A Train Brake Service Status Assessment System and Method Based on AI and Multi-Source Information

Abstract

This invention provides a system and method for evaluating the service status of train brakes based on AI and multi-source information, relating to the field of train braking. The system includes: a working condition simulation unit, a data acquisition unit, a data processing unit, a safety status assessment unit, and a result display and early warning unit. The method includes: S1, constructing an environment to support the operation and testing of the target brake, and providing a testing environment for the target brake; S2, acquiring multi-source physical signals during the operation of the target brake; S3, obtaining the correlation results between the multi-source physical signals, and generating corresponding primary fusion feature vectors based on the correlation results; S4, constructing a multi-working-condition mapping model library, and outputting safety status assessment results. This avoids the shortcomings of existing technologies that only judge single faults or local performance indicators, providing train brakes with system-level safety assessment and early warning capabilities covering multiple failure modes.

Description

Technical Field

[0001] This invention relates to the field of train braking, and in particular to a system and method for evaluating the service status of train brakes based on AI and multi-source information. Background Technology

[0002] The train braking system is the core line of defense for ensuring safe train operation, and its performance directly determines the train's maximum operating speed and stopping safety. As the core actuator of the braking system, the train brake typically consists of a brake disc, brake pads, brake calipers, and brake cylinders. The brake disc is fixed to the wheelset or axle, and the brake pads are mounted on the brake calipers. During friction braking, the train's braking control system converts braking commands into pressure signals, driving the brake cylinders to actuate. This causes the brake calipers to generate clamping force, forcing the brake pads to press against the brake disc, generating tangential friction at the friction pair interface. This converts the train's kinetic energy into heat energy, achieving deceleration and stopping. With the continuous increase in train speed and train weight, the thermal and mechanical loads on the braking system rise sharply, causing the brake's service condition to exhibit complex characteristics of multi-physics coupling and continuous performance evolution, which are difficult to directly observe and dynamically track. Therefore, there is an urgent need to develop intelligent state perception and assessment methods for complex operating conditions.

[0003] During actual train operation, due to long-term periodic braking and high-load service, brakes inevitably experience a series of specific physical and mechanical damage problems. The root cause lies in the fact that the train braking process is essentially a nonlinear dynamic process involving strong coupling of multiple physical fields, including thermodynamics, mechanics, vibration, and acoustics. Specifically, the friction interface generates a large amount of frictional heat and complex stresses in a very short time, leading to continuous microscopic damage within the material and a cumulative effect, ultimately evolving into large-area chipping of the brake pads, thermal fatigue cracks in the brake disc, and abnormal wear of the friction pair. Simultaneously, the stick-slip effect generated at the friction interface under specific operating conditions can excite low-frequency vibrations and structural noise in the braking system, significantly reducing ride smoothness and component lifespan. Even more serious is the fact that when trains traverse harsh climatic environments such as high temperatures, snow, and sandstorms, the intrusion of environmental media alters the original heat dissipation conditions and tribological properties of the friction interface, further accelerating various failure modes such as thermal fatigue and material degradation, severely threatening the service reliability of the train braking system.

[0004] Given the multi-dimensional service damage mechanisms and their exacerbated evolution under harsh environments, sudden failures such as disc cracking and brake pad chipping can lead to insufficient braking force, extended braking distance, and even train accidents. Therefore, early detection and prevention of service status are crucial. However, due to the continuous evolution of brake performance during long-term service, and the drastically different response characteristics of different failure modes under various physical fields and operating conditions, their operational status is difficult to accurately characterize using static data from a single time period. On the one hand, brake performance evolves continuously with braking cycles, and existing methods lack the ability to continuously monitor the service process dynamically. On the other hand, different failure modes exhibit different representations in thermal, mechanical, and vibrational physical fields, making effective fusion of multi-source information difficult. Furthermore, under the coupling effect of various environments and braking conditions, monitoring data exhibits strong nonlinearity and time-varying characteristics, and existing assessment methods lack generalization and adaptability under complex conditions. Therefore, achieving accurate perception and systematic assessment of the service status of train brakes under complex conditions is an urgent need for predictive maintenance and ensuring absolute train safety.

[0005] Currently, existing service status assessment systems have significant limitations in terms of both hardware testing platforms and software analysis methods, specifically in the following three aspects, in response to the aforementioned complex service characteristics and assessment requirements: Firstly, in terms of hardware support and testing platforms, existing test benches mostly focus on testing braking performance indicators or collecting single physical quantities under single or a few typical operating conditions, and have limited ability to simulate multiple environmental and braking conditions. For example, patent CN102749205A only simulates the driving environment through an air convection device, and although patent CN103033373A adds a water spraying device to simulate humid conditions, it still does not cover the simulation capabilities of complex environments such as high temperature, low temperature, ice and snow, and sandstorms. The data utilization method is still mainly based on result comparison or experience judgment. There is a lack of close coupling between test data and service safety status assessment, making it difficult to support systematic risk assessment for multiple failure modes.

[0006] Secondly, in terms of algorithmic diagnosis and information fusion, existing methods tend to focus on single physical dimension detection. For example, patent CN115163708A is still limited to the detection of a single physical quantity, making it difficult to cover the multidimensional characterization information of different failure modes in multi-physics fields. While patent CN116304556A involves multi-source information, it often uses simple splicing or fixed-weight fusion, failing to achieve dynamic adaptive allocation of feature weights based on braking conditions, environmental factors, and component degradation stages, thus making it difficult to fully leverage the complementary advantages of multi-source information. Furthermore, failure modes such as thermal cracking and thermal decay are often concurrent, but patents CN120720353A and CN120739818A focus on single or a few failure modes, lacking the ability to conduct in-depth correlation analysis and early warning of the evolutionary behavior of multiple failure modes.

[0007] Thirdly, regarding model adaptability and continuous evolution, due to the limitations of the aforementioned test benches in simulation, existing methods such as patents CN116304556A and CN115163708A are mostly designed based on single or a few operating conditions. The monitoring models or thresholds established by these methods show a significant decrease in generalization ability and lack adaptability when facing extreme environments such as high temperature, low temperature, ice and snow, and sandstorms. At the same time, most methods, such as patents CN120430215A and CN116304556A, rely on offline training with a large amount of historical data. Once the model parameters are determined, it is difficult to continuously update them during service. It is also difficult to use new operating condition data to continuously correct and optimize the evaluation model, thus limiting the adaptability and engineering application value of the model under long-term service conditions. Summary of the Invention

[0008] In view of this, the purpose of this invention is to develop a train brake service status assessment system that can be deeply integrated with a brake comprehensive performance test bench, make full use of multi-source heterogeneous monitoring data, realize the dynamic fusion of multi-physics field information and adaptive allocation of feature weights, and comprehensively assess the service safety risks of multiple failure modes. It also has online learning and model evolution capabilities to meet the actual needs of intelligent operation and maintenance and safety assurance of rail transit equipment.

[0009] In a first aspect, embodiments of the present invention provide a train brake service status assessment system based on AI and multi-source information, comprising: The operating condition simulation unit is used to carry out the operation test of the target brake and provide a test environment for the target brake. The data acquisition unit is used to acquire multi-source physical signals during the operation of the target brake; The data processing unit is used to obtain the correlation results and primary fusion feature vectors between the multi-source physical signals, wherein the primary fusion feature vectors are obtained by normalizing the multi-source physical signals. A safety status assessment unit is used to construct a multi-condition mapping model library, which includes multiple types of condition mapping models. Based on the acquired braking condition parameters and environmental condition parameters, the corresponding condition mapping model is selected. An advanced fusion feature vector is generated by adaptively weighting and reconstructing the primary fusion feature vector. Based on this advanced fusion feature vector, a main deep neural network and a multi-failure mode expert classifier are used to output a safety status assessment result. The safety status assessment results include: a comprehensive safety status index and a warning confidence level for at least one failure mode; The results display and early warning unit is used to visualize the security status assessment results and execute graded risk warnings according to preset strategies.

[0010] The embodiments of the present invention bring the following beneficial effects: 1. In terms of hardware support and testing platform, this invention integrates multiple environmental simulation functions such as high temperature, low temperature, ice and snow, and sandstorm through the working condition simulation unit, breaking through the limitation of existing test benches that can only simulate a single or a few typical working conditions, and providing hardware support for the service status of train brakes under the coupled effects of multiple environmental working conditions and braking working conditions; at the same time, through the synchronous acquisition and systematic utilization of multi-source physical signals, the test data and service safety status assessment are closely coupled, which strongly supports risk assessment for multiple failure modes.

[0011] 2. In terms of algorithm diagnosis and information fusion, this invention uses an attention weighting mechanism based on operating conditions to dynamically and adaptively allocate the primary fusion feature vector of multi-source physical signals according to real-time braking operating conditions and environmental operating conditions. This solves the shortcomings of existing methods that use simple splicing or fixed weight fusion, and fully leverages the complementary advantages of multi-source information. Furthermore, it introduces a parallel decision structure of a main deep neural network and a multi-failure mode expert classifier, which simultaneously covers multiple concurrent failure modes such as thermal anomalies, wear, flutter, and spalling, achieving a leap from local single fault judgment to parallel early warning of multiple failure modes.

[0012] 3. Regarding model adaptability and continuous evolution, the multi-condition mapping model library constructed in this invention contains various types of condition mapping models. It can adaptively select the corresponding evaluation model according to different environmental conditions and braking conditions, overcoming the shortcomings of existing methods that have a significant decrease in generalization ability under extreme environments. In addition, the system has a pending case library and an incremental learning mechanism, which can use newly added condition data to continuously correct and optimize the evaluation model online, breaking through the limitations of traditional model parameters being fixed and difficult to continuously update with the service process.

[0013] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0014] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0015] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the structure of a train brake service status assessment system based on AI and multi-source information provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the operation interface of the overall control module of the present invention; Figure 3 This invention relates to the distribution of the thermocouple array on the brake disc; Figure 4 The positions of the triaxial acceleration sensor and the strain gauge pressure sensor of this invention are shown. Figure 5 This is a flowchart of the data acquisition and signal processing of the present invention; Figure 6 This is a schematic diagram of the safety status assessment process of the present invention; Figure 7 This is a schematic diagram of the result display and early warning unit of the present invention; Figure 8 This is a schematic diagram of the environmental simulation module structure of the present invention; Figure 9 This is the curve showing the change in brake friction coefficient and brake disc temperature under low-temperature conditions according to the present invention. Figure 10 This invention presents the curves showing the relationship between the brake friction coefficient and brake disc temperature under icy and snowy conditions. Figure 11 This is a schematic diagram of the structure of the verification unit of the present invention; Figure 12 This is a schematic diagram of the deep neural network structure of the present invention; Figure 13 This is a schematic diagram of the rolling contact between the track wheel and the wheel of the present invention; The corresponding explanations for the labels in the attached diagram are as follows: 1 – Test bench base; 2 – LED lighting source; 3 – Capacitive microphone; 4 – Industrial camera; 5 – Thermocouple array; 6 – Torque sensor; 7 – Speed ​​sensor; 8 – Wheel-rail simulation module; 9 – Environmental simulation module; 10 – Environmental isolation chamber; 11 – Axle load simulation module; 12 – Strain gauge pressure sensor; 13 – Triaxial acceleration sensor; 14 – Target brake; 15 – Rigid gantry frame; 16 – Main control module; 17 – High and low temperature integrated unit; 18 – Environmental simulation 19 – Fan; 20 – Medium conditioning and mixing unit; 21 – First air inlet; 22 – Return outlet; 23 – Dust control valve; 24 – Dust dispensing device; 25 – Return regulating unit; 26 – Circulating fan; 27 – Air outlet; 28 – First exhaust circuit; 29 – Auxiliary aerodynamic device; 30 – Particle concentration detection device; 31 – Particle stratification filtration and collection device; 32 – Laser particle detection device; 33 – Air inlet; 34 – Second exhaust circuit. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] First, the working condition simulation unit provided in this application will be described: Reference Figure 1 and Figure 13 In the embodiments provided in this application, the working condition simulation unit includes: The test bench consists of a base 1, a rigid gantry frame 15, an environmental isolation chamber 10, an environmental simulation module 9, a wheel-rail simulation module 8, an axle load simulation module 11, a target brake 14, and a main control module 16.

[0019] The test bench base 1 is made of high-rigidity cast steel structure, which is used to support the entire test device and reduce ground vibration interference.

[0020] The rigid gantry frame 15 spans the top of the test bench base 1 and is used to provide installation support for the test execution of the target brake 14, ensuring the stable application of the braking load; Specifically, the rigid gantry frame 15 has undergone strength and stiffness optimization to ensure sufficient structural stability under maximum braking load, providing a stable mounting platform for brake calipers, sensors and cameras.

[0021] The environmental isolation chamber 10 is located outside the target brake 14, and its bottom end is sealed to the top of the test bench base 1 to form a closed environmental space for testing the target brake 14.

[0022] Specifically, the environmental isolation chamber 10 is set on the test bench base 1 and covers the target brake 14 to isolate the braking area from the external environment, so that the brake disc and brake pads are in a controlled airflow environment, thereby avoiding interference from external air disturbances, dust or humidity changes on the test results.

[0023] The environmental simulation module 9 is connected to the environmental isolation cavity 10 and is used to construct a controllable circulating airflow environment in the environmental isolation cavity 10.

[0024] Specifically, the environmental simulation module 9 is connected to the environmental isolation chamber 10 through the main airflow channel. It includes a high and low temperature integrated unit 17, an environmental simulation fan 18, a medium conditioning and mixing unit 19, and a reflux adjustment unit 25. It is used to construct an adjustable, controllable, and closed-loop stable circulating airflow environment in terms of temperature, flow rate, humidity, and medium composition. The conditioned airflow is delivered to the braking area in the environmental isolation chamber 10, and the discharged airflow is guided back to the front end through the reflux adjustment unit 25 to participate in the next cycle.

[0025] In the normal braking condition of this embodiment, the high and low temperature integrated machine 17 operates in normal temperature mode. The environmental simulation fan 18 drives the basic airflow to enter the environmental isolation chamber 10 after being shaped by the medium conditioning and mixing unit 19. A stable and uniform basic airflow environment is formed around the brake disc and brake pads to simulate the air convection state under normal vehicle driving conditions.

[0026] One end of the wheel-rail simulation module 8 passes through the environmental isolation cavity 10 and forms a rolling contact connection with the wheel of the axle load simulation module 11, such as... Figure 13 Its bottom end is set on the test bench base 1 to simulate the wheel-rail contact relationship, i.e. the wheelset rotation state, when the train is running. Specifically, the wheel-rail simulation module 8 consists of an AC servo motor, a reducer, and a track wheel. The servo motor drives the track wheel to rotate, and the track wheel forms rolling contact with the wheel in the axle load simulation module 11 to simulate the kinetic energy and rotational inertia during train operation. The servo motor is controlled by a frequency converter, which can accurately simulate the train's running speed and achieve constant torque or constant power control to simulate the train's operating conditions on different gradients.

[0027] The axle load simulation module 11 is mounted on the columns on both sides of the rigid gantry frame 15. It is used to apply a vertical load to the target brake 14 through hydraulic pressure to simulate the effect of different vehicle axle loads on the brake, and at the same time realize wheel-rail coupling through the rolling contact relationship between the wheel and the track wheel.

[0028] Specifically, the axle load simulation module 11 uses hydraulic cylinders to load and adjust the vertical load under different axle loads.

[0029] The target brake 14 includes a train brake disc, brake caliper, brake pad assembly, etc. All components are installed in accordance with the installation torque and process specified in railway standards to ensure that the test conditions are consistent with the actual service conditions. The main control module 16 is connected to the servo drive and hydraulic controller via a data bus to realize the synchronous issuance of all action commands and status monitoring.

[0030] Specifically, the main control module 16 is the central control unit of the entire system. It connects to the servo drive and hydraulic controller via a data bus to achieve synchronous issuance of all action commands and status monitoring. Its user interface is as follows: Figure 2 As shown.

[0031] In this embodiment, the operator sets parameters in the main control module 16, including but not limited to initial braking speed, braking pressure, axle load, and wheel-rail excitation.

[0032] In this embodiment, when the operator clicks the "System Self-Check" button, the main control module 16 will automatically check the connection and status of all sensors and actuators.

[0033] In this embodiment, the "synchronization status indicator" above the operation interface turns green, indicating that synchronization is normal and that all data acquisition channels have completed high-precision synchronization based on the time synchronization protocol.

[0034] In this embodiment, after the operator confirms that the self-test has passed, he clicks the "Start Test" button on the main control module 16. The wheel-rail simulation module 8 starts to work. The servo motor drives the main shaft and rail wheel to accelerate through the reducer according to the set instructions until the speed stabilizes at the target value. The axle load simulation module 11 operates synchronously, and the hydraulic servo system applies a constant vertical load to the target brake 14 according to the settings. In this embodiment, when the operator clicks the "Start Braking" button on the main control module 16, the system is automatically triggered. Under the command of the main control module 16, the brake caliper of the target brake 14 pushes the brake pads to clamp the brake disc with a set pressure, and the braking process begins.

[0035] like Figure 8 As shown, in this embodiment, the environmental simulation module 9 consists of a high and low temperature integrated unit 17, an environmental simulation fan 18, a medium conditioning and mixing unit 19, and a reflux adjustment unit 25, which constructs a stable and controllable low temperature test environment through a closed-loop airflow circuit.

[0036] The beneficial effects of the working condition simulation unit in this embodiment are as follows: With high rigidity and high stability bearing capacity, the test bench base and rigid gantry frame are made of high-rigidity cast steel and optimized for strength and stiffness. They can maintain structural stability under maximum braking load, effectively suppress ground vibration interference, provide a stable mounting reference for brakes, sensors and industrial cameras, ensure accurate and stable application of braking load, and greatly improve the repeatability, consistency and measurement accuracy of test data.

[0037] By using a fully enclosed isolation environment, the environmental isolation chamber completely seals and covers the braking area, isolating it from external air disturbances, dust, humidity fluctuations, and impurities. This ensures that the brake disc and brake pads are always in a pure, controllable, and stable testing environment, thereby avoiding test deviations caused by external factors and ensuring the authenticity and reliability of multi-physics field signal acquisition.

[0038] Through full-condition closed-loop environment simulation, the extreme service scenario environment simulation module, through the high and low temperature integrated unit, fan, medium conditioning and reflux regulation, forms a closed-loop airflow system, which can accurately reproduce complex service environments such as normal temperature, low temperature (-40℃), high temperature, wind and sand, ice and snow, and high humidity. The temperature control is stable and there is no thermal shock, which meets the performance testing and status assessment needs of trains in all regions, all seasons and extreme conditions, and significantly expands the applicability of the system.

[0039] By coupling real wheel-rail and axle loads, the wheel-rail simulation module, which highly replicates the actual vehicle operating conditions, accurately simulates wheel-rail contact, train kinetic energy, moment of inertia, and running speed, enabling constant torque / constant power control and adapting to different gradient conditions. The axle load simulation module precisely applies vertical loads hydraulically to replicate the axle load effects of different vehicles, ensuring that the test loads, motion states, and coupling relationships are highly consistent with the actual vehicle service conditions, and the evaluation results have strong engineering reference value.

[0040] Through centralized and collaborative control of the entire process, the high-automation and safety of the overall control module realizes unified scheduling, synchronized commands, and real-time monitoring of servo drive, hydraulic loading, environmental simulation, and data acquisition. It supports one-click parameter setting, automatic system self-testing, and high-precision time synchronization of channels, reducing human operation errors and improving the system's automation level, testing efficiency, and operational safety.

[0041] With comprehensive protection against low temperatures and extreme environments, it ensures long-term stability and reliability. For harsh working conditions such as low temperatures, a full set of antifreeze, anti-fog, waterproof, and sealing protection measures are taken for sensors, cameras, cables, and hydraulic oil to eliminate problems such as low-temperature drift, signal short circuits, and oil solidification, ensuring that sensors and actuators can work stably for a long time in extreme environments and that measurements are not distorted.

[0042] Through standardized installation and testing procedures, the results are traceable and comparable. The target brake is installed in strict accordance with railway standards, the testing process is standardized and controllable, and all operating parameters can be set, reproduced, and recorded, ensuring that the test results are standardized, traceable, and comparable across different scenarios, meeting the requirements of multiple scenarios such as product development, type testing, and operation and maintenance testing.

[0043] Based on the foregoing embodiments, this application provides a system and method for evaluating the service status of train brakes based on AI and multi-source information.

[0044] To facilitate understanding of this embodiment, the train brake service status evaluation system based on AI and multi-source information disclosed in this embodiment will first be described in detail, such as... Figure 1 As shown, the system includes: The operating condition simulation unit is used to carry out the operation test of the target brake and provide a test environment for the target brake.

[0045] For a detailed description of the operating condition simulation unit, please refer to the aforementioned embodiments, and it will not be repeated here.

[0046] The data acquisition unit is used to acquire multi-source physical signals during the operation of the target brake.

[0047] The data acquisition unit is used to simultaneously acquire multi-source physical signals during braking, including thermal, vibration, acoustic, mechanical, motion, and visual signals, to characterize the working state of the brake itself, as well as the vibration and acoustic signals of the test bench base components, and to monitor the overall state of the test bench.

[0048] The data processing unit is used to obtain the correlation results and primary fusion feature vectors between the multi-source physical signals, wherein the primary fusion feature vectors are obtained by normalizing the multi-source physical signals.

[0049] The correlation results between the multi-source physical signals include the coherence coefficient, cross-correlation delay, and transmission path gain index; this set of indexes is used to characterize the transmission path and attenuation characteristics of vibration energy, providing a basis for the preliminary identification of fault locations.

[0050] The safety status assessment unit is used to construct a multi-condition mapping model library, which includes multiple types of condition mapping models.

[0051] Based on the acquired braking condition parameters and environmental condition parameters, the corresponding condition mapping model is selected. The corresponding advanced fusion feature vector is generated by adaptively weighting and reconstructing the primary fusion feature vector. Based on the advanced fusion feature vector, the safety status assessment result is output through a main deep neural network and a multi-failure mode expert classifier.

[0052] The safety status assessment results include: a comprehensive safety status index and a warning confidence level for at least one failure mode.

[0053] The above system achieves a high degree of system architecture integration and a closed-loop system for the entire process of working condition-acquisition-processing-evaluation, organically integrating the four major units of working condition simulation, multi-source acquisition, data processing, and AI safety evaluation. This forms an integrated system from test environment construction to intelligent state discrimination, solving the problems of data fragmentation, decentralized evaluation, and disjointed processes in traditional braking tests, and significantly improving evaluation efficiency and engineering practicality.

[0054] By acquiring multi-source physical signals across the entire domain, the state characterization is more comprehensive and without blind spots. It simultaneously acquires six types of signals: thermal, vibration, acoustic, mechanical, motion, and visual, fully covering the strong coupling characteristics of multi-physical fields during the braking process. At the same time, it also takes into account the state monitoring of the brake body and the test bench, breaking through the limitations of incomplete evaluation information and easy misjudgment caused by single signals or a few signals.

[0055] Through in-depth analysis of multi-source signal correlation, it has the ability to initially identify and trace faults. By using indicators such as coherence coefficient, cross-correlation delay, and transmission path gain, it can quantitatively characterize the vibration propagation path and attenuation law, realize the location of abnormal sources and the initial identification of fault locations, provide prior information for subsequent accurate diagnosis, and improve the accuracy and directionality of fault identification.

[0056] By standardizing the primary fusion feature vector, the quality of model input is improved. Multi-source heterogeneous signals are normalized to form a primary fusion feature vector with unified dimension, no offset, and no magnitude interference. This eliminates data deviations caused by different physical quantities, different ranges, and different acquisition units, providing stable and high-quality input for AI evaluation.

[0057] By using adaptive feature weighting based on operating conditions, the system intelligently focuses on key state information. Based on the braking condition and environmental condition, feature weights are dynamically allocated, automatically strengthening features sensitive to the current state and weakening noise and redundant information. This achieves an upgrade from "fixed fusion" to "intelligent adaptive fusion," significantly improving the robustness and accuracy of assessments under varying operating conditions and complex environments.

[0058] Supported by a multi-condition model library, it has strong generalization capabilities across all scenarios. It has built-in multi-type condition mapping models and can automatically match the optimal evaluation model according to real-time conditions. This solves the problem that a single model cannot adapt to complex scenarios such as low temperature, high temperature, sandstorm, and snow, covering the entire service environment of trains and making it more widely applicable.

[0059] By using a main network and multiple expert classifiers for parallel decision-making, a comprehensive assessment of multiple failure modes is achieved. The main deep neural network outputs a comprehensive safety status index, which is combined with multiple specialized failure mode expert classifiers for parallel inference. This allows for the simultaneous output of overall health level and confidence levels for multiple types of fault warnings, such as thermal anomalies, wear, flutter, and spalling. This enables system-level, multi-mode, and quantitative early warning, overcoming the shortcomings of traditional methods that can only judge a single fault.

[0060] The evaluation results are quantifiable and interpretable, facilitating engineering applications and operation and maintenance decisions. The output includes a comprehensive safety status index of 0-100 and a warning confidence level of 0-1. The results are intuitive, quantifiable, comparable, and traceable, supporting tiered warnings and automatic report generation, directly supporting predictive maintenance and driving safety assurance.

[0061] As one possible design, the data acquisition unit includes: A motion signal sensing unit, wherein the motion signal sensing unit is used to acquire the motion signal of the target brake; A thermal signal sensing unit is used to acquire the thermal signal of the target brake. A vibration signal sensing unit is used to acquire the vibration signal of the target brake. A mechanical signal sensing unit is used to acquire the mechanical signal of the target brake; A visual signal sensing unit, wherein the visual signal sensing unit is used to acquire the visual signal of the target actuator; An acoustic signal sensing unit is used to acquire the acoustic signal of the target brake.

[0062] The motion signal sensing unit includes a torque sensor 6 and a speed sensor 7. The torque sensor 6 is located at one end of the wheel-rail simulation module 8, and the speed sensor 7 is located at the other end of the wheel-rail simulation module 8. The torque sensor 6 and the speed sensor 7 are used to acquire the motion signal of the target brake.

[0063] Specifically, the motion signal is acquired by a torque and speed acquisition device connected to the main shaft and is used to characterize the energy change during the braking process; The thermocouple array 5 is arranged radially along the brake disc and is used to obtain thermal signals by acquiring the temperature distribution and changes on the surface of the brake disc, such as... Figure 3 As shown; The triaxial acceleration sensor 13 is arranged on the outside of the brake caliper jaws and is used to obtain vibration signals by capturing frictional vibration and impact response. Acceleration data is obtained from an acceleration acquisition device installed on the brake caliper and key structure of the test bench, and is used to capture frictional vibration and impact response.

[0064] The strain gauge pressure sensor 12 is arranged in the hydraulic channel of the brake caliper or in a related force transmission part, and is used to obtain mechanical signals by acquiring and monitoring the braking force and its dynamic changes.

[0065] Specifically, the mechanical signal is provided by a pressure acquisition device located in the hydraulic channel of the brake caliper or related force transmission parts, used to monitor the braking force and its dynamic changes. Its location is similar to that of the three-dimensional acceleration sensor on the brake caliper. Figure 4 As shown; The industrial camera 4 is positioned vertically on the brake disc body and aligned with the friction surface of the brake disc to monitor changes in surface morphology and obtain visual signals. Specifically, the visual signal is obtained by an image acquisition device aligned with the friction surface of the brake disc, and is used to monitor changes in surface morphology.

[0066] The capacitive microphone 3 is positioned near the friction interface to acquire acoustic signals.

[0067] In this embodiment, the sensors are arranged in the most sensitive and representative positions of the braking system, including torque / speed sensors, thermocouple arrays, triaxial acceleration sensors, strain gauge pressure sensors, industrial cameras, and capacitive microphones.

[0068] Among them, the torque / speed sensor directly reflects the change in braking energy; Thermocouple arrays are arranged radially along the brake disc to fully capture the temperature field and thermal gradient; The triaxial accelerometer is placed close to the brake caliper to capture frictional vibrations and impacts with high sensitivity. The strain gauge pressure sensor is installed in the hydraulic channel to accurately reflect the dynamic characteristics of the braking force. An industrial camera is vertically aligned with the friction surface to accurately monitor surface cracks, spalling, and roughness changes; The capacitive microphone is positioned close to the friction interface, effectively picking up braking acoustic features. Its layout is highly compatible with the braking failure mechanism, ensuring complete capture of key degradation features.

[0069] By coordinating among sensor groups, multi-source signals are mutually complementary and mutually verified, significantly improving the system's anti-interference capability and robustness. Thermal, vibration, mechanical, visual, acoustic, and motion signals mutually corroborate and complement each other. When one type of signal is affected by environmental interference or noise, other modal signals can be used to maintain assessment stability, greatly improving monitoring reliability under complex operating conditions. The physical meaning of the signals is clear and strongly correlated with failure modes. Various signals directly correspond to typical failure mechanisms of the brake. Thermal signals correspond to thermal fatigue, thermal cracks, and thermal decay; vibration signals correspond to flutter, loosening, impact, and spalling; mechanical signals correspond to insufficient braking force and fluctuations in the coefficient of friction; visual signals directly reflect surface damage, spalling, and cracks; acoustic signals correspond to abnormal friction and interface instability; and motion signals correspond to braking performance and energy changes.

[0070] By complementing and verifying multi-source signals, a highly recognizable and interpretable data source is provided for subsequent AI evaluation and fault diagnosis.

[0071] Meanwhile, this application ensures the spatiotemporal consistency of multi-source information through synchronous acquisition across all channels. All sensors work synchronously based on a unified timestamp, ensuring that multi-physics field signals are strictly aligned in time and space, providing high-quality standard data that can be directly used for AI modeling for signal correlation analysis, feature fusion, and operating condition matching.

[0072] Meanwhile, it has strong adaptability to different sensor types and meets the requirements for long-term stable testing under extreme conditions. It adopts mature industrial-grade components such as thermocouples, accelerometers, pressure sensors, industrial cameras, and capacitive microphones. Combined with protective designs such as low temperature, dustproof, anti-fog, and sealing, it can work stably in harsh environments such as high and low temperatures, sandstorms, ice and snow, and dust, with no measurement distortion or drift.

[0073] This application incorporates motion signal sensing units, thermal signal sensing units, vibration signal sensing units, mechanical signal sensing units, visual signal sensing units, and acoustic signal sensing units. The structural design of these sensing units facilitates expansion and maintenance. Each sensing unit is independently configured with standardized interfaces, allowing for flexible addition, removal, or replacement based on brake type, material, and structural differences, without requiring modifications to the overall system. It is compatible with different brake discs such as cast iron and carbon ceramic, as well as various brake pad materials, demonstrating strong versatility and expandability.

[0074] In this embodiment, all raw signals are transmitted to the data processing unit in real time via the data bus, and the timestamp error meets the system synchronization requirements. In this embodiment, when the main control module 16 detects that the speed of the wheel-rail simulation module 8 is stable, it will automatically light up the "Start Acquisition" button and start synchronously acquiring all sensor signals. The operator can also manually click it at this time.

[0075] In this embodiment, the operator can observe the graphical display area of ​​the interface in real time through the main control module 16, including the comprehensive safety status index S and the real-time signal curve.

[0076] In this embodiment, the operator can switch to view the real-time waveforms of signals such as vibration, temperature, and pressure as needed.

[0077] One possible design is that, in the step of obtaining the correlation results between the multi-source physical signals in the data processing unit, The correlation results among the multi-source physical signals are obtained, and the correlation results are used to characterize fault location and the health status of the operating condition simulation unit. The correlation results among the multi-source physical signals include: Vibration transmission indices between signals and transmission paths and attenuation characteristics between signals; The vibration transmission indicators include: correlation coefficient and cross-correlation delay; The transmission path and attenuation characteristics include: transmission path gain index and attenuation characteristics; The vibration transmission index is used to generate a fault location sample set by combining the safety status assessment results. Each fault location sample in the fault location sample set includes a safety status assessment result and its corresponding vibration transmission index. The transmission path and attenuation characteristics are used to quantify the attenuation characteristics of vibration between different structural components and to generate a health state sample set of the working condition simulation unit. The health state sample set includes a health state of the working condition simulation unit and its corresponding transmission path and attenuation characteristics.

[0078] Specifically, such as Figure 5 As shown, the data processing unit is used to receive multi-source raw signals and perform preprocessing, feature extraction, and standardized fusion.

[0079] The data processing unit includes filtering, noise reduction, and baseline compensation steps to improve the availability and consistency of multi-source sensor signals under different environments and operating conditions. The filtering and noise reduction process is used to filter out noise interference and retain key feature information; The baseline compensation is used to correct for low-frequency drift and slow-varying offsets caused by changes in ambient temperature.

[0080] The data processing unit needs to extract key feature parameters from signals of different modes; Extraction of thermal characteristic parameters includes, but is not limited to, maximum temperature. Radial temperature gradient Average temperature rise The extraction of vibration characteristic parameters includes, but is not limited to, frequency band energy. Three main peak frequencies kurtosis The extraction of acoustic feature parameters includes, but is not limited to, sound pressure level (SPL), sharpness (S), and loudness (N); the extraction of mechanical feature parameters includes, but is not limited to, the coefficient of friction. Friction coefficient fluctuation rate Instantaneous braking power The extraction of visual feature parameters includes, but is not limited to, crack area ratio. Peeling rate Surface roughness .

[0081] Meanwhile, feature extraction also includes spatial and temporal relationship features obtained based on multi-source signal correlation analysis; by performing coherence analysis, cross-correlation calculation and transfer function estimation on signals synchronized in the time domain from sensing nodes at different spatial locations, the coherence coefficient, cross-correlation delay and transmission path gain index between signals are extracted; this set of indexes is used to characterize the transmission path and attenuation characteristics of vibration energy, providing a basis for the preliminary identification of fault locations.

[0082] Normalization is performed using the Z-score method. All standardized feature combinations form a unified, multi-dimensional primary fusion feature vector. This provides input to the safety status assessment unit.

[0083] As one possible design, the vibration transmission index is calculated as follows: The coherence coefficient between signals is calculated using the amplitude squared coherence coefficient: ; in, , Signals and The self-power spectral density; Let be the cross-power spectral density of the two signals; The closer the value is to 1, the higher the linear correlation between the two signals at frequency f. This index is used to determine whether vibration signals at different measuring points belong to the same excitation source, providing a basis for fault tracing. The cross-correlation delay is calculated by the time difference between the arrival of signals from two measuring points to help determine the vibration propagation direction and path length; Delay is calculated using generalized cross-correlation: ; in, This is a weighting function used to improve the accuracy of experimental estimations; Cross-correlation delay estimate , which is the time delay parameter at which the cross-correlation function reaches its maximum value.

[0084] One possible design is that the transmission path and attenuation characteristics are calculated as follows: The path gain is expressed using the transfer rate function as the path gain metric, and its calculation formula is as follows: ; in, , They are measuring points and measuring points To the same reference point The frequency response function; , They are measuring points and measuring points Value and reference point Cross-power spectral density between them; Indicates frequency From the measuring point to the measuring point Vibration transmission gain; Through coherence coefficient Determine the vibration transmission path; through cross-correlation time delay. Determine the time sequence of vibration arrival at different measuring points and determine the direction of transmission; use the transmissibility function. Otherwise, it is determined that there is a decay effect.

[0085] The beneficial effects of the data processing unit in this embodiment are as follows: by performing high-quality preprocessing on the original signal, the signal availability under strong noise conditions is improved, including: through filtering, noise reduction and baseline compensation, environmental noise and electrical interference are effectively filtered out, temperature drift and low frequency offset are corrected, so that the multi-source sensor signal remains stable, clean and reliable under complex conditions such as high and low temperature, wind and sand, ice and snow, and the signal consistency and availability are significantly improved.

[0086] Key features are extracted from five dimensions: thermal, vibration, acoustic, mechanical, and visual. These features cover all aspects of state information, including temperature field, vibration energy, friction characteristics, surface damage, and acoustic quality. The features are highly correlated with failure modes such as thermal fatigue, wear, flutter, spalling, and cracks, providing a highly recognizable and interpretable feature foundation for intelligent assessment.

[0087] By using four key indicators—coherence coefficient, cross-correlation delay, transmission path gain, and attenuation characteristics—the vibration excitation source, propagation direction, transmission path, and attenuation law can be quantitatively characterized. This enables accurate identification of abnormal signal sources and pinpointing of fault locations, achieving preliminary location and tracing of brake faults and significantly improving diagnostic accuracy and directionality.

[0088] By using vibration transmission and attenuation characteristics to quantitatively assess the health status of the working condition simulation unit itself, interference factors such as test bench loosening, deformation, and abnormal vibration can be identified, avoiding the impact of test bench failure on brake evaluation results and ensuring that test data are pure, effective, and reliable.

[0089] Z-score normalization is used to map signals with different physical dimensions, ranges, and amplitudes to a standard distribution, eliminating model training bias caused by differences in dimensions and values, and forming a stable, unified, and directly input primary fusion feature vector into the AI ​​model.

[0090] By combining vibration transmission indicators with safety assessment results to generate a fault location sample set, and by combining transmission attenuation characteristics with bench status to generate a bench health sample set, high-quality labeled data is provided for model training, validation, and incremental learning, continuously improving the system's accuracy and generalization ability.

[0091] Simultaneously extracting signal time-domain, frequency-domain, and spatial-domain features, along with multi-measurement-point correlation features, fully characterizes the strongly coupled dynamic characteristics of the braking process. This is more comprehensive and refined than traditional single-feature extraction, significantly improving the completeness of state representation under complex working conditions.

[0092] The coherence coefficient, cross-correlation delay, and transfer rate function are all based on mechanical vibration and signal processing theory. The source of the features is clear and the calculation logic is traceable, so that the subsequent AI evaluation is no longer a "black box" and meets the requirements of high safety and high interpretability of rail transit equipment.

[0093] One possible design is, such as Figure 6 As shown, the safety status assessment model in the safety status assessment unit includes: Feature layer, attention weighting layer, multiphysics state representation layer and decision layer; The feature layer is used to receive the primary fused feature vector; A multi-physics state representation layer is used to transform the primary fusion feature vector into a corresponding representation vector through M set physical fields; The attention weighting layer is configured to dynamically weight and reconstruct the representation vector of each physical field based on braking condition parameters and environmental condition parameters to generate a high-level fusion feature vector. The attention weighting layer adjusts the weights of each physical field representation vector in the following manner: The trainable parameters of the attention weighting layer are initialized based on historical data to generate the initialization parameter set of the attention weighting layer. The initialization parameter set consists of the initialization weight parameters of the physical field representation vector under various braking conditions and environmental conditions. Calculate the normalized eigenvector for each of the physical field representation vectors; The weights of the standardized feature vectors are set based on the initialization parameter set; A primary fusion feature vector is formed based on all the standardized feature vectors. ; The primary fusion feature vector Hidden representations of each feature are generated using a gating network. Then, the original attention score is calculated using the trained scoring vector. ; ; ; in, The attention score vector is obtained by initializing the parameter set. The input is the primary fusion feature vector; Features The original attention score; Features The feature transformation weight matrix; Calculate features using the softmax function Final weight : ; in, The original attention parameters for the j-th feature; The real-time operating conditions include: acquiring braking operating condition parameters and environmental operating condition parameters.

[0094] If, under the current operating condition A, the original score of feature a is... Much larger than other characteristics, then The weights that dominate the denominator are calculated as follows: It will automatically approach 1. Because the softmax function is used, the sum of all weights is always 1, so when... When the weights of the other features increase, their weights will automatically decrease proportionally, the magnitude of which depends on the weights of the features. Size.

[0095] In this embodiment, the attention weighting layer is used to adaptively adjust the influence of different features in the final prediction according to the real-time operating conditions. This layer uses pre-trained parameters to automatically identify which features are more likely to reflect potential risks or performance degradation trends under the current braking conditions and assign them higher weights. For features that contribute little to the judgment or have a lot of noise, their weights are reduced. The dynamic weighting mechanism can reflect the differences in the importance of key features of the brake under different operating conditions. For example, when the disc temperature rises significantly, the system can automatically increase the weight of thermal features; when there is abnormal vibration or surface damage, it can enhance the contribution of vibration and visual features. Through the above process, a high-level fusion feature vector that highlights key state information can be obtained. .

[0096] In this embodiment, the layer does not require manual weight setting and can be adaptively optimized based on the learning results of historical data, thereby significantly improving the system's adaptability and evaluation accuracy under different operating conditions.

[0097] Based on the automatic dynamic allocation of feature weights according to braking conditions and environmental conditions, no manual parameter adjustment is required: thermal characteristics are enhanced at high temperatures, vibration weight is increased when vibration is abnormal, and mechanical / visual weight is enhanced when wear is aggravated. Noise and redundant information are automatically suppressed, significantly improving the evaluation accuracy and robustness under varying conditions and extreme environments.

[0098] The system employs a gated network, attention scoring vector, tanh activation, and softmax normalization. The weights are automatically normalized and sum to 1, avoiding bias caused by imbalance in feature magnitudes. The weighting process is smooth, stable, and reproducible.

[0099] The decision layer is used to combine the high-level fusion feature vector. It uses a main deep neural network and multiple expert classifiers for specific failure modes to perform parallel decision calculations, and outputs a comprehensive safety status index and warning confidence of various failure modes.

[0100] Based on the above safety status assessment model, the entire process of "feature input - physical field mapping - adaptive weighting of operating conditions - intelligent decision-making" can be fully realized. The structure is rigorous and traceable, breaking the "black box" limitation of traditional AI models and meeting the high safety and high reliability requirements of rail transit.

[0101] By transforming primary features into dedicated representation vectors through M physical fields, the strong coupling characteristics of thermo-mechanical-vibration-acoustic-visual processes in the braking process are accurately matched. This approach is more in line with engineering mechanisms than purely data-driven models, resulting in more accurate and stable state representations.

[0102] The decision-making level outputs a comprehensive security status index. and the confidence level of early warning for various failure modes In the steps, Based on environmental operating parameters, environmental operating conditions are classified into various types, resulting in multiple types of operating conditions. Obtain environmental data and primary fusion feature vectors for each type of operating condition. ; The comprehensive safety status index is calculated using the performance index mapping method. And based on the real labels of various failure modes Construct a training sample set ; Through the training sample set Train a mapping relationship model for each of the aforementioned working conditions. The mapping model includes: a main deep neural network and multiple expert classifiers for specific failure modes; Select a target mapping relationship model based on the current type of working condition; The advanced fusion feature vector is input into the target mapping relationship model, and the comprehensive safety status index of the target brake and the warning confidence of various failure modes are output.

[0103] Specifically, the advanced fusion feature vector The input is fed into the main deep neural network, undergoes multiple nonlinear transformations, and outputs a comprehensive safety status index S, which is used to characterize the overall health status of the brake. When the value of S is lower than a preset threshold, the system determines that the brake is in a potential risk state. When S is continuously lower than the threshold, the system enters a high-risk mode and triggers the result display and warning unit.

[0104] High-level fusion feature vectors Several lightweight expert sub-classifiers are input in parallel to identify different types of failure modes, including but not limited to: thermal anomaly, flutter, spalling, and wear; each expert sub-classifier outputs a warning confidence level for each failure mode. And together with the S-value, it constitutes the basis for system risk assessment; when the confidence level of any failure mode... When the threshold is exceeded, the system automatically triggers an alarm; In one possible technical solution, in the step of constructing and training the mapping relationship model for each type of working condition, Using a comprehensive performance testing rig for train brakes, environmental data and preliminary fusion feature vectors were acquired under each type of operating condition. ,include: Braking tests were conducted under corresponding environmental simulation conditions, and multi-source signals, including thermal, vibration, acoustic, mechanical, motion, and visual signals, were collected simultaneously. These signals were then processed by the data processing unit and the safety status assessment unit to generate a primary fusion feature vector. .

[0105] Meanwhile, the comprehensive safety status index is calculated using the performance index mapping method. And based on the real labels of various failure modes Construct a training sample set : ; During training, the initial fused feature vector of each sample is... The attention-weighted layer of the input safety status assessment unit combines current operating parameters, such as ambient temperature, humidity, and dust concentration, and dynamically generates attention weights through a gating network. Weighted fusion is performed to obtain the advanced fusion feature vector. The data are input in parallel into the main deep neural network and each expert sub-classifier, which output the predicted value of the comprehensive security status index respectively. and the predicted confidence values ​​for each failure mode .

[0106] A multi-task learning framework is employed to jointly optimize the main network, attention-weighted layers, and all expert sub-classifiers. The total loss function is defined as the weighted sum of the mean squared error loss and the binary cross-entropy loss: ; in, The loss for the main task is calculated using the mean squared error loss function: ; For the first The loss of each expert sub-classifier is calculated using the binary cross-entropy loss function: ; and The weight parameter is used to balance the contributions of the main task and multiple classification tasks, and is generally set to a value of 1. ,

[0107] During training, mini-batch gradient descent is used, with the total loss function... To optimize the objective, the gradient of the loss function with respect to the parameters of each layer is calculated using the backpropagation algorithm, and the trainable parameters of the main deep neural network and all expert sub-classifiers are updated synchronously using the Adam optimizer.

[0108] During training, the following parameters need to be updated: the gating network parameters of the attention-weighted layer, and the weight matrices of each layer of the main network. and bias Hidden layer weight matrices of each expert subclassifier and bias Output layer weight vectors of each expert subclassifier and bias The parameters are iteratively updated until the loss function converges, thus obtaining the trained main deep neural network and expert sub-classifier models.

[0109] The mapping relationship model construction and training based on each of the above-mentioned working conditions adopts a multi-working-condition dedicated model training mechanism. Dedicated mapping models are established for low temperature, high temperature, wind and sand, and ice and snow, so that the system can deeply adapt to the braking characteristics of different extreme environments, and fundamentally solve the problem of poor generalization ability and inaccurate evaluation of single models under complex working conditions.

[0110] A multi-task joint learning framework is constructed to unify and optimize comprehensive safety status assessment and multi-failure mode identification, enabling the model to accurately identify faults such as thermal anomalies, wear, flutter, and spalling while learning the overall health status, thus achieving dual capabilities of global assessment and detailed diagnosis.

[0111] In the embodiments of this application, a joint loss function is constructed using mean squared error loss and cross-entropy loss, which takes into account both regression and classification tasks, making the training objective more reasonable, the model convergence more stable, less prone to overfitting, and maintaining high accuracy even after long-term use.

[0112] The model employs mini-batch gradient descent and the Adam optimizer to collaboratively update the parameters of the attention layer, main network, and expert classifier. This approach results in high training efficiency, thorough parameter optimization, and stable output even under varying operating conditions and strong noise.

[0113] The above training process is completely consistent for the four working conditions of low temperature, high temperature, wind and sand, and ice and snow. The only difference is that the training data are collected from different environmental simulation conditions.

[0114] In one possible technical solution, the main deep neural network structure is as follows: Figure 12 As shown, it includes: A multi-layer fully connected neural network (MLP) structure is used to fuse the high-level feature vector obtained after the attention weighting layer. Gradually mapped to a comprehensive safety status index The network contains multiple hidden layers, each of which performs the following sublinear transformation: ; ; ; in, and For the first The weight matrix and bias vector of the layer are parameters obtained through pre-training; The output layer uses the sigmoid activation function to make the output value take up to the range of [0,1], and then multiplies it by 100 to obtain the comprehensive security state index with a range of 0~100.

[0115] The main deep neural network adopts a multi-layer MLP structure to gradually map high-level fusion features into a comprehensive safety state index. It has strong nonlinear expression capabilities and can fully fit the complex relationship of multi-physics field coupling in the braking process.

[0116] By combining ReLU and sigmoid activation, the output stably falls within the 0~1 range, and is then scaled to a comprehensive safety status index of 0~100. The numerical values ​​are intuitive, standardized, and convenient for engineering judgment and early warning.

[0117] After being weighted by attention, the features are input into the main network, resulting in higher information density, smaller model size, and faster inference, enabling online real-time evaluation of the braking process.

[0118] In one possible technical solution, the plurality of expert classifiers for failure modes include: Each sub-classifier uses a single-layer MLP structure, and its calculation process is as follows: ; ; in, and The first The input weight matrix and bias vector of each expert subclassifier This is the weight vector from the hidden layer to the output layer. These are the output layer bias parameters.

[0119] The aforementioned multi-failure mode expert classifier structure adopts a lightweight single-layer MLP structure, which has low computational load, fast response, and does not affect the real-time performance of the system, making it suitable for vehicle / bench online monitoring.

[0120] Each classifier is specifically designed for a particular failure mode, making it highly targeted and accurate in identifying faults. It can output independent confidence scores for precise early warning and can perform parallel inference with the main network without interfering with each other. This achieves "overall health scoring + parallel diagnosis of multiple faults," breaking through the limitations of traditional methods that can only identify a single fault.

[0121] The closed-loop environment simulation module covering extreme service scenarios, which simulates all operating conditions, consists of a high and low temperature integrated unit, a fan, and a medium conditioning and reflux regulation system to form a closed-loop airflow system, can accurately reproduce complex service environments such as normal temperature, low temperature (-40℃), high temperature, sandstorm, ice and snow, and high humidity.

[0122] The low-temperature environment simulation examples are as follows: The high and low temperature integrated unit 17 selects the cooling mode to generate a low temperature working medium. The environmental simulation fan 18 applies power to the low temperature medium, causing it to form a continuous circulation flow between the high and low temperature integrated unit 17, the medium conditioning and mixing unit 19, and the braking area. The medium conditioning and mixing unit 19 adjusts the temperature and flow rate of the low temperature airflow to form a stable and repeatable low temperature environment field. The recirculation adjustment unit 25 guides the airflow after passing through the braking area back to the high and low temperature integrated unit 17 and adjusts the recirculation ratio and flow rate to form a closed and controllable low temperature environment circulation system.

[0123] In this embodiment, the temperature control range of the low-temperature environment is set to -40℃ to room temperature. The system monitors the internal temperature in real time through a temperature sensor feedback signal and dynamically adjusts the output power and medium flow rate of the high and low temperature integrated unit 17 to ensure that the low-temperature environment changes smoothly according to a preset gradient, avoiding thermal shock or condensation of the brake disc or brake pads due to sudden temperature changes. The brake friction coefficient and brake disc temperature change curve under low-temperature conditions are shown in the figure below. Figure 9 As shown.

[0124] In this embodiment, rubber waterproof insulating sleeves are installed on all sensors exposed to low-temperature environments; low-temperature anti-fog lenses are installed in front of industrial camera lenses; cold-resistant cables are used for data acquisition cables, and low-temperature sealant is applied to the terminals to prevent water vapor condensation from causing signal short circuits; low-temperature antifreeze hydraulic oil is added to the hydraulic channel of the strain gauge pressure sensor to avoid oil solidification at low temperatures, which would cause pressure measurement distortion.

[0125] The evaluation results show that, under low-temperature environmental conditions, the system can effectively identify the key state change characteristics during braking, distinguish the evolution trend of different failure risks, and make stable and repeatable assessments of the safety status of the brake, providing a basis for braking safety analysis under complex environmental conditions.

[0126] The following are examples of high-temperature environment simulation: In this embodiment, the environment simulation module 9 is as follows: Figure 8 As shown. The high and low temperature integrated machine 17 selects the heating mode to generate a high temperature medium. The high temperature airflow is driven by the environmental simulation fan 18 to circulate on the test bench. After being regulated by the medium conditioning and mixing unit 19, it is input to the working area of ​​the brake, forming a stable high temperature environment around the brake disc and brake pads. The gas discharged from the working area is returned to the high and low temperature integrated machine 17 through the return flow regulating unit 25 to realize the closed-loop circulation of the high temperature medium.

[0127] In this embodiment, the temperature control range of the high-temperature environment can be set from room temperature to 80°C, and the temperature fluctuation is controlled within ±3°C. The temperature distribution is monitored in real time by multiple temperature sensors. The main control module 16 adjusts the output power and medium flow of the high and low temperature integrated machine 17 through temperature feedback to keep the temperature field uniform and stable, so as to simulate working conditions such as continuous braking and high temperature thermal decay.

[0128] In this embodiment, the industrial camera, vibration sensor, temperature sensor, and signal cable are equipped with high-temperature resistant and heat-insulating protective structures; high-temperature resistant hydraulic oil is used in the hydraulic system to avoid inaccurate pressure measurement caused by a decrease in oil viscosity or evaporation.

[0129] Before starting signal acquisition, the temperature of the isolation chamber is raised to the set value and stabilized for 1 hour. Then, 5 minutes of static output is collected as the reference zero value under high temperature environment to eliminate zero drift and changes in the performance of electronic components caused by high temperature.

[0130] In this embodiment, the safety status assessment unit calls upon a mapping relationship model specifically trained for high-temperature operating conditions, and comprehensively analyzes thermal attenuation characteristics, friction pair temperature rise patterns, vibration mode changes, etc., based on a multi-source information fusion mechanism, thereby completing the service safety status assessment of the brake in a high-temperature environment.

[0131] In this embodiment, the safety status assessment unit performs a comprehensive analysis of the service status of the brake under high-temperature environmental conditions based on the collected multi-source feature information, outputs the confidence level of each failure mode, and further provides a comprehensive safety status index that characterizes the overall service safety level of the brake.

[0132] The evaluation results show that, under high-temperature conditions, the system can effectively identify the key state change characteristics during braking, distinguish the evolution trend of different failure risks, and make stable and repeatable assessments of the safety status of the brake, providing a basis for braking safety analysis under complex conditions such as high temperature and thermal decay.

[0133] The following is an example of simulating a windy, sandy, and dusty environment: In this embodiment, the high and low temperature integrated unit 17 operates in normal temperature mode as the airflow medium source, outputting basic working gas; the environmental simulation fan 18 applies power to the gas, causing the airflow to form a continuous flow between the high and low temperature integrated unit 17, the medium conditioning and mixing unit 19, and the brake working area; the medium conditioning and mixing unit 19 adjusts the flow rate and uniformity of the airflow, so that the airflow entering the braking area forms a stable and repeatable wind field structure; the airflow discharged from the braking area is returned to the high and low temperature integrated unit 17 via the return flow adjustment unit 25, thereby forming a closed-loop circulating wind field environment.

[0134] The dust release device 23 is arranged in the environmental isolation chamber 10 and is located in the airflow action area near the brake disc and brake pads. The dust release rate is adjusted by the dust control valve 22 so that the particles move with the airflow near the brake friction interface. The return flow adjustment unit 25 is used to guide and divide the return gas carrying dust. A portion of the gas is returned to the wind and sand generation path for recycling after particle separation and filtration, while the other portion is discharged from the system as needed to maintain the dust concentration and flow field stability in the chamber.

[0135] In this embodiment, the lens of the industrial camera is equipped with an automatic cleaning device that uses pulsed airflow to periodically remove dust adhering to the lens surface; the microphone is equipped with a windproof noise shield and a dustproof filter inside; the thermocouple and accelerometer are sealed with a labyrinth-type sealing structure to prevent dust intrusion from affecting measurement accuracy; and all electrical interfaces are protected with IP67 rating to ensure stable operation of the system in dusty environments.

[0136] In this embodiment, the brake disc and brake pads operate under normal loading conditions, and the system's data acquisition unit synchronously records multi-dimensional data such as brake pressure, temperature rise, and friction coefficient fluctuations.

[0137] In this embodiment, the safety status assessment unit calls a mapping relationship model specifically trained for sandstorm environments, and through multi-source information fusion and attention mechanism, identifies friction torque disturbances and performance degradation trends caused by particle erosion, thereby achieving braking safety assessment under sandstorm environments.

[0138] In this embodiment, the safety status assessment unit performs a comprehensive analysis of the service status of the brake under windy and sandy environmental conditions based on the collected multi-source feature information, outputs the confidence level of each failure mode, and further provides a comprehensive safety status index that characterizes the overall service safety level of the brake.

[0139] The evaluation results show that, under aeolian and sandy environmental conditions, the system can effectively identify key feature differences caused by particle erosion and changes in friction state, distinguish the evolution trend of different failure risks, and make stable and repeatable assessments of the safety status of the brake, providing a basis for brake safety analysis in complex service environments such as aeolian and sandy conditions.

[0140] In the sandstorm environment of this embodiment, external sand particles may enter the air intake area of ​​the verification unit, but under the combined effect of the isolation chamber, layered filtration and detection process, they only exist as background disturbances and do not participate in the dominant formation of abrasive particle detection features.

[0141] The collected thermal information, acoustic features, and other multi-source information are analyzed collaboratively. The multi-source features are weighted and fused through the evaluation unit to achieve a comprehensive assessment of the service safety status of the brake in windy, sandy, and dusty environments. The confidence level of each failure mode is output, and a comprehensive safety status index characterizing the overall service safety level of the brake is given, providing a basis for brake safety analysis under complex environmental conditions.

[0142] The following is an example of simulating an ice and snow environment: The high and low temperature integrated machine 17 generates a low temperature working medium. Driven by the environmental simulation fan 18, the medium enters the medium conditioning and mixing unit 19. In this unit, a water-containing medium is introduced into the airflow through a medium injection channel, so that the low temperature airflow and liquid water or high humidity saturated gas form a uniformly distributed low temperature water-containing airflow under the action of flow field shearing and turbulence. The water-containing airflow after being shaped by the medium conditioning and mixing unit 19 is sent into the brake working area, forming an ice and snow adhesion state on the surface of the brake disc and brake pads, with alternating frost, ice and melting.

[0143] The humid and cold gas discharged from the braking area is guided back to the high and low temperature integrated machine 17 and the medium conditioning and mixing unit 19 by the return regulating unit 25 to participate in the next cycle. The water content, gas flow rate and temperature in the return gas are coordinated and regulated to achieve controllable adjustment of the ice and snow formation rate, ice layer thickness and melting behavior in the braking area, thereby constructing a repeatable and adjustable non-steady-state ice and snow working environment.

[0144] In this embodiment, the system's data acquisition unit performs multi-dimensional synchronous recording, including the impact of the ice layer breaking at the moment of rupture, the evolution of the temperature field, the change of the friction coefficient, the evolution of the ice layer thickness, and acoustic characteristics, among other multi-dimensional features.

[0145] like Figure 10 As shown in the figure, this curve represents the test results of the friction coefficient and brake disc temperature changing with braking distance during braking in icy and snowy conditions. The horizontal axis represents the drag distance, the left vertical axis represents the friction coefficient, and the right vertical axis represents the brake disc temperature. It can be seen that in icy and snowy conditions, the friction coefficient changes rapidly with the drag distance in the initial stage of braking, exhibiting obvious fluctuations within a certain distance; simultaneously, the brake disc temperature also gradually increases during the braking process. In the partial figure, the friction coefficient curve shows staged fluctuations, reflecting the dynamic changes in the friction interface state under icy and snowy conditions.

[0146] In this embodiment, the continuous generation and rupture of ice layers cause the braking process to exhibit significant non-steady-state characteristics, thereby introducing coupling changes between multiple source signals. Traditional evaluation methods based on steady-state friction parameters or single-mode signals are difficult to accurately determine the safety status of the brake. Based on this, the system performs collaborative analysis of the collected thermal information, acoustic features, and other multi-source information. Through the evaluation unit, the multi-source features are weighted and fused to achieve a comprehensive assessment of the service safety status of the brake in icy and snowy environments. The system outputs the confidence level of each failure mode and further provides a comprehensive safety status index that characterizes the overall service safety level of the brake, providing a basis for brake safety analysis under complex environmental conditions.

[0147] Based on the full-condition closed-loop environment simulation in the above embodiments, the beneficial effects under each condition are as follows: The system features comprehensive low-temperature protection through all sensors, eliminating issues such as low-temperature drift, signal short circuits, and hydraulic oil solidification, ensuring accurate acquisition of multi-source signals. The system effectively identifies fluctuations in friction coefficient, temperature rise characteristics, and vibration changes at low temperatures, providing stable and repeatable evaluation results and offering a basis for safe braking in cold weather.

[0148] The high-temperature operating condition system employs a complete set of high-temperature resistant protection designs to prevent sensor drift, hydraulic oil failure, and image distortion caused by high temperatures, ensuring reliable testing under high-temperature conditions. The system can accurately capture high-temperature thermal decay, frictional characteristic degradation, abnormal vibration, and temperature rise patterns, providing stable and reliable evaluation in high-temperature environments.

[0149] The system is designed to withstand wind and sand / dust conditions with full-component dust protection: automatic camera cleaning, labyrinth seals for sensors, and wind and dust protection for microphones, ensuring long-term stable operation. The system can identify frictional torque disturbances, friction coefficient fluctuations, and wear acceleration trends caused by particle erosion, providing accurate and reliable assessments in windy and sandy environments.

[0150] In icy and snowy conditions, it can reproduce the unsteady icy and snowy environment of alternating freezing, frost, and melting, with controllable ice thickness and formation rate, highly matching the actual icy and snowy conditions of railway lines. It simultaneously captures strong unsteady characteristics such as ice fracturing impact, friction coefficient jumps, temperature transients, and acoustic abrupt changes. Through multi-source information fusion and adaptive weighting, it solves the problem that traditional single-signal methods cannot determine the dynamic changes of the ice and snow interface, achieving accurate safety assessment under icy and snowy conditions.

[0151] As one possible design, the train brake service condition assessment system also includes: The verification unit includes an exhaust pipe, a particle stratified filtration and collection device 31, and an abrasive particle detection device. The verification unit is used to analyze the collected abrasive particle samples to obtain analysis results, and to perform post-verification on the comprehensive safety status index and the early warning confidence of various failure modes output in the decision-making layer based on the analysis results. The discharge pipe is used to connect the environmental isolation chamber 10 and the particle stratification filtration and collection device 31 to form a closed loop. The abrasive particle detection device includes a laser particle detection device 32, a particle concentration detection device 30, and an auxiliary aerodynamic device 29; The laser particle detection device 32 is used to monitor the particle size distribution of the discharged gas particles in real time; The particle concentration detection device 30 monitors the concentration of each filter layer through a dynamic detection opening; The auxiliary aerodynamic device 29 works in conjunction with a negative feedback mechanism to regulate airflow; The analysis of collected abrasive particles yielded analytical results, which were then used to conduct post-validation of the comprehensive safety status index and the warning confidence levels for various failure modes output by the decision-making level. This included: This invention proposes a train brake service safety status assessment system based on multi-source information fusion. This embodiment further illustrates the specific application of the verification unit in the system of this invention. This unit is used to collect abrasive particles generated by the brake friction pair during the braking process, and after the safety status assessment process is completed, to independently analyze and post-verify the abrasive particle characteristics, thereby verifying and calibrating the assessment results to improve the accuracy of wear-related status judgment.

[0152] The integrated verification unit on the train brake comprehensive performance test bench includes a particle stratification filtration and collection device 31, an exhaust pipe, and an abrasive particle detection device, such as... Figure 11 As shown; In this embodiment, the environmental isolation cavity 10 not only prevents interference from external factors, but also prevents abrasive particles from escaping to the surroundings and polluting the environment. The inner surface is electroplated to reduce particle adhesion. The discharge pipe includes a first discharge loop 28 and a second discharge loop 34; the first discharge loop 28 connects the air outlet of the environmental isolation chamber 10 to the particle stratification filter collection device 31; the second discharge loop 34 forms a closed-loop air circulation, driven by the circulating fan 26. The particle layered filtration and collection device 31 contains multi-stage filter screens arranged from top to bottom according to particle size from small to large. Each stage of filter screen includes a nano-layer filter layer and an electrostatic particle adsorption layer, and is equipped with a vibration cleaning device. The abrasive particle detection device integrates a laser particle detection device 32, a particle concentration detection device 30, and an auxiliary aerodynamic device 29. The laser particle detection device 32 is used to monitor the particle size distribution of the exhaust gas particles in real time, the particle concentration detection device 30 monitors the concentration of each filter layer through a dynamic detection opening, and the auxiliary aerodynamic device 29 adjusts the airflow in conjunction with a negative feedback mechanism. In this embodiment, the abrasive particles generated during braking are collected by the layered filtration device 31 according to particle size and then quantitatively analyzed. In this embodiment, the abrasive particle detection device analyzes the collected abrasive particle samples to obtain the number, particle size distribution, morphology and metal composition characteristics of the abrasive particles. The morphology, composition and number of abrasive particles can directly reflect the real degradation mechanism of the friction pair material. For example, plate-like or layered abrasive particles indicate thermal fatigue or surface peeling process, sharp-edged abrasive particles characterize the failure mode dominated by abrasive wear, high iron content abrasive particles reflect the increased exposure of the brake disc substrate, and a sudden increase in the number of abrasive particles usually corresponds to the transition stage of the wear rate.

[0153] In this embodiment, abrasive data is not involved in the multi-source signal fusion in the real-time evaluation process. The abrasive degradation index output by this unit is independently compared and verified with the comprehensive safety status index S and the confidence level of wear-related failure modes given by the safety status evaluation unit. Its core function is to perform post-verification of the wear state and its changing trend output by the evaluation model.

[0154] When the fusion evaluation results of abrasive grain characteristics and multi-source signals are highly consistent, the system will improve the reliability of the evaluation results. When there is a significant deviation between the two, such as abrasive grains showing compositional characteristics that indicate severe thermal damage while multi-source signals do not show obvious abnormalities, the system will identify this case as a potential misjudgment or deviation and trigger a calibration prompt.

[0155] The reverse authenticity verification based on the microscopic morphology and macroscopic quantitative characteristics of abrasive particles provides a key basis for model weight optimization and threshold adaptive adjustment.

[0156] In this embodiment, all compared abrasive data and verification conclusions, especially those showing deviations or indicating typical failure modes, will be stored in the system's pending case library. This data can be used to drive incremental learning and long-term updates of the evaluation model, thereby significantly improving the accuracy and robustness of the comprehensive safety status assessment in wear identification.

[0157] As can be seen from this embodiment, the verification unit effectively verifies and calibrates the multi-source information fusion evaluation results by providing direct physical evidence independent of sensor signals, thereby enhancing the reliability of the entire system in determining wear status and providing a data foundation for continuous model optimization.

[0158] Through the above embodiments, this application achieves post-verification of AI assessment results by providing independent physical evidence. By collecting abrasive particles generated by the braking friction pair through a closed-loop airflow and a graded filtration device, direct physical evidence that is completely independent of multi-source sensor signals is formed. This objectively verifies the comprehensive safety status index and failure mode confidence level output by the safety status assessment unit, significantly improving the authenticity and credibility of the assessment results.

[0159] Abrasive grain characteristics can directly invert the degradation mechanism of friction pairs, making diagnosis more based on physical evidence; Laser particle detection, concentration detection, and composition and morphology analysis can be used to obtain the number, size distribution, morphology, and metal composition characteristics of abrasive particles, directly reflecting the true degradation mechanism of the braking friction pair. Among them, flaky / layered abrasive grains indicate thermal fatigue or surface spalling; Sharp-edged abrasive grains characterize abrasive wear; High iron content in abrasive particles reflects increased exposure of the brake disc substrate; A sudden increase in the number of abrasive particles corresponds to a jump in the wear rate.

[0160] By acquiring the aforementioned abrasive properties, failure determination gains clear physical meaning, rather than being purely data-driven.

[0161] Through closed-loop collection and graded filtration, the abrasive particle recovery rate is high and pollution-free. It adopts a dual-loop closed-loop airflow design and is equipped with an environmental isolation chamber to prevent abrasive particles from escaping. Based on multi-stage filter screens and electrostatic adsorption, it achieves graded collection according to particle size. The vibration cleaning device ensures long-term stable operation and accurate and repeatable test data.

[0162] Furthermore, abrasive analysis is only used for post-verification and does not participate in real-time multi-source information fusion. This ensures high-speed, continuous, and online evaluation of the braking process, while also enabling long-term calibration and correction of wear-related conditions.

[0163] When the abrasive grain characteristics deviate from the AI ​​evaluation results, the system automatically identifies them as samples to be calibrated and includes them in the unresolved case library for model weight optimization, threshold correction, and incremental learning, enabling the system's diagnostic capabilities to continuously evolve and become more stable and accurate in the long term.

[0164] In one possible technical solution, the collected abrasive particles are analyzed to obtain analytical results, and based on these results, the confidence levels of the comprehensive safety status index and various failure modes output by the decision-making level are subsequently verified, including: The abrasive particle detection device analyzes the collected abrasive particle samples to obtain the number, particle size distribution, morphology and structure and metal composition characteristics of the abrasive particles. The morphology, composition and number of abrasive particles can directly reflect the real degradation mechanism of the friction pair material. Flaky or layered abrasive grains indicate thermal fatigue or surface spalling processes, while sharp-edged abrasive grains characterize abrasive wear-dominated failure modes. High-iron-content abrasive grains reflect increased exposure of the brake disc substrate, and a sudden increase in the number of abrasive grains usually corresponds to a transition stage in the wear rate.

[0165] like Figure 7 As shown, in one possible technical solution, the result display and early warning unit is used to display the safety status index S and the failure mode early warning confidence level. The system provides visual displays and offers intuitive, tiered risk alerts to operators based on the established warning strategies.

[0166] The system provides three states: S≥85 indicates a healthy state, with a green display on the interface, indicating that the brake disc is performing well; 60≤S<85 indicates a state of concern, with an orange warning on the interface, suggesting predictive maintenance; S<60 indicates a risk state, with a red alarm on the interface, and the system issues an audible and visual alarm and automatically generates a diagnostic report which is uploaded to the main control module 16.

[0167] In this embodiment, the operator can click the "Failure Mode" button on the operation interface of the main control module 16 to view the failure mode and its warning confidence level; In this embodiment, when the system detects a continuous high-risk state, the result display and early warning unit will automatically feed back key feature data and multi-source signal segments to the evaluation unit for adaptive correction of the model threshold; at the same time, the system will trigger a cloud model correction task according to the strategy, so that the model parameters are incrementally updated in the cloud to improve the stability of subsequent evaluations. In this embodiment, the interface of the main control module 16 displays the real-time updated comprehensive safety status index and various real-time signal curves. The operator can click the "Generate Report" button, and the system will automatically create a diagnostic report that includes all parameters of this test, key data throughout the process, safety status index curves, failure mode confidence levels and conclusions. In this embodiment, the operator can click "View History" to review all data records of all braking experiments; the system will automatically save the complete multi-source data of this braking process for incremental learning of the model; In this embodiment, when the system determines that the confidence level of a certain braking process is in the gray range, the system will prompt the operator whether to store this data in the pending case library. After confirmation by the engineer, the complete multi-source data and related features will be stored in the pending case library for subsequent manual annotation and incremental learning of the model. In this embodiment, the operator can view the case by clicking "Case Library Management" and annotate it. The annotated data will be used for incremental learning of the system model, making the evaluation of similar working conditions more accurate in the future.

[0168] In addition, this application provides a train brake service safety status assessment system based on multi-source information fusion. This embodiment proposes a system implementation method in terms of general scalability and parameterized adaptation, enabling it to conduct service safety status assessments on brake components of different types, different material systems, and different geometric structures without changing the overall system structure.

[0169] In this embodiment, the system adopts a modular and parameterized software and hardware architecture, which can flexibly configure corresponding parameters according to the characteristics of the brake component under test. The adaptation content includes, but is not limited to, the material properties, geometric dimensions, ventilation structure, and friction pair pairing relationship of brake components such as brake discs, brake pads, and brake calipers.

[0170] As a concrete implementation of the system's general scalability and parameterized adaptability, this paper takes the replacement of the train brake from a cast iron brake disc + powder metallurgy brake pads to a carbon ceramic brake disc + organic composite brake pads as an example to illustrate the process of adjusting the system parameters after replacing the test components: In this embodiment, through the configurable parameter interface provided by the system software, the operator switches the brake disc material type from cast iron to carbon ceramic and inputs the corresponding thermophysical parameters, including thermal conductivity, specific heat capacity, coefficient of thermal expansion, and friction coefficient reference values; at the same time, the operator inputs the geometric dimension parameters of the new brake disc, including brake disc thickness, ventilation groove structure, and effective friction surface area; after receiving the above parameters, the system automatically adjusts the relevant thresholds in the data processing unit.

[0171] In this embodiment, based on the failure characteristics of carbon-ceramic brake discs, the existing sensor configuration is parametrically adjusted. For example, for industrial cameras, the image acquisition frame rate, resolution, and region of interest are adjusted, and a dedicated image processing algorithm for crack identification is enabled. All adjustments are completed through a software interface, without the need to replace or add / remove sensor hardware.

[0172] In this embodiment, by calling the attention weight parameters and decision layer network parameters that were pre-trained based on historical test data of carbon ceramic brake discs through the system interface, the system automatically loads the new model into the attention weighting layer and decision layer without redesigning the network structure or modifying the algorithm code.

[0173] In this embodiment, a small number of calibration tests are conducted on the new components used for the first time. The attention weighting layer automatically learns the importance distribution of each feature under the current operating condition through a gating network based on the real-time collected operating parameters and multi-source signal characteristics, and dynamically adjusts the attention weights so that the evaluation results gradually converge to the optimal accuracy during the subsequent braking process.

[0174] Through the above steps, the system of the present invention achieves rapid adaptation and accurate evaluation of braking components of different materials and structures without changing the overall architecture, redeveloping the software, or replacing the sensor hardware, fully demonstrating its modular and parametric design capabilities for general expansion.

[0175] In this embodiment, the system software provides a configurable parameter interface, allowing users to input corresponding thermophysical parameters, friction model coefficients, interface heat exchange coefficients, and related thresholds based on the material system of the brake component being tested, such as aluminum-based composite materials, cast iron materials, carbon ceramic materials, or other metal-based composite materials. For differences in geometric structure, the system parameters related to the brake disc thickness, ventilation groove structure, or effective area of ​​the friction surface are adjusted to enable the multi-source feature extraction process to match different structural components.

[0176] In this embodiment, the data acquisition unit supports flexible selection of different types of sensor components. Depending on the differences between the evaluation target and the object being measured, thermocouple arrays, infrared temperature measurement devices, fiber optic grating sensors, piezoelectric vibration sensors, acoustic emission sensors, and strain measurement devices can be selected. All sensors are integrated into the data acquisition unit through modular interfaces, and different sensor combinations will not affect the universality of the system's data processing flow and information fusion mechanism.

[0177] In this embodiment, the multi-source information fusion algorithm of the safety status assessment unit has general scalability; the system allows loading and running different fusion models, including multi-modal fusion methods based on attention mechanisms, feature weighting methods based on temporal correlation, or multi-source fusion methods based on statistical inference; different algorithm models are standardized before running to ensure that the feature normalization process, weight allocation logic, and calculation method of the comprehensive safety status index remain consistent, so that the system can maintain the comparability and stability of the assessment results after changing components, materials, or sensor configurations.

[0178] In this embodiment, the system's output unit provides standardized safety assessment results, including a comprehensive safety status index, key feature weights, and risk trends under typical operating conditions. Users can adjust the output format according to their assessment needs, such as generating structured test records or diagnostic reports, for subsequent analysis or maintenance plan development.

[0179] In this embodiment, to adapt to different operating environments such as high temperature, low temperature, snow or sandstorm environments, the system can work in conjunction with the corresponding environment simulation module 9. By loading environment-related compensation parameters, such as temperature drift correction coefficient, humidity interference factor or particle erosion correction factor, the fusion model can still maintain high stability under different external conditions.

[0180] Through the modular, parameterized, and algorithm-extended design described in this embodiment, the system of the present invention can not only cover a variety of braking component types and material systems, but also maintain the consistency and reliability of the evaluation process under different structural features, different sensor combinations, and different working conditions.

[0181] As a specific method for obtaining environmentally relevant compensation parameters, temperature drift correction factor, humidity interference factor and particle erosion correction factor are obtained by combining experimental calibration with historical data statistics.

[0182] In this embodiment, under standard environmental conditions, a braking test is conducted on the brake using a test bench, simultaneously acquiring multi-source physical signals, and obtaining a reference feature vector through a data processing unit. This serves as the benchmark for subsequent comparisons.

[0183] In this embodiment, the same braking condition is repeated under the same target environmental conditions for the same brake, and feature vectors are collected. Calculate the environmental deviation of each feature. This forms a calibration dataset.

[0184] In this embodiment, calibration test data under different environmental conditions were collected, with environmental parameters temperature T, humidity H, and dust concentration D as independent variables, and the deviation of each characteristic as the independent variable. Using the dependent variable, an environmental compensation model is established through multiple linear regression: ; in, , , Let be the environmental compensation parameter corresponding to the i-th feature.

[0185] In this embodiment, the compensation parameters are stored in the system in the form of a matrix. During actual online evaluation, the input features are corrected based on the environmental parameters collected in real time.

[0186] The modified feature vector is used to replace the original feature input in the safety status assessment unit, thereby eliminating the interference of environmental factors on multi-source signals.

[0187] It should be noted that the acquisition of the above compensation parameters can be completed in one go during the test bench calibration stage, and the compensation model can be dynamically called in actual operation after it is established. When the system application scenario is expanded to new environmental conditions, the compensation model can be incrementally updated through supplementary calibration tests.

[0188] A method for assessing the service status of train brakes based on AI and multi-source information employs any of the systems described above, and the method includes: S1. On the comprehensive performance test bench for train brakes, the target test environment is set through the main control module, and the initial braking speed, braking pressure, axle load and wheel-rail excitation conditions are set to simulate various typical service conditions faced by the train in actual operation; different types of brake discs and brake pads can also be selected; the test bench carries the target brake operation test and provides the test environment for the target brake; S2. Synchronous Signal Acquisition: During braking, the data acquisition unit synchronously acquires multi-source physical signals of the target brake, including but not limited to thermal, vibration, acoustic, mechanical, motion, and visual signals; simultaneously, the verification unit synchronously collects abrasive particles generated during braking; and acquires multi-source physical signals during the operation of the target brake. S3. Using the data processing unit, the raw signal collected in step S2 is preprocessed to extract key feature parameters related to brake performance degradation, and then standardized and fused to form a standardized feature vector. At the same time, based on the amplitude variation relationship of the multi-source signals in time and space, the fault is preliminarily screened and the location is identified to obtain the correlation results between the multi-source physical signals, and a corresponding primary fusion feature vector is generated based on the correlation results. S4. Safety Status Assessment: Input the standardized feature vector obtained in step S3 into the pre-stored mapping relationship model in the safety status assessment unit. Through multi-source information fusion calculation, dynamically output a comprehensive safety status index S from 0 to 100, and simultaneously output the warning confidence level for various failure modes.

[0189] Construct a multi-condition mapping model library, which includes multiple types of condition mapping models; Based on the acquired braking condition parameters and environmental condition parameters, the corresponding condition mapping model is selected. The corresponding advanced fusion feature vector is generated by adaptively weighting and reconstructing the primary fusion feature vector. Based on the advanced fusion feature vector, the safety status assessment result is output through a main deep neural network and a multi-failure mode expert classifier. The safety status assessment result includes: a comprehensive safety status index and a warning confidence level of at least one failure mode.

[0190] The method further includes: a. Results Display and Early Warning: Visualize the comprehensive safety status index and failure mode early warning confidence level output in step S4; and make judgments and issue early warnings according to the set graded early warning strategy: trigger a red alarm when S < 60, trigger an orange alert when 60 ≤ S < 85, and display the health status when S ≥ 85. b. Post-verification of abrasive particles: After the safety status assessment is completed, the verification unit is invoked to perform offline analysis on the collected abrasive particles to obtain the number, particle size distribution, morphology, and metal composition characteristics of the abrasive particles; the abrasive degradation index obtained from the analysis is compared and verified offline with the comprehensive safety status index S and the confidence level of wear-related failure modes output in step S4; if the verification is consistent, the credibility of the assessment results is enhanced; if there is a significant deviation, the case is marked as a sample to be verified, and a basis is provided for model optimization; c. Learning and Improvement: The system establishes a pending case library. When the confidence level output by the safety status assessment unit is in the gray area of ​​60% to 80%, the complete multi-source data and features of the braking process are automatically stored in the pending case library. After the pending cases are verified and labeled by personnel, the labeled data is used for incremental learning of the mapping relationship model to achieve continuous evolution of the system's diagnostic capabilities.

[0191] Through the above embodiments, the results of this application show the following beneficial effects of the early warning unit: The comprehensive safety status index S is divided into three levels: healthy (green), alert (orange), and risk (red). The interface is intuitive and operators can quickly judge the braking status, reducing misjudgment and operational difficulty.

[0192] Risk conditions automatically trigger audible and visual alarms and automatically generate a complete diagnostic report containing test parameters, state curves, failure confidence levels, and assessment conclusions, which is then uploaded to the central control module for rapid handling, record archiving, and accountability tracing.

[0193] It displays the comprehensive safety status index, curves of various physical quantities, and failure mode confidence in real time, and supports one-click viewing and switching, which facilitates test monitoring, working condition debugging, and fault reproduction analysis.

[0194] It allows users to view all historical test data, fully save multi-source signals, evaluation results, and early warning records, and provide data support for product optimization, operation and maintenance analysis, and model iteration.

[0195] Samples with confidence levels in the gray range are automatically suggested and stored in the case library. After manual annotation, they are used for incremental model learning to continuously improve the system's evaluation accuracy for complex and edge cases.

[0196] Continuous high-risk conditions are automatically fed back to the assessment unit, triggering threshold correction and cloud model updates, enabling the system to adapt to changes in operating conditions, component aging, and environmental differences, resulting in more stable long-term assessments.

[0197] This invention proposes a train brake service safety status assessment system based on multi-source information fusion. This embodiment proposes a method for implementing the system in terms of general scalability and parameterized adaptation, enabling it to conduct service safety status assessments on brake components of different types, material systems, and geometric structures without changing the overall system structure.

[0198] In this embodiment, the system adopts a modular and parameterized software and hardware architecture, which can flexibly configure corresponding parameters according to the characteristics of the brake component under test. The adaptation content includes, but is not limited to, the material properties, geometric dimensions, ventilation structure, and friction pair pairing relationship of brake components such as brake discs, brake pads, and brake calipers.

[0199] In this embodiment, the system software provides a configurable parameter interface, allowing users to input corresponding thermophysical parameters, friction model coefficients, interface heat exchange coefficients, and related thresholds based on the material system of the brake component being tested, such as aluminum-based composite materials, cast iron materials, carbon ceramic materials, or other metal-based composite materials. For differences in geometric structure, the system parameters related to the brake disc thickness, ventilation groove structure, or effective area of ​​the friction surface are adjusted to enable the multi-source feature extraction process to match different structural components.

[0200] In this embodiment, the data acquisition unit supports flexible selection of different types of sensor components. Depending on the differences between the evaluation target and the object being measured, thermocouple arrays, infrared temperature measurement devices, fiber optic grating sensors, piezoelectric vibration sensors, acoustic emission sensors, and strain measurement devices can be selected. All sensors are integrated into the data acquisition unit through modular interfaces, and different sensor combinations will not affect the universality of the system's data processing flow and information fusion mechanism.

[0201] In this embodiment, the multi-source information fusion algorithm of the safety status assessment unit has general scalability; the system allows loading and running different fusion models, including multi-modal fusion methods based on attention mechanisms, feature weighting methods based on temporal correlation, or multi-source fusion methods based on statistical inference; different algorithm models are standardized before running to ensure that the feature normalization process, weight allocation logic, and calculation method of the comprehensive safety status index remain consistent, so that the system can maintain the comparability and stability of the assessment results after changing components, materials, or sensor configurations.

[0202] In this embodiment, the system's output unit provides standardized safety assessment results, including a comprehensive safety status index, key feature weights, and risk trends under typical operating conditions. Users can adjust the output format according to their assessment needs, such as generating structured test records or diagnostic reports, for subsequent analysis or maintenance plan development.

[0203] In this embodiment, to adapt to different operating environments such as high temperature, low temperature, snow or sandstorm environments, the system can work in conjunction with the corresponding environment simulation module 9. By loading environment-related compensation parameters, such as temperature drift correction coefficient, humidity interference factor or particle erosion correction factor, the fusion model can still maintain high stability under different external conditions.

[0204] Through the modular, parameterized, and algorithm-extended design described in this embodiment, the system of the present invention can not only cover a variety of braking component types and material systems, but also maintain the consistency and reliability of the evaluation process under different structural features, different sensor combinations, and different working conditions.

[0205] The device provided in this application embodiment has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0206] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0207] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0208] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0209] In the description of the embodiments of this application, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments of this application. In the embodiments of this application, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in the embodiments of this application, as well as the features of different embodiments or examples.

[0210] Furthermore, 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 number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of embodiments of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0211] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order according to the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0212] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0213] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present application should be included within the scope of protection of the present application.

Claims

1. A train brake service status assessment system based on AI and multi-source information, characterized in that, include: The operating condition simulation unit is used to carry out the operation test of the target brake and provide a test environment for the target brake. The data acquisition unit is used to acquire multi-source physical signals during the operation of the target brake; A data processing unit is used to obtain a primary fusion feature vector among the multi-source physical signals, wherein the primary fusion feature vector is obtained by normalizing the multi-source physical signals; A safety status assessment unit is used to construct a multi-condition mapping model library, which includes multiple types of condition mapping models. Based on the acquired braking condition parameters and environmental condition parameters, a target condition mapping model is selected. A corresponding advanced fusion feature vector is generated by adaptively weighting and reconstructing the primary fusion feature vector. This advanced fusion feature vector is then sequentially input into the main deep neural network and a multi-failure mode expert classifier, outputting a safety status assessment result. The advanced fusion feature vector is used to rank the importance of the multi-source physical signals, and the safety status assessment result includes: a comprehensive safety status index and a warning confidence level for at least one failure mode; The results display and early warning unit is used to visualize the security status assessment results and execute graded risk warnings according to preset strategies.

2. The train brake service status assessment system based on AI and multi-source information according to claim 1, characterized in that, The data processing unit is also used for, The correlation results among the multi-source physical signals are obtained, and the correlation results are used to characterize fault location and the health status of the operating condition simulation unit. The correlation results among the multi-source physical signals include: Vibration transmission indices between signals and transmission paths and attenuation characteristics between signals; The vibration transmission indicators include: correlation coefficient and cross-correlation delay; The transmission path and attenuation characteristics include: transmission path gain index and attenuation characteristics; The vibration transmission index is used to generate a fault location sample set by combining the safety status assessment results. Each fault location sample in the fault location sample set includes a safety status assessment result and its corresponding vibration transmission index. The transmission path and attenuation characteristics are used to quantify the attenuation characteristics of vibration between different structural components and to generate a health state sample set of the working condition simulation unit. The health state sample set includes a health state of the working condition simulation unit and its corresponding transmission path and attenuation characteristics.

3. The train brake service status assessment system based on AI and multi-source information according to claim 1, characterized in that, The operating condition simulation unit includes: A rigid gantry frame (15) spans the top of the test bench base (1) to provide installation support for the test execution of the target brake (14) and ensure that the braking load is stably applied to the target brake (14). An environmental isolation chamber (10) is provided outside the target brake (14), and its bottom end is sealed to the top of the test bench base (1) to provide a closed environmental space for testing the target brake (14); An environmental simulation module (9) is connected to the environmental isolation cavity (10) and is used to construct a controllable circulating airflow environment in the environmental isolation cavity (10); The wheel-rail simulation module (8) passes through the environmental isolation cavity (10) at one end and forms a rolling contact connection with the wheel of the axle load simulation module 11. Its bottom end is set on the test bench base (1) to simulate the wheel-rail contact relationship and wheelset rotation state during train operation. The axle load simulation module (11) is set on the columns on both sides of the rigid gantry frame (15) to apply a vertical load to the target brake (14) by hydraulic pressure to simulate the effect of different vehicle axle loads on the brake, and at the same time realize wheel-rail coupling through the rolling contact relationship between the wheel and the rail wheel. The main control module (16) is connected to the servo driver and hydraulic controller via a data bus to realize the synchronous issuance of all action commands and status monitoring.

4. The train brake service status assessment system based on AI and multi-source information according to claim 3, characterized in that, The data acquisition unit includes: A motion signal sensing unit, wherein the motion signal sensing unit is used to acquire the motion signal of the target brake; A thermal signal sensing unit is used to acquire the thermal signal of the target brake. A vibration signal sensing unit, wherein the vibration signal sensing unit is used to acquire the vibration signal of the target brake; A mechanical signal sensing unit is used to acquire the mechanical signal of the target brake; A visual signal sensing unit, wherein the visual signal sensing unit is used to acquire the visual signal of the target actuator; An acoustic signal sensing unit is used to acquire the acoustic signal of the target brake.

5. The train brake service status assessment system based on AI and multi-source information according to claim 1, characterized in that, The safety status assessment model in the safety status assessment unit includes: Feature layer, attention weighting layer, multiphysics state representation layer and decision layer; The feature layer is used to receive the primary fused feature vector; A multi-physics state representation layer is used to transform the primary fusion feature vector into a corresponding representation vector through M set physical fields; The attention weighting layer is used to dynamically weight and reconstruct the representation vector of each physical field by combining braking condition parameters and environmental condition parameters to generate a high-level fusion feature vector. The decision layer is used to combine the advanced fusion feature vectors and perform parallel decision calculations through a main deep neural network and multiple expert classifiers for specific failure modes, outputting a comprehensive safety status index and warning confidence levels for various failure modes.

6. The train brake service status assessment system based on AI and multi-source information according to claim 5, characterized in that, The attention weighting layer is configured to dynamically weight and reconstruct the representation vector of each physical field based on braking condition parameters and environmental condition parameters to generate a high-level fusion feature vector. The attention weighting layer adjusts the weights of each physical field representation vector in the following manner: The trainable parameters of the attention weighting layer are initialized based on historical data to generate the initialization parameter set of the attention weighting layer. The initialization parameter set consists of the initialization weight parameters of the physical field representation vector under various braking conditions and environmental conditions. Calculate the normalized eigenvector for each of the physical field representation vectors; The weights of the standardized feature vectors are set based on the initialization parameter set; A primary fusion feature vector is formed based on all the standardized feature vectors. ; The primary fusion feature vector Hidden representations of each feature are generated using a gating network. Then, the original attention score is calculated using the trained scoring vector. ; ； ； in, The attention score vector is obtained by initializing the parameter set. The input is the primary fusion feature vector; Features The original attention score; Features The feature transformation weight matrix; Calculate features using the softmax function Final weight : ； in, Let be the original attention parameters for the j-th feature; The real-time operating conditions include: acquiring braking operating condition parameters and environmental operating condition parameters.

7. The train brake service status assessment system based on AI and multi-source information according to claim 5, characterized in that, In the decision-making layer, the steps for outputting the comprehensive safety status index and the early warning confidence levels for various failure modes are as follows: Based on environmental operating parameters, environmental operating conditions are classified into various types, resulting in multiple types of operating conditions. Obtain environmental data and primary fusion feature vectors for each type of operating condition. ; The comprehensive safety status index is calculated using the performance index mapping method. And based on the true labels of various failure modes, y i Construct a training sample set ; Through the training sample set Train a mapping relationship model for each of the aforementioned working conditions. The mapping model includes: a main deep neural network and multiple expert classifiers for specific failure modes; Select a target mapping relationship model based on the current type of working condition; The advanced fusion feature vector is input into the target mapping relationship model, and the comprehensive safety status index of the target brake and the warning confidence of various failure modes are output.

8. The train brake service status assessment system based on AI and multi-source information according to any one of claims 1-7, characterized in that, The system also includes: The verification unit includes an exhaust pipe, a particle stratified filtration and collection device (31), and an abrasive particle detection device. The verification unit is used to analyze the collected abrasive particle samples to obtain analysis results, and to perform post-verification on the comprehensive safety status index and the early warning confidence of various failure modes output in the decision-making layer based on the analysis results. The discharge pipe is used to connect the environmental isolation chamber (10) and the particle stratification filtration and collection device (31) to form a closed loop. The abrasive particle detection device includes a laser particle detection device (32), a particle concentration detection device (30), and an auxiliary aerodynamic device (29). The laser particle detection device (32) is used to monitor the particle size distribution of the exhaust gas particles in real time; The particle concentration detection device (30) monitors the concentration of each filter layer through a dynamic detection opening; The auxiliary aerodynamic device (29) works in conjunction with a negative feedback mechanism to regulate airflow; The collected abrasive particles were analyzed to obtain analytical results. Based on these results, the confidence levels of the comprehensive safety status index and various failure modes output by the decision-making level were subsequently validated, including: The abrasive particle detection device analyzes the collected abrasive particle samples to obtain the number, particle size distribution, morphology and structure and metal composition characteristics of the abrasive particles. The morphology, composition and number of abrasive particles can directly reflect the real degradation mechanism of the friction pair material. Flaky or layered abrasive grains indicate thermal fatigue or surface spalling processes, while sharp-edged abrasive grains characterize abrasive wear-dominated failure modes. High-iron-content abrasive grains reflect increased exposure of the brake disc substrate, and a sudden increase in the number of abrasive grains usually corresponds to a transition stage in the wear rate.

9. The train brake service status assessment system based on AI and multi-source information according to claim 8, characterized in that, The results display and early warning unit is used to analyze the comprehensive safety status index S and the early warning confidence levels of various failure modes. Visualize the situation and provide risk alerts to operators based on the set warning strategies; The system provides three states: S≥85 indicates a healthy state, displayed in green on the interface, indicating that the brake disc is performing well. When 60≤S<85, a state of alert is indicated, with an orange warning on the interface, suggesting predictive maintenance. S<60 indicates a risky state, triggering a red alarm on the interface. The system will issue an audible and visual alarm and automatically generate a diagnostic report, which will be uploaded to the main control module (16).

10. A method for evaluating the service status of train brakes based on AI and multi-source information using the system described in any one of claims 1-9, characterized in that, The methods include: S1. Conduct operational testing of the target brake and provide a testing environment for the target brake; S2. Acquire multi-source physical signals during the operation of the target brake; S3. Obtain the correlation results between the multi-source physical signals, and generate the corresponding primary fusion feature vector based on the correlation results; S4. Construct a multi-condition mapping model library, which includes multiple types of condition mapping models; Based on the acquired braking condition parameters and environmental condition parameters, the corresponding condition mapping model is selected. The corresponding advanced fusion feature vector is generated by adaptively weighting and reconstructing the primary fusion feature vector. Based on the advanced fusion feature vector, the safety status assessment result is output through a main deep neural network and a multi-failure mode expert classifier. The safety status assessment result includes: a comprehensive safety status index and a warning confidence level of at least one failure mode.

Images