Determining and measuring device for braking distance of freight train

Abstract

The invention discloses a device for determining and measuring the braking distance of a freight train, and belongs to the technical field of train braking safety protection, and the device comprises a multi-mode sensing module which integrates the following units: a wheel-rail adhesion monitoring unit which collects the friction coefficient and slip rate of a wheel-rail contact surface in real time based on a laser Doppler velocimeter and a strain gauge sensor; the environment sensing unit is used for detecting rail surface humidity, foreign matter and frost coverage states through a temperature and humidity sensor and an infrared camera; and a train state acquisition unit. According to the invention, through combination of multi-modal sensing and an advanced algorithm, comprehensive monitoring and accurate braking are realized; federal learning breaks a data barrier, and algorithm generalization is enhanced. Adhesion adjustment optimizes braking force distribution and anti-skid control, and stable and safe braking is guaranteed. The carbon emission optimization module promotes green energy conservation, the fault self-healing function, the response time is shortened, the self-healing rate is increased, the requirements of the heavy-load train are met, and the safety, reliability and sustainability of a train braking system are comprehensively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of train braking safety protection, and in particular relates to a device for determining and measuring the braking distance of a freight train. Background Art

[0002] Freight railway transportation has the characteristics of long transportation distance, large carrying capacity and low transportation cost. In freight railway transportation, the automatic train protection system is mainly used to determine the driving curve in real time according to the line data and the train's own status to achieve the train's overrun and overtaking prevention. During the train's operation, for the braking requirements on the train (such as normal braking or emergency braking), ATP will use the safety braking model to timely determine the safe braking distance. The safety braking model is a modeling of the actual braking process of the train. The model needs to take into account various adverse conditions during the train braking process, including speed and distance measurement errors, wheel-rail adhesion conditions, slope, etc. The safety braking model is used to control the train speed to avoid the risk of the train overrunning the signal and ensure the safety of the train operation.

[0003] Traditional methods for calculating the braking distance of freight trains have significant technical limitations. First, their static modeling is based on the assumption of a constant friction coefficient and lacks quantitative analysis of the dynamic changes in the wheel-rail adhesion coefficient and the effects of multi-physics field coupling. Under uneven load conditions, the braking force distribution error reaches 12%-18%. Second, the perception system with a single-source sensing architecture is susceptible to electromagnetic interference, making it difficult to fully monitor changes in the track surface state. Furthermore, industry data is isolated, the massive historical data of a single train is not effectively utilized, and there is a lack of cross-vehicle collaborative modeling, resulting in poor algorithm generalization and significant error amplification when applied across lines. Finally, existing methods do not incorporate carbon footprint into control objectives, have low braking energy recovery efficiency, and lack fault response capabilities. They use fixed threshold alarms, rely on manual labor when sensors fail, have an average response time of more than 30 seconds, and have a self-healing rate of less than 75%, which cannot meet the continuous braking and green development needs of heavy-load trains.

[0004] Based on this, the present invention designs a device for determining and measuring the braking distance of a freight train to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to solve the above-mentioned problem and to propose a device for determining and measuring the braking distance of a freight train.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A device for determining and measuring the braking distance of a freight train, comprising:

[0008] Multimodal sensing module: Integrates the following units:

[0009] Wheel-rail adhesion monitoring unit: Based on a laser Doppler velocimeter and strain gauge sensors, it collects the friction coefficient and slip rate of the wheel-rail contact surface in real time;

[0010] Environmental perception unit: detects rail surface humidity, foreign matter, and frost coverage through temperature and humidity sensors and infrared cameras;

[0011] Train status acquisition unit: collects train formation length, load distribution, brake cylinder pressure and on-board ATP command signals.

[0012] Federated Learning Engine:

[0013] Build a distributed braking model training network to aggregate braking behavior data from multiple trains and dynamically update the wheel-rail adhesion prediction model and braking phase division parameters;

[0014] A differential privacy mechanism is used to protect train operation data, and the model update cycle is ≤10 minutes.

[0015] Dynamic braking stage division module:

[0016] The braking process is divided into the response phase (0-t1), the propagation phase (t1-t2), the adhesion adjustment phase (t2-t3) and the stable phase (t3-t4);

[0017] The duration thresholds (t1 to t4) of each stage are dynamically adjusted based on the change rate of the adhesion coefficient. The calculation formula is:

[0018]

[0019] Among them, μcurrent is the real-time stickiness coefficient, μnominal is the nominal stickiness coefficient, and ki is the stage coefficient of federated learning optimization;

[0020] Braking distance calculation module:

[0021] The total braking distance is calculated using a four-stage integration model:

[0022]

[0023] Among them, ηprop is the braking propagation efficiency factor, γstable is the attenuation coefficient in the stable stage;

[0024] The multimodal sensing module supports wireless ad hoc network communication, and the data transmission delay is ≤20ms.

[0025] As a further description of the above technical solution:

[0026] The federated learning engine includes:

[0027] Local model training unit: A lightweight LSTM network is deployed locally on each train. The input is wheel-rail adhesion data, environmental data, and historical braking curves, and the output is the adjustment of the stage coefficient ki.

[0028] Global aggregation unit: A weighted average algorithm is used to fuse the parameters of each train model. The weight is determined by the data quality index (DQI). The DQI calculation formula is:

[0029] DQI = α·Data integrity + β·Sensor accuracy level + γ·Braking success rate

[0030] Among them, α+β+γ=1, default values ​​α=0.4, β=0.3, γ=0.3.

[0031] As a further description of the above technical solution:

[0032] The adhesion adjustment phase includes the following functions:

[0033] Dynamic braking force distribution: According to the load difference of each car in the train, the brake cylinder pressure is adjusted proportionally. The distribution formula is:

[0034]

[0035] Where Pj is the braking pressure of the jth carriage, Wj is its load, W is the average load, λ is the adhesion sensitivity coefficient (default is 0.5), az is the longitudinal acceleration, and δ is the acceleration sensitivity coefficient (default is 0.1);

[0036] Anti-skid control: When the wheelset slip rate is detected to be greater than 5%, the graded decompression strategy is triggered, prioritizing the reduction of braking force on the car with the highest slip rate.

[0037] As a further description of the above technical solution:

[0038] The device integrates a carbon emission optimization module, and its functions include:

[0039] Braking energy consumption model: Calculate carbon emissions per unit braking distance

[0040] C emit =f(P avg , t brake , η rec )=(1-η rec )·P avg ·t brake ,

[0041] Among them, Pavg is the average braking power, tbrake is the braking time, η rec is the regenerative braking energy recovery efficiency factor;

[0042] Multi-objective decision-making: Under the constraint of safe braking distance, the NSGA-II algorithm is used to optimize the braking curve to minimize carbon emissions.

[0043] As a further description of the above technical solution:

[0044] The device supports fault self-healing function:

[0045] When the wheel-rail adhesion monitoring unit fails, it automatically switches to the adhesion estimation algorithm based on the IMU (inertial measurement unit), and the error compensation amount is dynamically corrected by the federated learning engine;

[0046] When the on-board ATP communication is interrupted, the edge computing node is enabled to execute local braking decisions with a synchronization accuracy of ≥95%.

[0047] As a further description of the above technical solution:

[0048] The environmental perception unit also includes a millimeter-wave radar for detecting the distance and movement speed of obstacles in front of the train in real time. When an obstacle is detected, it triggers an early warning or emergency braking command based on the braking distance calculation result.

[0049] As a further description of the above technical solution:

[0050] The onboard ATP command signal collected by the train status acquisition unit includes target speed and speed limit information. The braking distance calculation module dynamically adjusts the braking strategy according to the ATP command signal to ensure that the train reaches the target speed within the specified distance.

[0051] As a further description of the above technical solution:

[0052] The dynamic braking force distribution function is also provided with a braking force upper limit threshold. When the calculated braking pressure of a certain carriage exceeds the upper limit threshold, the braking pressure of all carriages is reduced proportionally to prevent the wheels from locking due to excessive braking force.

[0053] As a further description of the above technical solution:

[0054] The carbon emission optimization module also includes a real-time carbon emissions display interface, which shows the driver the real-time carbon emissions and cumulative carbon emissions during the current braking process through the on-board display screen. In the fault self-healing function, when the system detects a fault and switches to a backup algorithm or mode, the fault information and current operating status data are sent to the ground control center through the on-board communication module to facilitate remote diagnosis and maintenance by ground personnel.

[0055] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0056] 1. By integrating wheel-rail adhesion, environmental perception, and train status acquisition units through a multimodal sensing module, the vulnerability of traditional single-source sensing is overcome, enabling comprehensive monitoring of the wheel-rail contact surface, environment, and train operating status. Combined with wireless ad hoc network communication, data transmission delay is ≤20ms, providing real-time, reliable data support for precise braking. The dynamic braking stage division module adjusts the duration of each stage based on the rate of change of the adhesion coefficient. The braking distance calculation module uses a four-stage integral model, addressing the lack of quantitative characterization capabilities of traditional static modeling under the multi-physics field coupling effect. It effectively copes with uneven load distribution conditions, reduces braking force distribution errors, improves the accuracy of braking distance calculations, and ensures safe train operation.

[0057] 2. A distributed training network is constructed through a federated learning engine to aggregate braking behavior data from multiple trains. Local model training units and global aggregation units work together to fuse model parameters from each train using a weighted average algorithm, breaking down data silos and effectively utilizing massive historical data from individual trains for model iteration. Furthermore, a differential privacy mechanism is employed to protect operational data. While ensuring data security, cross-vehicle collaborative modeling is achieved, significantly improving the generalization of the braking algorithm in complex scenarios and reducing errors in cross-line applications. 3. Dynamic braking force distribution during the adhesion adjustment phase precisely adjusts the brake cylinder pressure of each car based on factors such as car load differences, adhesion coefficient, and longitudinal acceleration, and sets an upper braking force threshold to prevent wheel lock. Anti-skid control triggers a graded decompression strategy when the wheelset slip ratio exceeds 5%, effectively preventing wheel lock and further improving train stability and safety during braking.

[0058] 4. The carbon emissions optimization module calculates carbon emissions using a braking energy consumption model and employs the NSGA-II algorithm to optimize the braking curve within safe braking distance constraints, minimizing carbon emissions. This module incorporates carbon footprint indicators into braking control targets, improves regenerative braking utilization, and promotes the development of green and energy-efficient freight train braking systems. Furthermore, a real-time carbon emissions display interface allows drivers to understand energy consumption in real time, enhancing their awareness of energy conservation.

[0059] 5. If the wheel-rail adhesion monitoring unit fails, it automatically switches to an IMU-based adhesion estimation algorithm, and the federated learning engine dynamically corrects the error compensation. If onboard ATP communication is interrupted, the edge computing node initiates local braking decisions with a synchronization accuracy of ≥95%, and transmits fault information and operating status data to the ground control center. Compared with traditional fixed-threshold alarm mechanisms, this significantly shortens fault response time and improves system self-healing rate, meeting the continuous braking requirements of heavy-load trains and ensuring safe operation and autonomous control capabilities in the event of a fault. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1This is a schematic diagram of the architecture of a device for determining and measuring the braking distance of a freight train proposed by the present invention;

[0061] Figure 2 This is a schematic diagram of the data flow of the federated learning layer of the device for determining and measuring the braking distance of a freight train proposed by the present invention;

[0062] Figure 3 This is a schematic diagram of the framework of the use steps of a device for determining and measuring the braking distance of a freight train proposed by the present invention. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0064] Please see the attached Figure 1 -Attached Figure 3 The present invention provides a technical solution: a device for determining and measuring the braking distance of a freight train, comprising: a multimodal sensing module: integrating the following units:

[0065] Wheel-rail adhesion monitoring unit: Based on a laser Doppler velocimeter and strain gauge sensors, it collects the friction coefficient and slip rate of the wheel-rail contact surface in real time;

[0066] Environmental perception unit: detects rail surface humidity, foreign matter, and frost coverage through temperature and humidity sensors and infrared cameras;

[0067] Train status acquisition unit: collects train formation length, load distribution, brake cylinder pressure and on-board ATP command signals.

[0068] Federated Learning Engine:

[0069] Build a distributed braking model training network to aggregate braking behavior data from multiple trains and dynamically update the wheel-rail adhesion prediction model and braking phase division parameters;

[0070] A differential privacy mechanism is used to protect train operation data, and the model update cycle is ≤10 minutes.

[0071] Dynamic braking stage division module:

[0072] The braking process is divided into the response phase (0-t1), the propagation phase (t1-t2), the adhesion adjustment phase (t2-t3) and the stable phase (t3-t4);

[0073] The duration thresholds (t1 to t4) of each stage are dynamically adjusted based on the change rate of the adhesion coefficient. The calculation formula is:

[0074]

[0075] Among them, μcurrent is the real-time stickiness coefficient, μnominal is the nominal stickiness coefficient, and ki is the stage coefficient of federated learning optimization;

[0076] Braking distance calculation module:

[0077] The total braking distance is calculated using a four-stage integration model:

[0078]

[0079] Among them, ηprop is the braking propagation efficiency factor, γstable is the attenuation coefficient in the stable stage;

[0080] The multimodal sensing module supports wireless ad hoc network communication, and the data transmission delay is ≤20ms.

[0081] The federated learning engine includes:

[0082] Local model training unit: A lightweight LSTM network is deployed locally on each train. The input is wheel-rail adhesion data, environmental data, and historical braking curves, and the output is the adjustment of the stage coefficient ki.

[0083] Global aggregation unit: A weighted average algorithm is used to fuse the parameters of each train model. The weight is determined by the data quality index (DQI). The DQI calculation formula is:

[0084] DQI = α·Data integrity + β·Sensor accuracy level + γ·Braking success rate

[0085] Among them, α+β+γ=1, default values ​​α=0.4, β=0.3, γ=0.3.

[0086] The adhesion adjustment phase includes the following functions:

[0087] Dynamic braking force distribution: According to the load difference of each car in the train, the brake cylinder pressure is adjusted proportionally. The distribution formula is:

[0088]

[0089] Where Pj is the braking pressure of the jth carriage, Wj is its load, W is the average load, λ is the adhesion sensitivity coefficient (default is 0.5), az is the longitudinal acceleration, and δ is the acceleration sensitivity coefficient (default is 0.1);

[0090] Anti-skid control: When the wheelset slip rate is detected to be greater than 5%, the graded decompression strategy is triggered, prioritizing the reduction of braking force on the car with the highest slip rate.

[0091] The device integrates a carbon emission optimization module with the following functions:

[0092] Braking energy consumption model: Calculate carbon emissions per unit braking distance

[0093] C emit =f(P avg , t brake , η rec )=(1-η rec )·P avg ·t brake ,

[0094] Among them, Pavg is the average braking power, tbrake is the braking time, η rec is the regenerative braking energy recovery efficiency factor;

[0095] Multi-objective decision-making: Under the constraint of safe braking distance, the NSGA-II algorithm is used to optimize the braking curve to minimize carbon emissions.

[0096] The device supports fault self-recovery function:

[0097] When the wheel-rail adhesion monitoring unit fails, it automatically switches to the adhesion estimation algorithm based on the IMU (inertial measurement unit), and the error compensation amount is dynamically corrected by the federated learning engine;

[0098] When the on-board ATP communication is interrupted, the edge computing node is enabled to execute local braking decisions with a synchronization accuracy of ≥95%.

[0099] The environmental perception unit also includes a millimeter-wave radar, which is used to detect the distance and speed of obstacles in front of the train in real time. When an obstacle is detected, it triggers a warning or emergency braking command based on the braking distance calculation result;

[0100] The onboard ATP command signal collected by the train status acquisition unit includes target speed and speed limit information. The braking distance calculation module dynamically adjusts the braking strategy based on the ATP command signal to ensure that the train reaches the target speed within the specified distance.

[0101] The dynamic braking force distribution function also has a braking force upper limit threshold. When the calculated braking pressure of a car exceeds the upper limit threshold, the braking pressure of all cars is proportionally reduced to prevent excessive braking force from causing wheel lock.

[0102] The carbon emission optimization module also includes a real-time carbon emissions display interface, which shows the driver the real-time and cumulative carbon emissions during the current braking process through the on-board display screen. In the fault self-healing function, when the system detects a fault and switches to a backup algorithm or mode, it sends fault information and current operating status data to the ground control center through the on-board communication module to enable ground personnel to perform remote diagnosis and maintenance.

[0103] Example 1: Calculation of braking distance under normal operating conditions

[0104] Assume that a freight train has a length of 20 carriages, a total load of 1,200 tons, and is running at a speed of 80 km / h (about 22.2 m / s) on a dry and straight track. The onboard ATP instruction requires the train to reduce its speed to 40 km / h (about 11.1 m / s) within 1,500 meters.

[0105] 1. Data collection and transmission After the train starts, the multimodal sensing module enters the working state:

[0106] Wheel-rail adhesion monitoring unit: Laser Doppler velocimeter and strain gauge sensor work together to collect real-time wheel-rail contact surface data and obtain the friction coefficient (μ = 0.3) (nominal adhesion coefficient (μ nominal =0.35), the slip rate is 2%, and the data is transmitted to the core processing unit via wireless ad hoc network communication with a delay of 15ms.

[0107] Environmental perception unit: The temperature and humidity sensors detected that the track surface humidity was 30%, the infrared camera confirmed that there were no foreign objects or frost cover, and the millimeter-wave radar detected that there were no obstacles within 500 meters ahead. Environmental data was also transmitted to the system with low latency.

[0108] Train status acquisition unit: obtains the length of the train, the uniform distribution of the load of each carriage (the load of a single carriage (Wj = 60 tons), the initial pressure of the brake cylinder (P base =500kPa, as well as the target speed and speed limit information in the on-board ATP command, and quickly upload the data.

[0109] 2. Model Training and Parameter Optimization: The federated learning engine uses historical braking data from multiple trains and optimizes the distributed training network to obtain the following phase coefficients: k1 = 0.2, k2 = 0.3, k3 = 0.4, and k4 = 0.1. Simultaneously, the local model training unit, based on a lightweight LSTM network, processes input data on wheel-rail adhesion, environmental conditions, and historical braking curves. The output parameter adjustments are then aggregated globally to improve model prediction accuracy.

[0110] 3. Braking stage division and distance calculation

[0111] Dynamic braking stage division: Dynamic braking stage division module is based on the formula Calculate the duration of each stage and get the response stage

[0112] Dissemination stage

[0113] Adhesion adjustment stage

[0114] Stable phase

[0115] Braking distance calculation: The braking distance calculation module adopts a four-stage integral model, combining the real-time speed v(t), the braking propagation efficiency factor (η prop =0.8), stable phase attenuation coefficient (γ stable =0.9), calculate the total braking distance by numerical integration method (such as trapezoidal integration method)

[0116]

[0117] Finally, we get (S total =1480) meters, meeting the 1500-meter braking distance required by the ATP directive.

[0118] Example 2: Braking Adjustment in Complex Environments

[0119] When a train is traveling at 70 km / h (about 19.4 m / s) and encounters rain and snow, the track surface humidity rises to 80%, the wheel-rail adhesion coefficient drops to (μ=0.15), and the on-board ATP instruction requires the speed to be reduced to 30 km / h (about 8.3 m / s) within 1200 meters.

[0120] 1. Dynamic response and data processing

[0121] Environmental data triggers adjustments: The environmental sensing unit detects changes in rail surface humidity and a decrease in the adhesion coefficient, immediately transmitting the data to the system. The system then determines that it will enter complex environment braking mode.

[0122] Dynamic braking force distribution: The adhesion adjustment phase starts. Taking a carriage with a load of 60 tons (Wj = 60 tons) as an example, according to the formula:

[0123]

[0124] (where (λ=0.5), ) tons, (μ nominal =0.35)), calculate

[0125] Achieve precise distribution of braking force.

[0126] Anti-skid control intervention: When the slip rate of a certain wheelset is detected to reach 6%, exceeding the threshold of 5%, the anti-skid control function immediately triggers the graded decompression strategy, giving priority to reducing the braking force of the car, and through three steps of decompression (reducing 20kPa each time), the slip rate is controlled within a safe range.

[0127] 2. Braking distance correction and safety assurance

[0128] Recalculate the stage duration: The dynamic braking stage division module recalculates the stage duration according to the new adhesion coefficient and extends the adhesion adjustment stage to

[0129] (due to the decrease in adhesion coefficient, the adjustment time is extended), the stable phase is extended to

[0130] Braking distance recalculation: The braking distance calculation module recalculates the braking distance based on the updated parameters to ensure that the train stops safely within 1,200 meters. At the same time, the carbon emission optimization module synchronously adjusts the braking curve to reduce carbon emissions while ensuring safety.

[0131] Example 3: Self-healing in a fault scenario

[0132] When the train was traveling at a speed of 60 km / h (about 16.7 m / s), the wheel-rail adhesion monitoring unit suddenly experienced a sensor failure and the on-board ATP communication was interrupted.

[0133] 1. Fault detection and switching

[0134] Wheel-Rail Adhesion Monitoring Unit (WRMU) Failure Handling: The system uses the sensor's heartbeat detection mechanism to detect a WRMU failure within 200ms and automatically switches to the IMU-based adhesion estimation algorithm. The federated learning engine dynamically corrects the IMU algorithm's error compensation within 30 seconds, using real-time data from other normally operating trains, keeping the adhesion coefficient estimation error within ±5%.

[0135] Response to ATP communication interruption: After the on-board ATP communication is interrupted, the edge computing node starts the local braking decision-making function within 100ms, and calculates the braking strategy using a preset emergency braking model based on local data such as the train's current speed, load, and brake cylinder pressure.

[0136] 2. Security and remote collaboration

[0137] Local braking execution: The edge computing node sends instructions to the train braking system based on the calculation results to control the braking force output, and at the same time stores the braking status data (such as speed changes and braking force) locally in a cycle of 500ms.

[0138] Remote Communication and Support: The system transmits fault information (including wheel-rail adhesion monitoring unit failures and ATP communication interruptions) and current operating status data (speed, position, braking parameters, etc.) to the ground control center via the onboard communication module. After receiving the data, the ground control center uses an expert system to diagnose the fault and feeds back the optimized braking strategy to the train, assisting the edge computing node in adjusting braking decisions to ensure safe train operation.

[0139] Directions:

[0140] 1. System startup and initialization

[0141] Device installation and activation

[0142] Multimodal sensing modules are deployed on the train bogie and locomotive to ensure that the detection range of the laser Doppler velocimeter, strain gauge sensor, temperature and humidity sensor, infrared camera and millimeter-wave radar covers the wheel-rail contact surface and the front track.

[0143] The device is powered by the vehicle power supply, the wireless ad hoc network communication module is started (delay ≤ 20ms), and a data link is established with the vehicle ATP system and brake control unit (BCU).

[0144] Parameter calibration

[0145] Initialize the federated learning engine and load the pre-trained wheel-rail adhesion prediction model and braking phase division parameters (default k1 to k4 coefficients).

[0146] Input train formation information (number of carriages, load distribution), brake cylinder pressure threshold, and carbon emission optimization target weight.

[0147] 2. Real-time monitoring and data collection

[0148] Multimodal sensing operation

[0149] Wheel-rail adhesion monitoring: A laser Doppler velocimeter collects wheelset rotational speed in real time, and a strain gauge sensor calculates the friction coefficient (μ). Combined with the slip rate (≤5% threshold), the wheel-rail adhesion status is output.

[0150] Environmental perception: Infrared cameras detect foreign objects and frost on the rail surface, and temperature and humidity sensors simultaneously upload rail surface humidity; millimeter-wave radar continuously scans the distance and movement speed of obstacles ahead.

[0151] Train status collection: Real-time acquisition of onboard ATP instructions (target speed, speed limit), brake cylinder pressure, car load distribution and train length.

[0152] Data preprocessing

[0153] Abnormal data (such as sensor noise and communication packet loss) is filtered and compensated through edge computing nodes to ensure the data quality index (DQI ≥ 0.8) input into the federated learning engine.

[0154] 3. Dynamic Update of Federated Learning Model

[0155] Local model training

[0156] The lightweight LSTM network of each train receives local wheel-rail adhesion data, environmental data, and historical braking curves to calculate the adjustment of the stage coefficients k1 to k4.

[0157] The adhesion sensitivity coefficient λ and acceleration sensitivity coefficient δ are dynamically optimized in combination with the longitudinal acceleration (az).

[0158] Global aggregation and privacy protection

[0159] Every 10 minutes, the local model parameters are uploaded to the cloud through the differential privacy mechanism (noise injection), and the global aggregation unit generates a new model by weighted average according to the DQI weight (default α = 0.4, β = 0.3, γ = 0.3).

[0160] The updated braking phase division parameters (t1~t4) and carbon emission optimization strategy are synchronized to the entire fleet.

[0161] 4. Braking process control and optimization

[0162] Dynamic braking phase execution

[0163] Response phase (0-t1): The braking signal is triggered according to the ATP command, and the on-board BCU activates the brake cylinder pressure.

[0164] Propagation phase (t1-t2): Dynamically distribute the braking force according to the difference in vehicle load (formula: Pj = P0 × (Wj / W)^λ × (1 + δaz)) to prevent wheel locking.

[0165] Adhesion adjustment phase (t2-t3): If the slip ratio is greater than 5%, the graded decompression strategy is triggered, prioritizing the reduction of braking force on the car with the highest slip ratio.

[0166] Stable phase (t3-t4): Regenerative braking energy recovery is enabled, and the braking force is smoothed by the attenuation coefficient γ{stable}.

[0167] Braking distance calculation and warning

[0168] The four-stage integration model calculates the total braking distance Stotal in real time, combined with the millimeter-wave radar obstacle detection results:

[0169] If the obstacle distance is ≤1.2×Stotal, an audible and visual warning will be triggered;

[0170] If the distance is ≤0.8×Stotal, the emergency brake (EB) command is triggered.

[0171] Carbon emissions optimization

[0172] The NSGA-II algorithm generates a Pareto optimal braking curve under the constraint of safe braking distance, and displays carbon emissions (C = 0.85 × Pavg × tbrake / ηregen) in real time on the vehicle screen.

[0173] 5. Fault Self-Recovery and Emergency Response

[0174] Sensor failure response

[0175] If the wheel-rail adhesion monitoring unit fails, it automatically switches to the IMU inertial measurement unit to estimate the adhesion coefficient, and the federated learning engine compensates for the error (accuracy ≥ 90%).

[0176] When the on-board ATP communication is interrupted, the edge node executes the braking decision based on local historical data (synchronization accuracy ≥ 95%).

[0177] Fault reporting and maintenance

[0178] After the system switches to backup mode, it immediately sends the fault code, current speed and braking status to the ground control center through the 4G / 5G module.

[0179] After remote diagnosis, maintenance personnel can push repair instructions or lock the faulty module.

[0180] 6. Human-computer interaction and operation

[0181] Driver interface operation

[0182] The on-board display shows real-time:

[0183] Progress bar for the braking phase (response / propagation / adhesion adjustment / stabilization);

[0184] Comparison of the real-time adhesion coefficient μcurrent with the nominal value μnominal;

[0185] Dynamic carbon emission curve and energy-saving suggestions (such as "extending the taxiing distance can reduce carbon emissions by 10%").

[0186] The driver can manually adjust the braking strategy priority (safety / energy saving / balance mode).

[0187] Ground monitoring

[0188] The ground control center receives the federated learning model update logs, fault reports, and carbon emission statistics of the entire fleet for macro-scheduling optimization.

[0189] 7. Maintenance and Upgrade

[0190] Regular calibration

[0191] The laser Doppler velocimeter and millimeter-wave radar are calibrated every month to ensure that the detection accuracy error is ≤2%.

[0192] The federated learning benchmark model is updated quarterly to adapt to seasonal wheel-rail adhesion changes (such as winter frost patterns).

[0193] Software upgrade

[0194] Remotely update the braking algorithm through OTA (such as NSGA-III replacing NSGA-II) and optimize the DQI weight parameters.

[0195] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A device for determining and measuring the braking distance of a freight train, characterized in that: include: Multimodal sensing module: Integrates the following units: Wheel-rail adhesion monitoring unit: Based on a laser Doppler velocimeter and strain gauge sensors, it collects the friction coefficient and slip rate of the wheel-rail contact surface in real time; Environmental perception unit: detects rail surface humidity, foreign matter, and frost coverage through temperature and humidity sensors and infrared cameras; Train status acquisition unit: collects train formation length, load distribution, brake cylinder pressure and on-board ATP command signals. Federated Learning Engine: Build a distributed braking model training network to aggregate braking behavior data from multiple trains and dynamically update the wheel-rail adhesion prediction model and braking phase division parameters; A differential privacy mechanism is used to protect train operation data, and the model update cycle is ≤10 minutes. Dynamic braking stage division module: The braking process is divided into the response phase (0-t1), the propagation phase (t1-t2), the adhesion adjustment phase (t2-t3) and the stable phase (t3-t4); The duration thresholds (t1 to t4) of each stage are dynamically adjusted based on the change rate of the adhesion coefficient. The calculation formula is: (i＝1,2,3,4) Among them, μcurrent is the real-time stickiness coefficient, μnominal is the nominal stickiness coefficient, and ki is the stage coefficient of federated learning optimization; Braking distance calculation module: The total braking distance is calculated using a four-stage integration model: Among them, ηprop is the braking propagation efficiency factor, γstable is the attenuation coefficient in the stable stage; The multimodal sensing module supports wireless ad hoc network communication, and the data transmission delay is ≤20ms.

2. A device for determining and measuring the braking distance of a freight train according to claim 1, characterized in that: The federated learning engine includes: Local model training unit: A lightweight LSTM network is deployed locally on each train. The input is wheel-rail adhesion data, environmental data, and historical braking curves, and the output is the adjustment of the stage coefficient ki. Global aggregation unit: A weighted average algorithm is used to fuse the parameters of each train model. The weight is determined by the data quality index (DQI). The DQI calculation formula is: DQI = α·Data integrity + β·Sensor accuracy level + γ·Braking success rate Among them, α+β+γ=1, default values ​​α=0.4, β=0.3, γ=0.

3.

3. The device for determining and measuring the braking distance of a freight train according to claim 1, characterized in that: The adhesion adjustment phase includes the following functions: Dynamic braking force distribution: According to the load difference of each car in the train, the brake cylinder pressure is adjusted proportionally. The distribution formula is: Where Pj is the braking pressure of the jth carriage, Wj is its load, W is the average load, λ is the adhesion sensitivity coefficient (default is 0.5), az is the longitudinal acceleration, and δ is the acceleration sensitivity coefficient (default is 0.1); Anti-skid control: When the wheelset slip rate is detected to be greater than 5%, the graded decompression strategy is triggered, prioritizing the reduction of braking force on the car with the highest slip rate.

4. The device for determining and measuring the braking distance of a freight train according to claim 1, characterized in that: The device integrates a carbon emission optimization module, and its functions include: Braking energy consumption model: Calculate carbon emissions per unit braking distance C emit =f(P avg ,t brake ,or rec )=(1-th rec )·P avg ·t brake Among them, Pavg is the average braking power, tbrake is the braking time, η rec is the regenerative braking energy recovery efficiency factor; Multi-objective decision-making: Under the constraint of safe braking distance, the NSGA-II algorithm is used to optimize the braking curve to minimize carbon emissions.

5. The device for determining and measuring the braking distance of a freight train according to claim 4, characterized in that: The device supports fault self-healing function: When the wheel-rail adhesion monitoring unit fails, it automatically switches to the adhesion estimation algorithm based on the IMU (inertial measurement unit), and the error compensation amount is dynamically corrected by the federated learning engine; When the on-board ATP communication is interrupted, the edge computing node is enabled to execute local braking decisions with a synchronization accuracy of ≥95%.

6. The device for determining and measuring the braking distance of a freight train according to claim 1, characterized in that: The environmental perception unit also includes a millimeter-wave radar for detecting the distance and movement speed of obstacles in front of the train in real time. When an obstacle is detected, it triggers an early warning or emergency braking command based on the braking distance calculation result.

7. The device for determining and measuring the braking distance of a freight train according to claim 1, characterized in that: The onboard ATP command signal collected by the train status acquisition unit includes target speed and speed limit information. The braking distance calculation module dynamically adjusts the braking strategy according to the ATP command signal to ensure that the train reaches the target speed within the specified distance.

8. The device for determining and measuring the braking distance of a freight train according to claim 3, characterized in that: The dynamic braking force distribution function is also provided with a braking force upper limit threshold. When the calculated braking pressure of a certain carriage exceeds the upper limit threshold, the braking pressure of all carriages is reduced proportionally to prevent the wheels from locking due to excessive braking force.

9. The device for determining and measuring the braking distance of a freight train according to claim 5, characterized in that: The carbon emission optimization module also includes a real-time carbon emissions display interface, which shows the driver the real-time carbon emissions and cumulative carbon emissions during the current braking process through the on-board display screen. In the fault self-healing function, when the system detects a fault and switches to a backup algorithm or mode, the fault information and current operating status data are sent to the ground control center through the on-board communication module to facilitate remote diagnosis and maintenance by ground personnel.

Images