite; it directly originates from the factory technical specifications of the preset model braking system. This means that after the system completes braking in a brand-new state under normal operating conditions, the rated pressure value during the stable phase is the same. For example, for the preset model EP2002 braking system, its rated steady-state pressure value (standard pressure value) is typically set at 500 kPa. This value serves as the benchmark for measuring system degradation. The brake cylinder steady-state pressure value is the actual pressure measurement value after the system enters the stable phase during each braking process. The difference between this value and the standard pressure value is the offset. The absolute value of the offset directly reflects the degree of degradation. The essence of this correlation logic is based on the braking system's performance indicator, pressure control accuracy. When the system experiences degradation phenomena such as wear, leakage, or aging of control components, the steady-state pressure will inevitably deviate from the rated standard, and the more severe the degradation, the greater the deviation. For example, when the pressure is slightly insufficient, the offset may be -30 kPa (absolute value 30 kPa); as the leakage worsens, the offset may increase to -80 kPa (absolute value 80 kPa), which intuitively reflects the progression of the degradation.

[0076] Step S420 focuses on constructing the degradation trajectory, reconstructing the temporal evolution of degradation through the ordered arrangement of data. Actual train operation involves multiple continuous or intermittent air braking processes. Each braking action generates a corresponding set of steady-state pressure offset data, which forms the basic units for trajectory construction. The choice of arrangement method must align with the actual application scenario: if analyzing the degradation pattern of the system over time is required, the data can be arranged chronologically according to the braking process, for example, recording the offset of representative braking actions per hour; if focusing on the impact of the number of braking actions on degradation is required, the data can be arranged chronologically according to the cumulative number of braking actions, for example, recording the offset data once every 100 braking actions. The significance of this orderly arrangement lies in restoring the continuity and gradualness of degradation. For example, in the initial 5000 braking cycles of a certain braking system, the offset gradually increases from 0 kPa to -20 kPa. Subsequently, in the 5000 to 10000 braking cycles, the offset slowly expands from -20 kPa to -50 kPa. After arranging according to the number of braking cycles, the resulting trajectory can show the stage characteristics of slow degradation and accelerated degradation, accurately matching the evolution law of faults such as seal wear and pipeline leakage.

[0077] Step S430, which calculates the final degradation pressure value, is a crucial step connecting the offset and the actual braking pressure. Its calculation logic directly stems from physical meaning: the standard pressure value is the ideal pressure of the system without degradation, and the offset is the pressure deviation caused by degradation. Adding the two together yields the actual brake cylinder pressure value under degradation conditions. For example, the standard pressure value of the EP2002 braking system is 500 kPa. If the offset during a braking test is -30 kPa, the corresponding degradation pressure value is 500 kPa + (-30 kPa) = 470 kPa. When degradation intensifies and the offset becomes -80 kPa, the degradation pressure value becomes 500 kPa + (-80 kPa) = 420 kPa. This calculation method ensures that the degradation pressure value accurately reflects the actual operating state of the system, retaining the baseline properties of the standard pressure while incorporating degradation characteristics through the offset, thus avoiding the limitations of relying solely on theoretical or measured values. At the same time, the calculation results correspond to the degradation trajectory constructed in step S420. The offset corresponding to each node on the trajectory can be converted into a specific degradation pressure value through this formula, forming a complete data chain of time / number of times, offset, and degradation pressure value.

[0078] The advantage of this entire generation process lies in its ability to establish an objective evaluation standard for the degree of degradation through quantified offsets, to reconstruct the dynamic evolution of degradation through orderly arrangement, and to output realistic pressure data through precise calculation. These three elements work together to ensure high fidelity in the degradation trajectory and degradation pressure values. The generated trajectory data can present the complete degradation path of the system from normal state to minor faults and then to severe faults, while the degradation pressure values ​​can be directly used to verify the accuracy of fault diagnosis algorithms, providing reliable basic data support for subsequent data-driven maintenance technology of rail transit braking systems.

[0079] Figure 5 This is a flowchart provided by the present invention for batch generating performance degradation data of train air braking systems. For example... Figure 5 As shown, according to an embodiment of the present invention, batch generation of train air brake system performance degradation data includes: in step S440, based on the modeling method and parameter settings of the extended stochastic process model, combined with actual braking process data, and superimposed with normally distributed random noise, generating brake cylinder pressure data that conforms to the natural degradation characteristics; in step S450, the change curve of the brake cylinder steady-state pressure value of a single air braking process is a single cycle data, and all single cycle data are arranged in order of time or number of braking to form complete degradation trajectory data.

[0080] Specifically, the process of batch generating performance degradation data of train air braking systems is achieved through a progressive logic of single-cycle data generation and full-track data integration. These two steps are closely linked, ensuring both the natural degradation characteristics of individual data sets and the integrity and engineering practicality of the overall trajectory. Figure 5 The flowchart corresponds to this.

[0081] The key to step S440 is generating single-cycle brake cylinder pressure data with natural degradation characteristics, which requires a full integration of four key elements: modeling methods, parameter constraints, actual data calibration, and noise superposition. The modeling method for the extended stochastic process model is the extended Wiener process framework, which covers modeling logic for linear, nonlinear, and step-degradation processes. The parameter settings are precise values ​​obtained after adaptation to physical constraints and the detection model, including the drift coefficient. diffusion coefficient Power index etc., for example Set the pressure to 0.00015 to 0.00025 kPa each time. Controlled between 1 and 3 brake, Values ​​range from 0.8 to 1.2. Incorporating actual braking process data enhances data fidelity. This actual data includes real braking cycle parameters of the preset braking system model, such as the duration of the pressure rise phase, the maximum pressure maintenance time, and the fluctuation range of the stable phase. It also includes pressure response characteristics under different operating conditions, such as the pressure changes during start-up braking and emergency braking. By embedding these actual characteristics into the modeling process, the generated pressure data closely matches real braking behavior. For example, in emergency braking scenarios, the pressure rise rate is faster, and the parameters for this phase are adjusted accordingly during modeling. Based on this, normally distributed random noise is superimposed. The noise parameters are consistent with those mentioned earlier, with a mean typically of 0 and a variance determined statistically based on actual fluctuations. The superposition occurs after the single-cycle pressure curve is generated. By superimposing random noise at each pressure sampling point, the control accuracy error of the braking system is simulated. For example, random fluctuations of ±1 to 3 kPa are superimposed on the stable phase pressure value to prevent the pressure curve from becoming ideally smooth and to present natural, small fluctuation characteristics. The final generated brake cylinder pressure data must fully include the three stages of pressure rise, maximum pressure, and pressure stabilization, and be able to reflect the degradation characteristics of the corresponding fault mode. For example, in the overcharge mode, the pressure value in the maximum pressure stage is higher than the standard pressure and gradually increases as the degradation process progresses.

[0082] The key to step S450 is integrating discrete single-cycle data into a complete degradation trajectory, with the crucial aspect being the ordered arrangement of the data and the continuity of the trajectory. A single cycle data point is the complete pressure change curve of a single air braking process. Each curve contains key information such as time or number of braking operations, pressure values ​​at each stage, and steady-state pressure offset. For example, in the cycle data of a certain braking operation, the pressure rise phase lasts 2 seconds, the maximum pressure is 520 kPa, the steady-state pressure is 510 kPa, and the offset is 10 kPa. The arrangement method needs to be selected according to the application scenario. When arranged in chronological order, the actual time of the braking process is used as the axis, connecting the cycle data in chronological order, suitable for analyzing degradation patterns under continuous train operation. When arranged in order of braking frequency, the cumulative number of braking operations is used as the axis, arranging the cycle data of the first, second, and so on up to the nth braking operation, suitable for focusing on the impact of braking frequency on degradation. During the arrangement process, it is necessary to ensure the continuity of the trajectory, that is, the steady-state pressure offset of the previous cycle matches the initial degradation state of the next cycle. For example, if the steady-state offset of the 1,000th braking is 8 kPa, the degradation of the 1,001st braking should continue to evolve based on 8 kPa to avoid physical meaningless pressure abrupt changes. The final complete degradation trajectory data can clearly show the entire process of the braking system gradually degrading from a normal state with an offset close to 0 to minor and severe faults. For example, the first 10,000 brakings in the trajectory represent minor pressure underfill with an offset of 0 to 15 kPa, the 10,011 to 30,000 brakings represent moderate underfill with an offset of 15 to 30 kPa, and after 30,000 brakings represent severe underfill with an offset of 30 to 50 kPa. Moreover, the pressure curve of each cycle can reflect the fault characteristics and random fluctuations of the corresponding stage, providing comprehensive and high-fidelity batch data support for subsequent fault diagnosis algorithm training and life prediction model verification.

[0083] The advantage of the entire batch generation process lies in ensuring the naturalness and authenticity of single-cycle data through step S440, and achieving the integrity and coherence of the entire degradation cycle through step S450. The batch generation not only produces isolated pressure data points, but also complete datasets containing fault evolution patterns, random fluctuation characteristics, and stage change features. For example, for the EP2002 braking system, 100 sets of degradation trajectory data containing three fault modes—underpressure, stable stage step changes, and pressure oscillation—can be generated in batches. Each set contains cycle data from 50,000 braking cycles, satisfying the data-driven technology's demand for large amounts of data while accurately reproducing degradation scenarios under different operating conditions, significantly enhancing the engineering application value of the data.

[0084] This invention addresses the need for generating performance degradation data for train air braking systems. It constructs a complete technical system from basic characteristic mining to batch data output, forming a closed-loop logic encompassing operating condition and fault feature extraction, accurate modeling of multiple degradation types, model optimization and calibration, and batch data generation. Using a pre-defined braking system model as a carrier, this technical system deeply deconstructs the three natural time-series stages of brake cylinder pressure changes: pressure rise, maximum pressure, and pressure stabilization. The system extracts various typical fault modes, organically integrating pressure stage characteristics with fault evolution laws, providing a realistic physical basis for subsequent modeling. By extending the Wiener process to build a unified modeling framework, it cleverly adapts to different degradation types such as linear, nonlinear, and stepped degradation, clearly defining the power-law constraints for nonlinear degradation and the stage division criteria for stepped degradation. This effectively solves the pain points of existing technologies, such as fragmented modeling frameworks, ambiguous parameter definitions, and poor scenario adaptability.

[0085] Regarding the improvement of data fidelity, this invention overcomes the limitations of traditional simulations that simply superimpose noise. By synergistically constraining the diffusion coefficient and normally distributed random noise, it accurately simulates the control precision error and random interference of the braking system. Simultaneously, it combines actual braking process data to calibrate model parameters, ensuring that the generated data not only conforms to the natural degradation patterns of mechanical components but also reproduces the pressure response characteristics under different operating conditions, successfully covering the degradation characteristics across all scenarios from normal states to extreme failures. Compared to traditional field data acquisition methods, this invention eliminates the need for long-term train tracking, significantly shortening the data acquisition cycle and reducing costs while enabling flexible generation of extreme operating condition data. Compared to traditional simulation methods, the generated data, due to its deep coupling of pressure stage characteristics and failure modes, exhibits significantly improved fidelity and engineering practicality, effectively compensating for the shortcomings of existing technologies.

[0086] The technical value of this invention lies not only in providing highly reliable and high-fidelity basic data support for data-driven fault diagnosis and life prediction technologies, but also in providing a scientific basis for the design optimization, operation and maintenance decisions, and fault early warning of rail transit braking systems through standardized and replicable modeling and generation processes. Its application can help improve the reliability and safety of train braking system operation, promote the development of rail transit braking technology towards precision and intelligence, provide technical support for the safe and efficient operation of the rail transit industry, and possess broad engineering application prospects and significant technological innovation significance.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of generating performance degradation data for a train air brake system, the method comprising: The application relates to a method for generating train air brake system performance degradation data. The method comprises the following steps: Based on a preset model brake system, three stages of pressure rise, maximum pressure and pressure stabilization of brake cylinder pressure change are analyzed, and multiple typical fault modes are refined; Based on an extended random process model with non-monotonicity and independent increment, the linear, nonlinear or step-type performance degradation process of the train air brake system is modeled by combining the three stages of brake cylinder pressure change and the multiple typical fault modes; Normal distribution random noise is superimposed on the performance degradation trajectory generated by modeling to simulate the control accuracy error of the actual brake system, and the train air brake system performance degradation process model is optimized; 2. The method of claim 1, wherein, The train air brake system performance degradation data is batch generated by using the train air brake system performance degradation process model.

3. The method of claim 1, wherein, The typical fault modes include at least three of pressure undercharge, pressure overcharge, too fast inflation, step change in the stable stage and pressure oscillation in the stable stage. The extended random process model is an extended Wiener process; wherein, is a drift coefficient, is a diffusion coefficient, is a diffusion function.

4. The method of claim 3, wherein, The modeling formula of the linear performance degradation process is The nonlinear performance degradation process is realized by combining a time scale conversion function on the basis of the linear performance degradation process; wherein, t is time, b is a power exponent, when b > 0 and b < 1 is convex degeneration, when b > 1 is concave degeneration.

5. The method of claim 3, wherein, The time scale conversion function is The modeling mode of the step-type performance degradation process is The degradation process is divided into stages based on the pressure rise stage, maximum pressure stage and pressure stabilization stage of brake cylinder pressure change and the typical fault modes corresponding to each pressure stage; Linear, convex or concave performance degradation trajectories are used for modeling in the same stage; 6. The method of claim 3, wherein, Parameter switching is realized by adjusting the drift coefficient and the diffusion coefficient in different stages, and then the trajectories in each stage are spliced to form a composite degradation trajectory. The mean of the normal distribution random noise is a preset mean, and the variance is a preset variance; 7. The method of claim 4, wherein, the concave degradation corresponds to a power exponent greater than 1 and close to 1; the convex degeneration corresponds to a power exponent is less than 1 and close to 1.

8. The method of claim 3, wherein, the drift coefficient The maximum and minimum values of the drift coefficient are determined by When the trajectory is generated, the diffusion coefficient is controlled within a preset range, so that the control accuracy of the brake cylinder pressure meets a preset pressure accuracy value. In combination with the recognition ability of the degradation detection model, it is ensured that The corresponding degradation rate can cover the whole-period degradation process from normal to failure and can be effectively captured by the degradation detection model.

9. The method of claim 1, wherein, The brake cylinder steady-state pressure offset is used as a constraint basis, and when the offset exceeds a first preset offset, a fault state is determined, and the maximum offset does not exceed a second preset offset; The generation mode of the performance degradation trajectory and the degradation pressure value of the train air brake system is: The degradation degree of the train air brake system is represented by the offset of the brake cylinder steady-state pressure value and the standard pressure value, and the greater the absolute value of the offset, the higher the degradation degree, wherein the standard pressure value is the rated steady-state pressure value of the preset model brake system; The degradation trajectory of the train air brake system is composed of offset data of corresponding steady-state pressure values in multiple air brake processes arranged in time or brake number order; 10. The method of claim 1, wherein, The degradation pressure value is obtained by adding the generated offset data and the standard pressure value. The batch generation of train air brake system performance degradation data comprises: Based on the modeling method and parameter setting of the extended random process model, actual brake process data is combined, and the normal distribution random noise is superimposed to generate brake cylinder pressure data conforming to the natural degradation characteristics. The change curve of the brake cylinder steady pressure value of the single air braking process is single cycle data, all the single cycle data are arranged in time or braking times order, and constitute complete degradation track data.

Images