High-temperature detection and cooling method and system for brake pad of truck on long downhill section

Abstract

The invention discloses a long downhill section truck brake pad high temperature detection and cooling method and system, and the method comprises the steps: collecting the surface temperature data of a truck brake pad in real time according to high-precision infrared temperature sensor arrays and thermal imagers disposed at the two sides of a long downhill section, and generating a time-space correlation temperature field distribution diagram; based on the temperature field distribution diagram, dynamically adjusting a temperature safety threshold curve through reinforcement learning, and outputting a dynamic early warning trigger signal; according to the dynamic early warning trigger signal, a corresponding spraying instruction is generated so as to drive a piezoelectric ceramic microvalve array of a roadside spraying device to conduct spraying cooling on the brake pad; and according to the temperature attenuation rate and infrared thermal image feedback data of the brake pad after spraying, the temperature safety threshold curve and the spraying instruction are corrected through a Bayesian optimization algorithm, and closed-loop safety regulation and control are formed. According to the embodiment of the invention, the temperature of the brake pad can be monitored in real time, accurate spraying cooling control is realized, and safe and stable braking performance of a truck is ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of temperature detection, and in particular to a method and system for detecting and cooling high-temperature brake pads of trucks on a long downhill section. Background Art

[0002] With the rapid development of the road transport industry, trucks are increasingly traveling on long downhill sections. Due to the unique terrain and driving characteristics of long downhill sections, trucks generate a significant increase in brake heat during braking, leading to a rapid increase in brake pad temperature. Excessively high brake pad temperatures not only degrade brake system performance and potentially cause brake failure, but can also lead to accidents and pose a serious threat to traffic safety. Therefore, monitoring and timely cooling of high-temperature brake pads on trucks traveling on long downhill sections is particularly important.

[0003] Existing brake temperature monitoring methods often rely on single temperature sensors or traditional temperature measurement equipment. This approach often struggles to obtain comprehensive, real-time information on brake pad temperature, especially in complex downhill driving conditions. Furthermore, the lack of comprehensive consideration of factors such as vehicle speed, load, and slope significantly limits the accuracy and effectiveness of this traditional monitoring method. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for high-temperature detection and cooling of truck brake pads on long downhill sections, so as to address the shortcomings of the existing technology, monitor the temperature of the brake pads in real time, realize precise spray cooling control, and ensure that the truck maintains safe and stable braking performance during the downhill process.

[0005] One embodiment of the present application provides a method for detecting and cooling high-temperature brake pads of trucks on a long downhill slope, the method comprising: High-precision infrared temperature sensor arrays and thermal imagers deployed on both sides of the long downhill section collect real-time surface temperature data of the truck's brake pads. Combined with the vehicle's onboard optical borehole detection (OBD) system to obtain vehicle speed, load, and slope angle parameters, a multimodal data fusion algorithm is used to generate a spatiotemporally correlated temperature field distribution map. This multimodal data fusion algorithm uses an adaptive Kalman filter to eliminate environmental noise and compensates for temperature measurement errors based on the vehicle's motion trajectory. Based on the temperature field distribution map, a deep convolutional neural network is used to identify the brake pad contour. The target area is locked in combination with a spatiotemporal attention mechanism. Based on real-time meteorological data and a historical brake failure case library, the temperature safety threshold curve is dynamically adjusted through reinforcement learning to output a dynamic warning trigger signal. Based on the dynamic warning trigger signal, a fluid mechanics simulation model is called to predict the water flow coverage range through vortex flow field simulation. The spray duration is adjusted based on the heat capacity characteristics of the brake pad material, and a corresponding spray instruction is generated to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pad; Based on the temperature decay rate of the brake pad after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and spray instructions are corrected through the Bayesian optimization algorithm. At the same time, the high temperature event data is uploaded to the traffic management platform, triggering the roadside LED warning screen and the on-board terminal to synchronize warnings, forming a closed-loop safety control.

[0006] Optionally, the high-precision infrared temperature sensor array and thermal imager deployed on both sides of the long downhill section collects surface temperature data of the truck brake pads in real time, and simultaneously obtains vehicle speed, load, and slope angle parameters in combination with the on-board OBD system, and generates a spatiotemporally correlated temperature field distribution map through a multimodal data fusion algorithm, wherein the multimodal data fusion algorithm uses an adaptive Kalman filter to eliminate environmental noise and compensates for temperature measurement errors based on the vehicle motion trajectory, including: Based on the raw temperature data collected by a high-precision infrared temperature sensor array and a thermal imager, a nonlinear phase synchronization algorithm is used to dynamically compensate for multi-sensor clock bias and generate a temporally and spatially aligned temperature data stream. The synchronization algorithm uses millimeter-wave radar signals reflected from the edge of the vehicle's wheel hub as a time reference, achieving sub-millisecond synchronization accuracy. The vehicle speed, load, and slope angle parameters acquired by the onboard OBD system are input into the multi-physics field coupling model. A real-time calculation function for the heat generation power of the brake pad is constructed based on the vehicle dynamics equation. This function is then jointly calibrated with the spatiotemporally aligned temperature data stream to generate a temperature-mechanical correlation feature matrix. The temperature-mechanical correlation characteristic matrix is processed to eliminate environmental interference. An adaptive Kalman filter algorithm is used to model rain and fog scattering noise as a time-varying Gaussian mixture distribution. The noise covariance matrix is iteratively updated through expectation maximization. Based on the continuous Bezier curve fitting results of the vehicle motion trajectory, the sensor measurement position deviation is dynamically compensated to output environmentally corrected brake pad temperature field distribution data. The corrected brake pad temperature field distribution data is input into a spatiotemporal correlation encoder. The spatiotemporal evolution pattern of the brake pad surface temperature gradient is extracted through tensor decomposition technology, and the vehicle load and slope angle parameters are integrated to generate a three-dimensional temperature field distribution map. The encoder adopts a dynamic weight allocation strategy to perform nonlinear enhancement mapping on the high-temperature area.

[0007] Optionally, based on the temperature field distribution map, a deep convolutional neural network is used to identify the brake pad contour, a spatiotemporal attention mechanism is combined to lock the target area, and according to real-time meteorological data and a historical brake failure case library, a temperature safety threshold curve is dynamically adjusted through reinforcement learning to output a dynamic warning trigger signal, including: Based on the three-dimensional temperature field distribution map, a multi-scale feature fusion module of a deep convolutional neural network is used to extract the gradient mutation features of the brake pad edge. Geometric constraints are established in combination with the thermal expansion coefficient of the brake pad material to generate a pixel-level segmentation mask of the brake pad contour. The segmentation mask is input into the spatiotemporal attention mechanism. Based on the historical temperature change rate of the brake pad and the real-time wind speed data, the temperature sensitivity weight of each area is calculated. The gated recurrent unit is used to predict the heat diffusion path in the next few seconds and target high-risk target areas. The system uses a historical brake failure case library and transfer learning technology to map the temperature-failure time series in the case into equivalent feature vectors under the current vehicle load and slope angle conditions. Combined with humidity and air density parameters from real-time meteorological data, the Q-learning algorithm in reinforcement learning dynamically adjusts the slope and intercept of the temperature safety threshold curve. The dynamically adjusted temperature safety threshold curve is compared frame by frame with the real-time temperature field distribution. When the temperature in the target area continuously crosses the threshold and the rate of change exceeds the preset critical value, a graded warning signal is triggered. Among them, the first-level warning triggers the vibration prompt of the vehicle terminal, and the second-level warning simultaneously activates the roadside LED warning screen.

[0008] Optionally, the method of calling a fluid mechanics simulation model according to the dynamic warning trigger signal, predicting the water flow coverage range through vortex flow field simulation, adjusting the spraying duration based on the heat capacity characteristics of the brake pad material, and generating a corresponding spraying instruction to drive the piezoelectric ceramic microvalve array of the roadside spraying device to spray and cool the brake pad, including: Based on the dynamic warning trigger signal, a fluid dynamics simulation engine based on the Lattice Boltzmann method is invoked to perform vortex flow field topological decomposition on the water flow of the roadside sprinkler. The three-dimensional coverage of the water flow under the influence of air resistance and gravity is predicted by combining the truck's driving speed and the spatial coordinates of the brake pads. Based on the heat capacity characteristics and phase change latent heat parameters of the brake pad material, a heat conduction-convection coupling equation was constructed. The temperature decay curves for different spraying times were solved through finite element discretization, and a greedy algorithm was used to select the spraying time that met the cooling rate requirements. The three-dimensional coverage area and the spray duration are input into the digital twin control module, which generates high-frequency switching instructions through pulse width modulation technology of the piezoelectric ceramic microvalve. At the same time, the resonant frequency of the ultrasonic atomizer is adjusted to control the water mist particle size, so that the nanoscale water film evenly covers the brake pad surface; LiDAR is used to monitor the truck's position in real time. When the truck enters the spraying area, the nozzle deflection angle is dynamically corrected according to the wheel speed. Through distributed collaborative control of the microvalve array, sub-millisecond synchronization of the water jet direction and the vehicle's motion trajectory is achieved.

[0009] Optionally, the temperature safety threshold curve and spraying instructions are corrected using a Bayesian optimization algorithm based on the temperature decay rate of the brake pad after spraying and the infrared thermal imaging feedback data. At the same time, the high temperature event data is uploaded to the traffic management platform, triggering the roadside LED warning screen and the vehicle terminal to issue a synchronous warning, forming a closed-loop safety control, including: Based on the infrared thermal imaging feedback data of the brake pad after spraying, the temperature decay rate curve is extracted, and a thermal relaxation model is constructed based on fractional differential equations to quantify the deviation between the cooling effect and the theoretical prediction value; The deviation is input into the Bayesian optimization algorithm, and a nonlinear mapping relationship between the temperature safety threshold curve parameters and the spray instruction set is established through Gaussian process regression. The Pareto optimal solution set is generated by combining Monte Carlo sampling to correct the temperature safety threshold and the spray instruction. The vehicle ID, temperature peak, and spray response time in the high-temperature event data are feature-encoded, compressed using a lightweight encryption algorithm, and uploaded to the traffic management platform. The platform verifies the data integrity through blockchain technology and triggers an upgrade in the flashing frequency of the roadside LED warning screen. It also synchronizes warnings with the on-board terminal through the 5G-V2X communication link between the on-board terminal and the roadside equipment, achieving closed-loop safety regulation of the road section.

[0010] Optionally, the vehicle speed, load, and slope angle parameters acquired by the onboard OBD system are input into a multi-physics field coupling model, a real-time calculation function for the heat generation power of the brake pad is constructed based on the vehicle dynamics equation, and a joint calibration is performed with the time-space aligned temperature data stream to generate a temperature-mechanical correlation feature matrix, including: Based on vehicle speed, load, and slope angle parameters acquired by the on-board OBD system, a nonlinear vehicle dynamics model is used to construct a differential equation for the brake pad force. The implicit Runge-Kutta algorithm is used to dynamically solve the brake pad friction torque, generating a high-frequency mechanical characteristic sequence containing transient friction work and slip rate. The nonlinear vehicle dynamics model introduces a random roughness parameter of the tire-road contact surface as a disturbance term. The high-frequency mechanical characteristic sequence is input into the thermal-mechanical coupling field analysis module. Based on the thermal conductivity coefficient of the brake pad material and the micromorphology data of the friction interface, the non-Fourier heat conduction equation is solved by fractional order discretization to generate the spatiotemporal distribution function of the heat flux density on the brake pad surface. The solution process uses adaptive mesh encryption technology to capture the nonlinear mutation of the heat flux boundary layer. The spatiotemporal distribution function is subjected to multi-scale tensor decomposition to extract the coupling characteristics of the brake pad surface temperature gradient and friction torque. Simultaneously, the spatiotemporally aligned infrared temperature data streams are fused, and a temperature-mechanical calibration mapping relationship is constructed through a generative adversarial network to generate a physically consistent three-dimensional joint feature tensor. The three-dimensional joint feature tensor is input into the evolutionary multi-objective optimization module constrained by physical information. The Pareto frontier search space is constructed based on the thermodynamic entropy increase principle and the friction power dissipation equation. The NSGA-III algorithm is used to simultaneously optimize the three indicators of temperature gradient distribution uniformity, friction work equivalent error and computational efficiency. The population diversity is dynamically adjusted through the adaptive mutation operator, and finally the temperature-mechanical correlation feature matrix that satisfies the multi-physical field coupling constraints is output.

[0011] Another embodiment of the present application provides a high-temperature detection and cooling system for brake pads of trucks on long downhill sections, the system comprising: An acquisition module collects real-time surface temperature data of truck brake pads using high-precision infrared temperature sensor arrays and thermal imagers deployed on both sides of the long downhill section. It also uses the onboard OBD system to obtain vehicle speed, load, and slope angle parameters. A multimodal data fusion algorithm is used to generate a spatiotemporally correlated temperature field distribution map. The algorithm uses an adaptive Kalman filter to eliminate ambient noise and compensates for temperature measurement errors based on the vehicle's motion trajectory. A locking module is configured to identify the brake pad contour based on the temperature field distribution map using a deep convolutional neural network, lock the target area using a spatiotemporal attention mechanism, dynamically adjust the temperature safety threshold curve through reinforcement learning based on real-time meteorological data and a historical brake failure case library, and output a dynamic warning trigger signal; a generation module for invoking a fluid mechanics simulation model based on the dynamic warning trigger signal, predicting the water flow coverage range through vortex flow field simulation, adjusting the spray duration based on the heat capacity characteristics of the brake pad material, and generating a corresponding spray instruction to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pad; The correction module is used to correct the temperature safety threshold curve and spraying instructions based on the temperature decay rate of the brake pad after spraying and the infrared thermal image feedback data through the Bayesian optimization algorithm. At the same time, the high temperature event data is uploaded to the traffic management platform, triggering the roadside LED warning screen and the on-board terminal to issue synchronous warnings, forming a closed-loop safety control.

[0012] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when run.

[0013] Yet another embodiment of the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the above methods.

[0014] Compared with the existing technology, the present invention provides a method for high-temperature detection and cooling of truck brake pads on long downhill sections. According to the high-precision infrared temperature sensor array and thermal imager deployed on both sides of the long downhill section, the surface temperature data of the truck brake pads are collected in real time to generate a time-space correlated temperature field distribution map; based on the temperature field distribution map, the temperature safety threshold curve is dynamically adjusted through reinforcement learning, and a dynamic early warning trigger signal is output; according to the dynamic early warning trigger signal, a corresponding spray instruction is generated to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pads; according to the temperature decay rate of the brake pads after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and the spray instruction are corrected through the Bayesian optimization algorithm to form a closed-loop safety regulation, so that the temperature of the brake pads can be monitored in real time, and precise spray cooling control can be achieved, ensuring that the truck maintains safe and stable braking performance during the downhill process. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A hardware block diagram of a computer terminal for a method for detecting and cooling high-temperature brake pads on trucks on long downhill sections provided by an embodiment of the present invention; Figure 2 A schematic flow chart of a method for detecting and cooling high-temperature brake pads on a truck on a long downhill slope provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of a high-temperature detection and cooling system for truck brake pads on a long downhill slope provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention.

[0017] The embodiment of the present invention first provides a method for detecting and cooling high temperature of truck brake pads on a long downhill section. The method can be applied to electronic devices such as computer terminals, specifically ordinary computers.

[0018] The following describes it in detail by taking running on a computer terminal as an example. Figure 1The hardware structure block diagram of a computer terminal for a method for detecting and cooling high temperature of brake pads of trucks on a long downhill slope provided by an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0019] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions that, when executed, enable a processor to execute any one of the methods for detecting and cooling high-temperature brake pads of trucks on a long downhill slope.

[0020] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0021] The internal memory provides an environment for the operation of computer programs in non-volatile storage media. When the computer program is executed by the processor, the processor can execute any method for high-temperature detection and cooling of truck brake pads on a long downhill section.

[0022] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0023] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0024] See also Figure 2 The embodiment of the present invention provides a method for detecting and cooling high temperature of brake pads of trucks on a long downhill section, which may include the following steps: S201: Using high-precision infrared temperature sensor arrays and thermal imagers deployed on both sides of the long downhill section, real-time surface temperature data of the truck's brake pads is collected. Vehicle speed, load, and slope angle parameters are acquired through the onboard OBD system. A spatiotemporal temperature field distribution map is generated using a multimodal data fusion algorithm. The multimodal data fusion algorithm employs an adaptive Kalman filter to eliminate ambient noise and compensates for temperature measurement errors based on the vehicle's motion trajectory. By deploying an array of infrared temperature sensors and thermal imagers, the system captures the brake pad surface temperature distribution in real time. Combined with vehicle dynamic parameters (such as speed, load, and slope) from the onboard OBD (on-board diagnostic tool), an adaptive Kalman filter is employed to perform spatiotemporal alignment and noise suppression on the multi-source data. Specifically, sensor data is synchronized using timestamps and a spatial interpolation algorithm to eliminate measurement position offsets caused by vehicle motion (such as changes in viewing angle due to brake pad rotation), and the temperature data is mapped into a unified three-dimensional coordinate system. Furthermore, the Kalman filter's dynamic covariance matrix update mechanism effectively filters out environmental noise such as rain, fog, and airflow (e.g., interference from infrared radiation scattering caused by raindrops), ultimately generating a highly accurate temperature field distribution map. This overcomes the temperature measurement inaccuracies associated with traditional single sensors due to environmental interference and dynamic measurement errors. Multimodal data fusion enables real-time, high-precision modeling of the brake pad temperature field, providing a reliable data foundation for subsequent early warning and spray control, and avoiding the risk of brake failure due to false or missed detections.

[0025] Specifically, based on the raw temperature data collected by a high-precision infrared temperature sensor array and a thermal imager, a nonlinear phase synchronization algorithm can be used to dynamically compensate for multi-sensor clock deviations to generate a temporally and spatially aligned temperature data stream. The synchronization algorithm uses the millimeter-wave radar signal reflected from the edge of the vehicle's wheel hub as a time reference to achieve sub-millisecond synchronization accuracy. To address the issue of multi-sensor clock asynchrony (e.g., a 0.5ms delay between the infrared array and the thermal imager), millimeter-wave radar is used to capture the rotation angle signal from the wheel hub edge (e.g., 12 tooth slots trigger 12 pulses per revolution) and use this as the global time reference. Nonlinear phase synchronization algorithms (e.g., phase-locked loop-based clock correction) dynamically calculate the clock offset of each sensor, and cubic spline interpolation is used to align data timestamps, achieving a time synchronization error of ≤0.1ms and a spatial alignment error of ≤2mm between sensors. This eliminates the temporal and spatial misalignment of temperature data caused by sensor asynchrony, ensuring the accuracy of subsequent temperature field modeling (e.g., increasing spatial resolution to 5mm), providing high-precision input for spray positioning.

[0026] Monitoring truck brake pad temperature on long downhill sections requires the coordinated operation of multiple sensors. Because infrared sensors and thermal imagers are distributed along both sides of the road, hardware clock differences (such as crystal oscillator frequency drift) and signal transmission delays (such as varying fiber lengths) can lead to temporal misalignment (typically 1-10 milliseconds) and spatial offset (such as the position of the brake pad within the sensor's field of view) in the collected data. To achieve precise data alignment, the physical motion of the vehicle's wheels is used as a unified time reference, dynamically compensating for clock offsets through a nonlinear phase synchronization algorithm.

[0027] Millimeter-wave radar signal capture and time reference generation: Millimeter-wave radars operating at a frequency of 77 GHz (gigahertz) are deployed on both sides of the road. This frequency band has high resolution (wavelength of approximately 3.9 mm) and can accurately capture the geometric features of the edges of truck wheel hubs (such as bolt holes and heat dissipation holes).

[0028] As the wheel rotates, the uneven structure on its edge reflects millimeter-wave signals, forming a periodic pulse sequence. For example, a 12-hole wheel generates a reflected pulse every 30 degrees of rotation (360 degrees / 12 holes). Considering the wheel circumference (e.g., 2 meters) and the vehicle speed (e.g., 80 km / h, or 22.22 meters / second), the pulse interval can be calculated as (2 meters / 22.22 meters / second) × (30 degrees / 360 degrees) = 0.075 seconds, corresponding to a wheel speed of approximately 800 revolutions per minute (RPM).

[0029] The radar signal processing unit extracts the pulse arrival time (TOA) and generates a global timestamp sequence, which serves as the time reference for multi-sensor synchronization with an accuracy of up to 0.1 millisecond (i.e., sub-millisecond level).

[0030] Dynamic compensation of multi-sensor clock deviation: Clock drift modeling: Sensor clock deviation is primarily caused by temperature changes. For example, the frequency of an infrared sensor's crystal oscillator drifts by 50 ppm (parts per million) when the ambient temperature rises by 10°C, resulting in a cumulative deviation of 50 microseconds per second. By using temperature sensors to monitor the temperature of each node in real time, a crystal oscillator frequency-temperature curve is established to predict clock offset.

[0031] Nonlinear phase synchronization algorithm: This algorithm uses an improved phase-locked loop (PLL) structure and introduces nonlinear feedback control (such as a fuzzy PID controller) to dynamically adjust the virtual offset of the sensor sampling clock. For example, if a sensor data timestamp is detected to be 2 milliseconds delayed from the millimeter-wave reference, the algorithm gradually shortens the sampling interval of subsequent data packets until the offset is zero.

[0032] Data resampling and alignment: The raw data streams from each sensor are interpolated and resampled according to the corrected time base. For example, if a sensor's clock drift causes missing data at a certain moment, cubic spline interpolation is used to supplement the temperature value at the current moment (e.g., 305°C) using the temperature values at the previous and next moments (e.g., 300°C in the previous frame and 310°C in the next frame).

[0033] Space-time coordinate mapping: Based on the truck's real-time speed (derived from the onboard OBD system) and the sensor spacing (e.g., 5 meters between adjacent sensors), the system calculates the time it takes for the brake pad to enter each sensor's field of view. For example, when the vehicle is traveling at 72 km / h (20 m / s), it takes 0.5 seconds for the brake pad to move from sensor A to sensor B (10 meters apart). Therefore, the system activates data collection from sensor B 0.5 seconds in advance.

[0034] Multi-sensor data is mapped to a unified coordinate system through spatial interpolation (such as bilinear interpolation). For example, the center of a brake pad has coordinates (x1, y1) in the field of view of sensor A and (x2, y2) in the field of view of sensor B. The system fits its actual spatial position (x, y) based on the vehicle's motion trajectory and then weightedly fuses the temperature data from the two sensors.

[0035] The vehicle speed, load, and slope angle parameters acquired by the onboard OBD system are input into the multi-physics field coupling model. A real-time calculation function for the heat generation power of the brake pad is constructed based on the vehicle dynamics equation. This function is then jointly calibrated with the spatiotemporally aligned temperature data stream to generate a temperature-mechanical correlation feature matrix. Based on vehicle dynamics equations and combined with slope angle correction for normal force, the heat generation rate of the brake pad is calculated in real time. By combining calibrated temperature data (such as surface temperature measured by infrared) with the theoretical heat generation rate, a temperature-mechanical mapping relationship (such as the linear proportionality coefficient between temperature gradient and friction work) is constructed, forming a multidimensional feature matrix that includes temperature, friction work, and slip rate. This fusion of physical models and measured data overcomes the limitations of traditional single data sources, improves the reliability of the analysis of the correlation between temperature field and mechanical state (e.g., calibration error ≤ 5%), and provides a multidimensional decision-making basis for dynamic early warning.

[0036] The heat generated by brake pads is directly related to the vehicle's dynamic parameters. It is necessary to integrate the vehicle's OBD data (speed, load, slope angle) with the infrared temperature data to build a multi-physics field coupling model to accurately calculate the heat generation power.

[0037] Vehicle dynamics modeling and mechanical parameter calculation: Equivalent mass calculation: Based on the slope angle θ (e.g., an 8% slope corresponds to arctan(0.08) = 4.57 degrees) and the vehicle's gross mass m_vehicle (e.g., 30 tons), calculate the equivalent mass m_eq supported by the brake pads. This formula incorporates the tire-road friction coefficient μ (μ≈0.7 for dry asphalt roads). The equivalent mass m_eq is calculated as the vehicle mass multiplied by (sinθ + μ·cosθ). For example, for a 30-ton vehicle on an 8% slope, m_eq = 30,000 kg × (sin4.57° + 0.7 × cos4.57°) = 30,000 kg × 0.113 ≈ 3,390 kg.

[0038] Friction torque calculation: Based on the brake hydraulic pressure (OBD parameter, such as 10 MPa) and the effective friction radius of the brake pad (e.g., 0.2 m), calculate friction torque T = pressure × area × friction coefficient × radius. Assuming a single-side brake pad area of 0.05 square meters and a friction coefficient of 0.4, T = 10 × 10^6 Pa × 0.05 m² × 0.4 × 0.2 m = 40,000 Newton meters (Nm).

[0039] Calculation of heat generation power: The instantaneous heat power P_heat is equal to the friction torque T multiplied by the wheel angular velocity ω (ω = vehicle speed / tire radius. For example, 20 m / s corresponds to an angular velocity of 40 rad / s for a wheel with a radius of 0.5 m). This means P_heat = T × ω = 40,000 Nm × 40 rad / s = 1.6 megawatts (MW). This power is converted into heat flux density on the brake pad surface.

[0040] Multi-physics coupling and data calibration: Thermal-mechanical coupled field simulation: Finite element analysis (FEA) software (such as COMSOL) is used to build a three-dimensional model of the brake pad. The heat flux density is used as the boundary condition, combined with the material's thermal conductivity coefficient (such as 50 W / m·K for cast iron) to simulate the surface temperature distribution.

[0041] Joint calibration: Aligns the simulated temperature field with the measured infrared data. For example, if the simulation predicts a temperature of 400°C at a certain point after 5 seconds, while the measured value is 380°C, the system automatically adjusts the thermal conductivity to 55 W / m·K, bringing the simulation result closer to the measured value.

[0042] Characteristic Matrix Generation: The calibrated temperature field, heat flux, and mechanical parameters (pressure, slip) are organized into a three-dimensional tensor based on a spatiotemporal grid. For example, a 10×10 spatial grid (each grid is 1 cm²) is sampled 100 times within 10 seconds, forming a 10×10×100 three-dimensional matrix, where each element contains a temperature value (e.g., °C), a heat generation rate (e.g., W / m²), and a mechanical state label (e.g., pressure level).

[0043] The temperature-mechanical correlation characteristic matrix is processed to eliminate environmental interference. An adaptive Kalman filter algorithm is used to model rain and fog scattering noise as a time-varying Gaussian mixture distribution. The noise covariance matrix is iteratively updated through expectation maximization. Based on the continuous Bezier curve fitting results of the vehicle motion trajectory, the sensor measurement position deviation is dynamically compensated to output environmentally corrected brake pad temperature field distribution data. Environmental interference (such as rain, fog, and dust) can cause infrared temperature measurement accuracy to decrease (typical error ±20°C), requiring signal processing and trajectory compensation to improve data reliability.

[0044] Rain and fog noise modeling and filtering: Gaussian Mixture Model (GMM) construction: Rain and fog noise is classified into three categories: light rain (attenuation coefficient 0.1), moderate rain (0.3), and heavy rain (0.5). Each category corresponds to a Gaussian distribution with mean μ and variance σ². For example, in light rain mode, the noise mean is -5°C (infrared signals are attenuated, resulting in low temperature readings), and the variance σ² = 2.

[0045] Adaptive Kalman Filter: In the prediction-update loop, the expectation-maximization (EM) algorithm is used to iteratively estimate the current noise category. The filter's state variable is the actual temperature, and the observed variable is the noisy temperature. If the measured temperature falls below the theoretical value by more than 5°C for 10 consecutive frames, the system enters light rain mode and switches the noise covariance matrix to Σ = diag(2, 2).

[0046] Vehicle trajectory fitting and position compensation: Bezier curve fitting: A cubic Bezier curve is fitted using the wheel hub positions detected by the millimeter-wave radar as control points. For example, four control points P0(x0,y0), P1(x1,y1), P2(x2,y2), and P3(x3,y3) define a smooth trajectory. When the parameter t∈[0,1], the curve coordinates B(t) = (1-t)^3·P0 + 3t(1-t)^2·P1 + 3t^2(1-t)·P2 + t^3·P3.

[0047] Dynamic compensation: The system predicts the brake pad's position within the sensor's field of view based on the fitted trajectory. For example, if the vehicle deviates 0.5 meters to the right, the system controls the infrared sensor gimbal to rotate 0.1 degrees to the left. (Assuming the sensor's field of view is 30 degrees and the detection distance is 10 meters, 0.1 degrees corresponds to a lateral displacement of approximately 1.7 centimeters.)

[0048] Temperature field correction and output: The filtered temperature data is combined with the trajectory compensation results. For example, the original temperature measurement at a point is 250°C, which is corrected to 270°C after noise filtering. Then, trajectory compensation is added (due to sensor aiming deviation, the temperature measurement is 5°C lower), and the final output is the actual temperature of 275°C.

[0049] The output data format is a spatiotemporal grid temperature field. Each grid point contains a timestamp (such as Unix time 1625097600.123), spatial coordinates (x=1.2m, y=0.8m), temperature value (275℃) and confidence level (such as 0.95).

[0050] The corrected brake pad temperature field distribution data is input into a spatiotemporal correlation encoder. The spatiotemporal evolution pattern of the brake pad surface temperature gradient is extracted through tensor decomposition technology, and the vehicle load and slope angle parameters are integrated to generate a three-dimensional temperature field distribution map. The encoder adopts a dynamic weight allocation strategy to perform nonlinear enhancement mapping on the high-temperature area.

[0051] The core task of the spatiotemporal correlation encoder is to fuse multidimensional data into an intuitive three-dimensional temperature field distribution map to support subsequent early warning decisions.

[0052] Tensor decomposition and feature extraction: Tucker decomposition: Decomposes the temperature-mechanics correlation matrix (dimensions time × space × physical parameters) into a core tensor and three factor matrices (time factor, space factor, and physical parameter factor). For example, the core tensor size is 3 × 3 × 2, representing three time modes, three space modes, and two physical parameter modes.

[0053] Feature explanation: The time factor may reveal the rapid rise stage of the brake pad temperature in the initial braking period (such as a heating rate of 10°C / s in the 0-5 seconds) and the subsequent saturation stage (the rate drops to 2°C / s in the 5-10 seconds). The spatial factor shows that the temperature at the outer edge is higher than that in the center (the gradient difference is about 50°C).

[0054] Dynamic weight allocation strategy: High-temperature area enhancement: For areas with temperatures exceeding 300°C, the weight coefficient is increased from 1.0 to 2.0, and the color mapping intensity is enhanced during encoding (for example, from red to dark red).

[0055] Gradient smoothing: A bilateral filter is used to smooth the temperature field. The filter parameters have a spatial standard deviation of σ_s = 2 pixels and a grayscale standard deviation of σ_r = 10°C. This filter suppresses random noise while preserving sharp edges (such as sudden temperature changes caused by brake pad cracks).

[0056] Three-dimensional temperature field generation: A three-dimensional temperature distribution map is constructed by adding vehicle load (e.g., 30 tons) and slope angle (e.g., 8%) as additional dimensions. For example, the X-axis represents the lateral position of the brake pad (0-0.5 meters), the Y-axis represents the longitudinal position (0-0.3 meters), and the Z-axis represents the combined load-slope parameter (e.g., 30 tons - 8% corresponds to Z=1, and 25 tons - 6% corresponds to Z=2).

[0057] The visualization output supports multi-view switching, such as the top view showing the lateral temperature gradient and the side view revealing the heat conduction characteristics in the thickness direction.

[0058] For example, a long downhill section of a mountain highway (5 kilometers long, with an average gradient of 7%) experienced frequent brake pad overheating incidents due to heavy trucks' frequent braking. The highway administration deployed high-precision infrared temperature sensor arrays (100 sets spaced 50 meters apart) and thermal imagers (one every 200 meters) on both sides of the road, integrating data from the vehicle's on-board diagnostics (OBD) system. The following describes the complete implementation process of these steps, using a 35-ton six-axle truck as an example.

[0059] Step 1: Multi-sensor clock synchronization and spatiotemporal alignment Event trigger: The truck drove into the starting point of a long downhill slope at 65 km / h (about 18 m / s), triggering the roadside millimeter-wave radar to start monitoring.

[0060] Millimeter wave radar signal capture: A 77GHz millimeter-wave radar (wavelength approximately 3.9mm, resolution 0.1m) deployed on the roadside scans the edge of a truck's wheel hub. The wheel hub has a 12-hole structure, a diameter of 1m, and a circumference of 3.14m.

[0061] The radar detects a reflected pulse every 30 degrees of wheel rotation (corresponding to one bolt hole). The wheel rotation period is calculated based on the vehicle speed: wheel circumference 3.14 meters, vehicle speed 18 m / s → rotation period = 3.14 m / 18 m / s ≈ 0.174 seconds (corresponding to a speed of approximately 345 rpm).

[0062] Generate a global timestamp sequence: When each bolt hole passes through the radar field of view, record a timestamp accurate to 0.1 millisecond (e.g. t1=1625097600.000s, t2=1625097600.174s, …).

[0063] Infrared sensor clock synchronization: When the truck passes infrared sensor No. 3 (150 meters from the starting point), the sensor's local clock accumulates a 2.1 millisecond deviation due to temperature drift.

[0064] Nonlinear phase synchronization algorithm: Establish a clock drift model: Based on the internal temperature sensor data of the sensor (currently 35°C), the crystal oscillator frequency drift is 60 ppm (parts per million), accumulating 60 microseconds of deviation per second.

[0065] Dynamic compensation: A fuzzy PID controller is used to adjust the sampling interval, spreading the 2.1 millisecond deviation over the next 10 data packets, shortening the sampling interval by 0.21 milliseconds per packet.

[0066] Data resampling: Cubic spline interpolation is performed on the original data stream, and the timestamp is corrected to align with the millimeter wave reference. For example, the original data recorded a temperature of 320°C at t=1625097600.002s. After correction, it is mapped to t=1625097600.000s, and the interpolated temperature is 318°C.

[0067] Thermal imager spatial alignment: The thermal imager (30-degree field of view, 640×480 pixel resolution) detected the truck's right front wheel brake pad at image coordinates (x=300, y=200).

[0068] Based on the vehicle's motion trajectory (Bezier curve fitting), the brake pad position is predicted to be (x=320, y=190) 0.5 seconds later. The gimbal is controlled to rotate 0.2 degrees to track the target.

[0069] Results: The data from 10 groups of infrared sensors and two thermal imagers were synchronized at the sub-millisecond level (deviation < 0.5ms), and the spatial alignment error was < 3 cm.

[0070] Step 2: Multi-physics coupling model construction and joint calibration Data processing: The vehicle's OBD system uploads data in real time: speed 65km / h, load 35 tons, slope 7%, brake hydraulic pressure 12MPa.

[0071] Vehicle dynamics calculations: Calculation of equivalent mass: slope 7% (corresponding to inclination angle arctan(0.07)=4°), friction coefficient μ=0.6 (wet asphalt road surface), equivalent mass m_eq=35000kg×(sin4°+0.6×cos4°)=35000×0.6×0.9976≈20895kg.

[0072] Friction torque calculation: The effective radius of the brake pad on one side is 0.25 meters, the friction area is 0.08 square meters, and the friction coefficient is 0.35 → Torque T = 12×10^6Pa×0.08m²×0.35×0.25m=84,000Nm.

[0073] Heat generation power: wheel angular velocity ω = vehicle speed / tire radius = 18m / s ÷ 0.5m = 36rad / s → P_heat = 84000Nm × 36rad / s = 3.024MW.

[0074] Thermal-mechanical coupling simulation and calibration: Finite element modeling: The brake pad is made of carbon-ceramic composite material (thermal conductivity 40 W / m·K), and a three-dimensional mesh model (10,000 elements) is established.

[0075] Heat flux mapping: Distribute 3.024 MW of power to a friction area of 0.08 m² → average heat flux of 37.8 W / mm².

[0076] Combined calibration: Simulations show the brake pad's outer edge reaches 450°C after 5 seconds, but infrared measurements show 420°C. The system automatically adjusts the thermal conductivity to 45 W / m·K to match the simulated temperature curve with the measured data.

[0077] Output: Generates a temperature-mechanical correlation feature matrix, which contains temperature values (unit: °C), heat flux density (W / mm²), and pressure distribution labels for a 100×100 spatial grid.

[0078] Step 3: Environmental interference elimination and trajectory compensation Interference scenario: Sudden moderate rain (rainfall intensity 15 mm / h) causes infrared temperature measurement values to be attenuated by rain and fog scattering. Simultaneously, the truck shifts 1.2 meters to the left due to the slippery road surface.

[0079] Adaptive Kalman Filter: Noise modeling: The rain and fog attenuation coefficient is 0.3, and the noise distribution is a Gaussian mixture model (mean -10°C, variance σ² = 4).

[0080] Filtering process: Prediction stage: Based on the thermodynamic model, the temperature of the next frame is predicted to rise by 12°C (e.g., current 400°C → predicted 412°C).

[0081] Update phase: The infrared measurement value is affected by rain and fog and shows 390°C. The calculated Kalman gain K = 0.7 → Correction value = 412 × 0.3 + (390 + 10) × 0.7 = 401°C.

[0082] Bezier curve trajectory compensation: The millimeter-wave radar detected that the truck was 1.2 meters to the left, and the cubic Bezier curve control points were fitted: P0 (150m, 0m), P1 (160m, -0.5m), P2 (170m, -1.0m), and P3 (180m, -1.2m).

[0083] Predict the brake pad position to be (x=175m, y=-1.1m) in 0.5 seconds. Control the infrared sensor gimbal to rotate 0.25 degrees to the right to correct the aiming point.

[0084] Results: After filtering, the temperature data error was reduced from ±20°C to ±5°C, and trajectory compensation made the spatial alignment error <5 cm.

[0085] Step 4: Spatiotemporal correlation coding and 3D temperature field generation Data fusion: Input the corrected temperature field data (401°C), load 35 tons, and slope 7% into the encoder.

[0086] Tensor decomposition: Use Tucker decomposition to decompose the 100×100×100 feature matrix into a core tensor (5×5×3) and a factor matrix: The time factor shows that the heating rate is 15℃ / s in the first 5 seconds and drops to 3℃ / s in the last 5 seconds.

[0087] The spatial factor shows that the temperature at the outer edge is 80℃ higher than that at the center.

[0088] The physical parameter factors show that for every 5 tons increase in load, the peak temperature rises by 25°C.

[0089] Dynamic weight allocation: Regions with temperatures > 350°C are assigned a weight of 2.0 and are highlighted in dark red in the 3D plot.

[0090] A bilateral filter (σ_s = 2 pixels, σ_r = 15°C) is used to smooth the temperature gradient and retain the local high temperature points (e.g., 430°C) caused by the brake pad cracks.

[0091] 3D visualization output: X-axis: lateral position of the brake pad (0-0.5 meters), Y-axis: longitudinal position (0-0.3 meters), Z-axis: load-slope combination parameter (35 tons - 7% corresponds to Z=3).

[0092] It shows that the maximum temperature is 425℃ at the outer edge and 345℃ in the center area. The thermal gradient distribution is consistent with the braking friction characteristics.

[0093] Specifically, in another implementation, the vehicle speed, load, and slope angle parameters obtained by the on-board OBD system are input into the multi-physics field coupling model. A real-time calculation function for the heat generation power of the brake pad is constructed based on the vehicle dynamics equation. The function is then jointly calibrated with the time-space aligned temperature data stream to generate a temperature-mechanical correlation feature matrix, including: Based on vehicle speed, load, and slope angle parameters acquired by the on-board OBD system, a nonlinear vehicle dynamics model is used to construct a differential equation for the brake pad force. The implicit Runge-Kutta algorithm is used to dynamically solve the brake pad friction torque, generating a high-frequency mechanical characteristic sequence containing transient friction work and slip rate. The nonlinear vehicle dynamics model introduces a random roughness parameter of the tire-road contact surface as a disturbance term. A nonlinear dynamic equation is established, introducing road roughness as a random perturbation. The implicit Runge-Kutta method (0.1ms step size) is used to solve for friction torque (e.g., peak torque 1200 N·m). This outputs a friction work sequence (e.g., instantaneous power P = 25 kW) sampled at 1 kHz. This accurately characterizes the mechanical state of the brake pad (with a solution error of ≤2%), providing high-fidelity input data for thermal-mechanical coupling analysis.

[0094] First, the onboard OBD system collects vehicle speed (in km / h), load (in tons), and slope angle (in degrees) in real time via the CAN bus. These parameters are input into a nonlinear vehicle dynamics model based on the Newton-Euler equations, accounting for the vehicle's gravity component, air resistance, and tire friction during a long downhill slope. For example, when a truck is traveling at 80 km / h on a 5% slope and carrying a 30-ton load, the model calculates the equivalent moment of inertia acting on the brake pads.

[0095] To accurately capture the friction torque of the brake pad, the model incorporates a random roughness parameter (e.g., using ISO 8608 standard road roughness grade B) of the tire-road interface. This roughness parameter simulates the microscopic undulations of the actual road surface through a random process generator and acts as a perturbation term in the dynamic equations. For example, when a truck travels on a dilapidated asphalt road, the roughness parameter introduces high-frequency vibrations, leading to transient fluctuations in the brake pad's friction torque.

[0096] The differential equations are numerically solved using the implicit Runge-Kutta (IRK) algorithm. IRK uses fourth-order accuracy (RK4) to ensure computational stability, with a time step of 10 milliseconds, synchronized with the OBD system sampling rate. The algorithm iteratively solves the implicit equation for friction torque within each time step, outputting the transient friction work (in joules) and slip ratio (dimensionless, ranging from 0 to 1). For example, in an emergency braking scenario, the slip ratio might suddenly increase from 0.2 to 0.8, triggering ABS intervention.

[0097] The resulting high-frequency mechanical feature sequence contains 100 data points per second, covering key indicators such as peak friction power and slip gradient. These data are aligned via timestamps and provide input for subsequent thermodynamic analysis.

[0098] The high-frequency mechanical characteristic sequence is input into the thermal-mechanical coupling field analysis module. Based on the thermal conductivity coefficient of the brake pad material and the micromorphology data of the friction interface, the non-Fourier heat conduction equation is solved by fractional order discretization to generate the spatiotemporal distribution function of the heat flux density on the brake pad surface. The solution process uses adaptive mesh encryption technology to capture the nonlinear mutation of the heat flux boundary layer. Construct a non-Fourier heat conduction equation, use an adaptive grid (minimum size 0.01mm) to capture the heat flux boundary layer (thickness 0.2mm), and solve the heat flux density distribution (such as peak value q"=5×10 6 W / m²). Revealing the microscopic heat flux mutation characteristics (resolution up to 10μm), improving the physical consistency of temperature field prediction.

[0099] The core task of the thermal-mechanical coupled field analysis module is to establish a mapping relationship between the brake pad surface temperature field and the friction work. First, the thermal conductivity of the brake pad material (for example, 50 W / (m·K) for a cast iron brake pad) and the microscopic topography of the friction interface (roughness profiles acquired using a laser confocal microscope with a resolution of 1 μm) are input.

[0100] A non-Fourier heat conduction equation (Cattaneo-Vernotte model) replaces the traditional Fourier law to capture the thermal wave effects of transient heat transfer in brake pads. The equation is numerically solved using a fractional discretization (e.g., using Caputo fractional derivatives with an order of α = 0.8), with a time step consistent with the mechanical characteristic sequence (10 milliseconds). For example, when the brake pad temperature rises sharply, the fractional-order model more accurately reflects the delayed nature of heat flow propagation.

[0101] Adaptive Mesh Refinement (AMR) technology dynamically optimizes computing resource allocation. The initial mesh is uniformly divided (cell size 2 mm). In areas where temperature gradients exceed a threshold (e.g., 100°C / mm), the mesh is automatically refined to 0.5 mm. For example, at the edge of a brake pad, where the friction contact area is small and the temperature gradient is high, the mesh is refined to improve computational accuracy. Furthermore, a quadtree data structure is used to manage the mesh hierarchy, ensuring the efficiency of the refinement process.

[0102] The resulting spatiotemporal distribution function is stored as a three-dimensional matrix with the dimensions time × spatial coordinate × temperature / heat flux. For example, a typical output matrix might contain 1000 time points (corresponding to 10 seconds of data), a 100×100 spatial grid, and the corresponding heat flux values (unit: W / m²).

[0103] The spatiotemporal distribution function is subjected to multi-scale tensor decomposition to extract the coupling characteristics of the brake pad surface temperature gradient and friction torque. Simultaneously, the spatiotemporally aligned infrared temperature data streams are fused, and a temperature-mechanical calibration mapping relationship is constructed through a generative adversarial network to generate a physically consistent three-dimensional joint feature tensor. Tucker tensor decomposition (core tensor rank = 5) was used to extract multi-scale features. A Generative Adversarial Network (GAN) network (5 layers each of generator and discriminator) was used to calibrate infrared temperature with simulation data. The resulting output was a 3D feature tensor (256×256×32) containing physical quantities such as temperature gradient and friction work. This achieved physically consistent fusion of multi-source data (calibration error ≤ 3%), providing highly reliable input for the optimization module.

[0104] Multi-scale tensor decomposition uses the Tucker decomposition algorithm to break down the spatiotemporal distribution function into a core tensor and factor matrices. For example, a 1000×100×100 tensor might be decomposed into a 10×10×10 core tensor and three corresponding factor matrices. Wavelet transforms are used during the decomposition process to extract multi-scale features: low-scale components reflect the overall temperature rise trend of the brake pad, while high-scale components capture local hotspots.

[0105] When fusing infrared temperature data streams, a spatiotemporal alignment strategy is employed. For example, the infrared camera collects data at a 30 Hz frequency and uses an interpolation algorithm to align it with the 10 ms timestamps of the mechanical feature sequence. Spatial alignment is achieved by calibrating the pixel coordinates of the brake pad in the thermal image (for example, using ArUco markers to assist in positioning).

[0106] The generator of a generative adversarial network (GAN) uses a U-Net architecture, taking as input decomposed tensor features and infrared temperature data, and outputting a calibrated three-dimensional temperature field. The discriminator, a convolutional neural network (CNN), determines whether the generated results conform to physical constraints (such as energy conservation). Training uses the Wasserstein distance as the loss function, combined with physical regularization terms (such as enforcing a linear correlation between heat flux and frictional work). For example, if infrared data detects an abnormally high temperature in a certain area but the mechanical model does not predict the corresponding frictional work, the GAN will correct the temperature distribution in that area to align with physical laws.

[0107] The resulting joint feature tensor contains three channels: temperature gradient (°C / mm), friction torque (Nm), and heat flux (W / m²). For example, the center of a brake pad might have both high friction torque (200 Nm) and high heat flux (1e5 W / m²), while the edge might have a steep temperature gradient (150°C / mm) and low torque.

[0108] The three-dimensional joint feature tensor is input into the evolutionary multi-objective optimization module constrained by physical information. The Pareto frontier search space is constructed based on the thermodynamic entropy increase principle and the friction power dissipation equation. The NSGA-III algorithm is used to simultaneously optimize the three indicators of temperature gradient distribution uniformity, friction work equivalent error and computational efficiency. The population diversity is dynamically adjusted through the adaptive mutation operator, and finally the temperature-mechanical correlation feature matrix that satisfies the multi-physical field coupling constraints is output.

[0109] Optimization objectives were defined as follows: temperature gradient uniformity (standard deviation σ ≤ 15°C / mm), friction work error (≤ 5%), and computation time (≤ 50ms). The NSGA-III algorithm (population size 100) searches for a Pareto optimal solution through adaptive mutation (dynamically adjusted mutation rate between 0.1 and 0.3), outputting a feature matrix with a compression ratio of ≥ 80%. This achieves multi-objective collaborative optimization (improving computational efficiency by 60%), ensuring the feature matrix achieves both high precision and real-time performance, supporting real-time safety decision-making.

[0110] The design goal of the evolutionary multi-objective optimization module is to balance model accuracy and computational resource consumption. First, based on the second law of thermodynamics, an entropy increase constraint is defined: the rate of change of the brake pad's total entropy must be greater than or equal to the entropy generated by frictional work (unit: J / (K·s)). Furthermore, the frictional power dissipation equation requires that at least 95% of the input mechanical energy be converted into heat.

[0111] The Pareto front search space consists of three dimensions: Temperature gradient distribution uniformity: quantified by calculating the standard deviation (σ) of the temperature gradient on the brake pad surface, with the goal of minimizing σ; Friction work equivalent error: The root mean square error (RMSE) between the friction work predicted by the mechanical model and the infrared temperature inversion value. The goal is to minimize the RMSE. Computational efficiency: The time taken for a single iteration (unit: ms). The goal is to minimize this time.

[0112] Optimization was performed using the NSGA-III (third-generation non-dominated sorting genetic algorithm) with a population size of 500, a crossover probability of 0.9, and an initial mutation probability of 0.1. An adaptive mutation operator dynamically adjusts the mutation rate based on population diversity metrics (such as Hamming distance). As the population becomes more homogeneous, the mutation rate is increased to 0.3 to enhance exploration capabilities. For example, in the early stages of optimization, the algorithm may quickly converge to a low σ region, then escape the local optimum by increasing the mutation rate.

[0113] The final output feature matrix undergoes Pareto optimality screening to ensure the optimal combination of the three metrics. For example, an optimal solution might correspond to a combination of σ = 15°C / mm, RMSE = 2.3%, and a computational time of 8 ms, meeting both accuracy and real-time requirements.

[0114] S202, based on the temperature field distribution map, using a deep convolutional neural network to identify the brake pad contour, combining a spatiotemporal attention mechanism to lock the target area, and dynamically adjusting the temperature safety threshold curve through reinforcement learning based on real-time meteorological data and a historical brake failure case library, and outputting a dynamic warning trigger signal; A deep convolutional neural network (such as the ResNet-50 architecture) performs semantic segmentation on the temperature distribution map, extracting gradient features at the brake pad edge (e.g., areas with sudden temperature changes). It also uses differences in thermal expansion coefficients to distinguish the boundary between the brake pad and the wheel hub. A spatiotemporal attention mechanism dynamically assigns weights by analyzing the temperature change rate (e.g., areas with a temperature increase of more than 5°C per second) and the impact of wind speed on heat dissipation, pinpointing high-risk areas (e.g., localized hotspots). A reinforcement learning module uses a Q-learning algorithm to adjust the slope and intercept of the safety threshold (e.g., lowering the threshold by 10% on rainy days) based on historical case studies (e.g., temperature-time series from the past 1,000 brake failure events) and real-time meteorological data (e.g., humidity affects heat dissipation efficiency), generating adaptive warning signals. This overcomes the limitations of traditional fixed temperature thresholds by dynamically learning from the environment and historical data to intelligently optimize the warning threshold, significantly reducing the false alarm rate (e.g., from 15% to 2%) and accurately locating high-risk areas, avoiding false alarms or delayed responses caused by inappropriate global threshold settings.

[0115] Specifically, based on the three-dimensional temperature field distribution map, a multi-scale feature fusion module of a deep convolutional neural network can be used to extract the gradient mutation features of the brake pad edge, and geometric constraints can be constructed in combination with the thermal expansion coefficient of the brake pad material to generate a pixel-level segmentation mask of the brake pad contour. A deep convolutional network with multi-scale feature fusion (such as the ResNet-50 architecture) is used to extract the temperature gradient features of the brake pad edge (such as the area with a gradient value ≥ 5°C / mm) through the feature maps of different convolutional layers (such as 1×1, 3×3, and 5×5 convolution kernels). Combined with the thermal expansion coefficient of the brake pad material (such as the thermal expansion coefficient of cast iron α = 11×10 -6 / °C), constructing geometric constraints to eliminate contour deformation errors caused by thermal expansion and generating pixel-level segmentation masks (e.g., resolution 0.5mm / pixel). This achieves high-precision brake pad contour recognition (positioning error ≤ 2mm), avoids missegmentation caused by thermal deformation, provides a precise spatial reference for subsequent target area detection, and improves the reliability of high-temperature area detection.

[0116] The multi-scale feature fusion module of the deep convolutional neural network (DCNN) utilizes the U-Net++ architecture, with a ResNet-50 backbone network. It fuses feature maps from different layers through skip connections. The input 3D temperature field distribution map (512×512×3, with channels corresponding to temperature gradients, spatial coordinates, and material properties) first undergoes four layers of downsampling (stride 2, 3×3 convolution kernel), generating feature maps of sizes 256×256×64, 128×128×128, 64×64×256, and 32×32×512, respectively. During the upsampling phase, resolution is restored layer by layer through transposed convolution (4×4 convolution kernel). These features are then concatenated with the feature maps from the corresponding downsampling layers to form multi-scale fused features.

[0117] To address the sudden gradient changes at the edges of brake pads, the network introduces a dedicated edge detection branch in the decoder. This branch uses the Sobel operator combined with a learnable convolution kernel (5×5) to extract sharp gradient changes in the temperature field. For example, when the temperature gradient at the edge of the brake pad increases sharply due to high temperatures (e.g., a temperature difference of more than 200°C within a 10cm range from 300°C to 500°C), this branch enhances the response in such areas.

[0118] Combined with the geometric constraints of the thermal expansion coefficient of the brake pad material, a physical information embedding strategy is adopted. For example, the thermal expansion coefficient of the cast iron brake pad is 12×10 -6 When generating the segmentation mask, the thermal expansion deformation of each pixel in the temperature field (deformation = original size × temperature change × expansion coefficient) is calculated and input into the spatial transformer network (STN) as a spatial deformation field, dynamically adjusting the deformation tolerance of the segmentation boundary. Specifically, if the temperature of a region rises by 200°C, the corresponding lateral expansion is approximately 2.4mm. The network reserves this deformation margin in the segmentation mask to avoid contour misjudgment due to thermal expansion.

[0119] The resulting pixel-level segmentation mask (binary image, resolution 512×512) is post-processed with CRF (conditional random field) to optimize edge smoothness and is spatially aligned with the original image from the infrared thermal imager to ensure pixel-level segmentation accuracy (error < 2 pixels).

[0120] The segmentation mask is input into the spatiotemporal attention mechanism. Based on the historical temperature change rate of the brake pad and the real-time wind speed data, the temperature sensitivity weight of each area is calculated. The gated recurrent unit is used to predict the heat diffusion path in the next few seconds and target high-risk target areas. The spatiotemporal attention mechanism calculates temperature sensitivity weights (e.g., W = 0.9 for high-risk areas) by analyzing historical temperature change rates (e.g., areas with a heating rate ≥ 8°C / s over the past 10 seconds) and real-time wind speed (e.g., the convection heat dissipation coefficient h = 25 W / m²·K at a wind speed of 4 m / s). A gated recurrent unit (GRU) uses the heat conduction equation to predict heat diffusion paths over the next 5 seconds, targeting areas where temperatures may exceed the material tolerance threshold (e.g., the critical temperature of cast iron is 450°C). Dynamically focusing on high-risk areas (with a prediction accuracy of ≥ 95%) avoids computational redundancy in global monitoring, provides early warning of local overheating risks, and provides a spatially localized basis for tiered response.

[0121] The spatiotemporal attention mechanism consists of a spatial attention module (SAM) and a temporal attention module (TAM) in parallel. The spatial attention module receives segmentation masks and real-time temperature field data. It generates a spatial weight map through channel attention (SE Block) and spatial convolution (3×3), focusing on high-temperature areas of the brake pad (e.g., the weight of areas with temperatures >400°C is increased to 0.9, while the weight of low-temperature areas is reduced to 0.1). The temporal attention module processes the historical temperature change rate series (with a time window of 30 seconds and a sampling rate of 10Hz), using one-dimensional causal convolution (convolution kernel size 7) to extract mutation patterns in the temporal dimension (e.g., transient events with a temperature change rate exceeding 10°C / s).

[0122] Real-time wind speed data is acquired from roadside ultrasonic anemometers (sampling rate 1kHz), filtered through a sliding average, and then fed into a temperature sensitivity weight calculation unit. This unit uses a multi-layer perceptron (MLP) with a 64-32-16 architecture to map the wind speed vector (including magnitude and direction) into a spatial weight correction coefficient. For example, when the wind speed is 8m / s and the direction is opposite to the vehicle's travel direction, the heat dissipation efficiency on the windward side of the brake pad increases, and the corresponding area's temperature sensitivity weight decreases by 20%.

[0123] The input of the Gated Recurrent Unit (GRU) is a temporally weighted temperature field sequence (5-second time step, 0.1-second step interval), with a hidden layer dimension of 256. The GRU controls information flow by resetting and updating gates to predict the heat diffusion path within the next three seconds. For example, if the temperature in the center of the brake pad is detected to be rising at a rate of 15°C / s, the GRU simulates the heat conduction process along the radial direction of the brake pad, superimposes the convective cooling effect caused by wind speed, and outputs a probability distribution map of the temperature field at the next moment (resolution 128×128).

[0124] High-risk target areas are identified based on a dynamic comparison of temperature field predictions with safety thresholds. The system uses two criteria: 1) the predicted temperature exceeds the material phase transition threshold (e.g., 650°C for cast iron brake pads); and 2) the temperature gradient points toward critical structures, such as the brake pad mounting bolts. Areas meeting either condition are marked as red high-risk zones. Adjacent regions are then merged using a morphological closing operation (kernel size 5×5) to generate the final bounding box for the target area.

[0125] The system uses a historical brake failure case library and transfer learning technology to map the temperature-failure time series in the case into equivalent feature vectors under the current vehicle load and slope angle conditions. Combined with humidity and air density parameters from real-time meteorological data, the Q-learning algorithm in reinforcement learning dynamically adjusts the slope and intercept of the temperature safety threshold curve. Historical temperature-failure sequences (e.g., a temperature of 380°C for 12 seconds causes failure) are extracted from the case library and mapped to equivalent features (e.g., equivalent time T_eq = 10 seconds) under current vehicle parameters (20-ton load, 8% slope) through transfer learning (e.g., feature domain adaptation). Incorporating real-time humidity (e.g., RH = 70% reduces heat dissipation efficiency) and air density (ρ = 1.2 kg / m³), the state-action value function is updated using Q-learning, dynamically adjusting the threshold curve parameters (e.g., slope from 0.8 to 0.6). This achieves adaptive threshold curve adaptation to operating conditions (adjustment error ≤ 3%), addressing the failure of traditional fixed thresholds in complex environments and improving warning accuracy to over 98%.

[0126] The historical brake failure case database contains over 100,000 records, each storing vehicle load (ranging from 0-50 tons), slope angle (0-15°), temperature-time series (sampling rate 1Hz), and failure type (such as brake pad cracking and brake fluid vaporization). Transfer learning utilizes a pre-trained Transformer encoder (6 layers, 8 heads) to map the temperature-time series from historical cases into 128-dimensional feature vectors, which are then aligned to current vehicle parameters through domain adaptation. For example, data from a historical case with a 30-ton load and a 10° slope is converted to an equivalent feature vector for a current 35-ton load and an 8° slope through linear scaling (load variance coefficient = current load / historical load) and nonlinear interpolation (slope angle influence factor).

[0127] Real-time meteorological data, including humidity (ranging from 0-100% RH) and air density (1.15-1.29 kg / m³), is collected from roadside weather stations and fed into the threshold adjustment model. Humidity affects heat dissipation efficiency (for every 10% increase in humidity, convective heat dissipation efficiency decreases by approximately 3%), while air density modifies the thermal conductivity (for every 0.1 kg / m³ increase in density, conductivity increases by 1.2%).

[0128] The state space of the Q-learning algorithm is defined as a four-tuple: the current maximum temperature (quantized to 50°C intervals, such as 300-350°C), the load (in 5-ton intervals), the slope angle (in 1° intervals), and the humidity (in 10% intervals). The action space consists of adjusting the threshold curve parameters: slope change (±0.1°C / s) and intercept change (±10°C). The reward function is designed as follows: a successful warning (triggering 5 seconds before brake failure) is rewarded with +100, a false alarm (failure to trigger due to the threshold being too high) is penalized with -50, and a late warning (triggering after failure) is penalized with -200. An ε-greedy strategy (ε = 0.1) is used to balance exploration and exploitation. The Q-table is updated with a learning rate α = 0.01 and a discount factor γ = 0.9.

[0129] The dynamically adjusted temperature safety threshold curve is output as a piecewise linear function. For example, under the conditions of a 40-ton load, a 12° slope, and 70% humidity, the threshold curve has an initial slope of 0.8°C / s and an intercept of 350°C. After Q-learning optimization, the slope is adjusted to 1.2°C / s and the intercept is reduced to 320°C, enabling earlier response to high-temperature risks.

[0130] The dynamically adjusted temperature safety threshold curve is compared frame by frame with the real-time temperature field distribution. When the temperature in the target area continuously crosses the threshold and the rate of change exceeds the preset critical value, a graded warning signal is triggered. Among them, the first-level warning triggers the vibration prompt of the vehicle terminal, and the second-level warning simultaneously activates the roadside LED warning screen.

[0131] By comparing the real-time temperature field with a dynamic threshold frame-by-frame (e.g., 30 frames per second), a Level 2 warning (activating a red flashing LED screen) is triggered when the target area temperature exceeds the threshold (e.g., T>380°C) for three consecutive frames with a rate of change ≥5°C / s. If only a single frame exceeds the threshold with a rate of change <3°C / s, a Level 1 warning (low-frequency vibration of the vehicle terminal) is triggered. Warning signals are synchronized in real time with the roadside unit (RSU) via the CAN bus. This implements a graded response mechanism (reducing the false alarm rate to below 1%), balancing warning sensitivity with false alarm risk, ensuring accurate alerts at varying risk levels.

[0132] Real-time temperature field data (frame rate 30Hz) is compared frame by frame with the dynamic threshold curve through a sliding window (window size 1 second, overlap rate 50%). The system sets two levels of trigger conditions: Level 1 Warning: The target area temperature exceeds the threshold curve for three consecutive frames (i.e., 0.1 seconds), with an instantaneous rate of change ≥ 5°C / s. Once triggered, the vehicle terminal sends a command to the steering wheel vibration module via the CAN bus (frequency 50Hz, amplitude 0.5mm), and a yellow warning icon is displayed on the instrument panel.

[0133] Level 2 Warning: If the temperature in the target area exceeds the threshold curve for a sustained period of 1 second, with an average rate of change ≥10°C / s, the system activates the roadside LED warning screen via the 5G-V2X broadcast protocol (latency <10ms), displaying the red text "Brake Overheat" and a recommended deceleration value (e.g., "Slow down to 30 km / h") and simultaneously sends a coordinated warning message (SAE J2735 standard message format) to the vehicle behind.

[0134] The warning triggering logic uses a finite state machine (FSM) to manage state transitions. For example, after a Level 1 warning is triggered, if the temperature falls below the threshold within 5 seconds, the state is reset. If it continues to exceed the threshold, it is upgraded to Level 2 after 1 second. All warning events are recorded in the black box memory, including the trigger time, temperature peak, vehicle status, and response action, for subsequent closed-loop optimization.

[0135] S203: Based on the dynamic warning trigger signal, a fluid dynamics simulation model is invoked to predict the water flow coverage range through vortex flow field simulation. The spray duration is adjusted based on the heat capacity characteristics of the brake pad material, and a corresponding spray instruction is generated to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pad; When the warning is triggered, the Lattice Boltzmann method is used to simulate the three-dimensional diffusion of the spray water flow under the influence of air resistance, gravity, and vehicle motion (e.g., matching the water flow with the relative velocity at a vehicle speed of 20 km / h). This method predicts the water mist coverage area (e.g., an elliptical area with a diameter of 1.2 meters). Furthermore, based on the brake pad material's heat capacity (e.g., cast iron's specific heat capacity is 460 J / kg·K) and phase change latent heat (e.g., water evaporation absorbs heat at 2260 kJ / kg), a finite element method is used to calculate the mapping between spray duration and temperature decay (e.g., spraying for 3 seconds can reduce the temperature by 80°C). The piezoelectric ceramic microvalve uses PWM pulses to control its opening (e.g., a 75% duty cycle corresponds to a water flow rate of 8 L / min), ensuring uniform water mist coverage and preventing secondary safety hazards (e.g., hydroplaning due to accumulated water). By combining physical simulation with material properties, precise control of spray parameters is achieved, avoiding the waste of resources or insufficient cooling caused by blind spraying in traditional spray systems. The system also ensures that the water flow is synchronized with the vehicle's motion trajectory, improving cooling efficiency (e.g., increasing cooling rate by 40%) and safety.

[0136] Specifically, based on the dynamic warning trigger signal, a fluid dynamics simulation engine based on the Lattice Boltzmann method can be invoked to perform vortex flow field topological decomposition on the water flow of the roadside sprinkler. The three-dimensional coverage of the water flow under the influence of air resistance and gravity can be predicted by combining the truck's driving speed and the spatial coordinates of the brake pads. The Lattice Boltzmann method (D3Q19 model) is used to simulate the interaction between water flow and air. Inputs include truck speed (e.g., 20 m / s), nozzle elevation angle (30°), and gravitational acceleration (g=9.8 m / s²). The velocity distribution (e.g., central velocity 8 m / s) and coverage (major axis 2.5 m, minor axis 1.2 m) of the vortex flow field are calculated. Topological decomposition is used to identify the primary vortex structure (e.g., regions with vorticity Ω ≥ 50 s⁻¹) and predict the diffusion trajectory of the water mist in the dynamic airflow. Millimeter-level predictions of water flow coverage (with an error of ≤ 5 cm) are achieved, ensuring precise coverage of high-temperature areas of the brake pads and avoiding water waste or insufficient coverage.

[0137] When a dynamic warning trigger signal is generated, the system first initiates a fluid dynamics simulation engine based on the Lattice Boltzmann method (LBM). LBM is a mesoscopic simulation method based on the collision and migration of microscopic particles, particularly suitable for simulating turbulent flows under complex boundary conditions. In this application, the simulation engine uses the D3Q19 discrete velocity model (3D space, 19 velocity directions) to model the water flow. First, the fluid domain is initialized based on the design parameters of the roadside sprinkler (such as a 2.5mm nozzle diameter and an initial water velocity of 8m / s), dividing the spray area into a voxel grid with a resolution of 0.5mm. Next, vortex flow field topological decomposition techniques (such as the vorticity-velocity potential decomposition method) are used to extract the characteristics of the water flow structure. For example, when a truck is traveling at 60km / h, the water flow forms a paraboloid inclined downstream due to air resistance, while at rest it exhibits a symmetrical diffusion pattern.

[0138] To accurately predict the water flow coverage, the system integrates the truck's speed and the brake pad's spatial coordinate data. The brake pad's spatial coordinates are obtained through a combination of onboard GPS and roadside lidar positioning, with an accuracy of ±5cm. In the simulation, the truck's speed is converted into a velocity field boundary condition relative to the water flow: when the vehicle travels at v = 15m / s, the equivalent initial velocity of the water flow in the vehicle coordinate system becomes v_water = v_nozzle - v_vehicle. Simultaneously, the acceleration of gravity (9.8m / s²) and air resistance (calculated using the k-ε turbulence model) are dynamically incorporated into the fluid equations. For example, on a downhill section with a 5% gradient, the water flow will accelerate toward the bottom of the slope due to gravity, requiring simulation to compensate for the spray angle.

[0139] The simulation engine ultimately outputs a three-dimensional coverage probability cloud map, where each voxel contains the probability of water flow reaching the brake pad (0-1) and the coverage time (in milliseconds). For example, within 200 milliseconds of the spray activation, the probability of water flow covering the center of the brake pad can reach 95%, while the edge areas may drop to 80% due to turbulent disturbances. This result serves as the physical basis for subsequent spraying decisions.

[0140] In the simulation, the truck's velocity, v_vehicle (in m / s), is converted into a velocity field boundary condition relative to the water flow: when the vehicle travels at v_vehicle = 15 m / s (54 km / h), the equivalent initial velocity of the water flow in the vehicle coordinate system becomes v_water = v_nozzle - v_vehicle. (For example, if the nozzle velocity, v_nozzle, is 8 m / s, then v_water = -7 m / s, which must be compensated for by tilting the nozzle forward.) Simultaneously, the acceleration of gravity (9.8 m / s²) and air resistance (calculated using the k-ε turbulence model) are dynamically added to the flow equations.

[0141] Real-time changes in v_vehicle trigger nozzle deflection control to ensure that the water flow tracks the moving brake pads; the choice of v_nozzle affects the spray duration decision (high-pressure spraying can shorten the time, but requires higher energy consumption).

[0142] Parameter adjustability: v_nozzle can be dynamically adjusted according to water source pressure (e.g., down to 5 m / s in drought areas), while v_vehicle relies entirely on OBD real-time data.

[0143] Based on the heat capacity characteristics and phase change latent heat parameters of the brake pad material, a heat conduction-convection coupling equation was constructed. The temperature decay curves for different spraying times were solved through finite element discretization, and a greedy algorithm was used to select the spraying time that met the cooling rate requirements. A heat conduction-convection equation was established, and the finite element method (0.5 mm mesh size) was used to solve the temperature decay curve corresponding to the spray duration Δt (e.g., 3 seconds, 5 seconds) (e.g., a temperature drop of ΔT = 120°C at Δt = 3 seconds). A greedy algorithm selected Δt = 3 seconds as the optimal solution (satisfying ΔT ≥ 100°C and water consumption ≤ 5L). This optimized spray duration (reducing energy consumption by 30%) minimized water consumption while maintaining cooling effectiveness, improving system sustainability.

[0144] The thermal response characteristics of brake pads are determined by their material parameters. For example, a semi-metallic brake pad has a specific heat capacity of C = 500 J / (kg·K) and a thermal conductivity of k = 50 W / (m·K). Ceramic brake pads, on the other hand, have a latent heat of phase change of L = 200 kJ / kg (due to amorphous transition at high temperatures). The system automatically loads parameters from the material database based on the brake pad model provided by the vehicle's OBD (on-board diagnostics) and constructs the coupled heat conduction-convection equation: Equation Construction: During the spraying process, heat dissipates through conduction (internal), convection (water contact), and phase change (heat absorption by the material). The governing equations include a Fourier heat conduction term, a Newtonian cooling term (convection coefficient h = 5000 W / (m²·K)), and a phase change source term (activated when the temperature exceeds 600°C).

[0145] Finite element discretization: The brake pads were meshed using hexahedral elements with a minimum element size of 0.1 mm to capture surface temperature gradients. A time step of Δt = 0.01 seconds was set, and the total simulation duration covered the expected spraying time (1 to 10 seconds). For example, at 3 seconds into the spraying process, the system calculated the temperature change at each node and calculated the overall cooling rate (e.g., from 800°C to 300°C).

[0146] Greedy algorithm optimization: With a cooling rate of ≥100°C / s as the constraint, the greedy algorithm iterates starting with the shortest spray duration (1s). Each iteration increases the duration by 0.5s and evaluates the cooling effect until the minimum duration that satisfies the condition is found. For example, if a 2.5s spray duration can reduce the temperature from 750°C to 400°C, this duration is selected to avoid wasting water.

[0147] The key to this step is balancing computational accuracy and real-time performance. Through GPU parallel acceleration (such as the NVIDIA CUDA architecture), the finite element solver can complete the simulation of a single spray duration in 50ms, meeting the needs of real-time decision-making.

[0148] The three-dimensional coverage area and the spray duration are input into the digital twin control module, which generates high-frequency switching instructions through pulse width modulation technology of the piezoelectric ceramic microvalve. At the same time, the resonant frequency of the ultrasonic atomizer is adjusted to control the water mist particle size, so that the nanoscale water film evenly covers the brake pad surface; Based on simulation results, the digital twin module generates control parameters: a piezoelectric microvalve PWM frequency of 1 kHz (a 60% duty cycle corresponds to a flow rate of 6 L / min), and an ultrasonic atomizer resonant frequency of 1.7 MHz (generating a 10 μm water mist). Through closed-loop feedback control (e.g., lidar monitoring of water mist distribution), these parameters are dynamically adjusted to achieve a uniform water film thickness of 50 ± 5 μm. This achieves uniform nanoscale water film coverage (≥95%), improving cooling efficiency (40% higher than traditional spraying) and preventing secondary safety hazards caused by localized water accumulation.

[0149] The digital twin control module is the "intelligent hub" of the sprinkler system. Its core task is to convert simulation results into executable physical operations: Piezoelectric ceramic microvalve control: The microvalve array uses PZT-5H piezoelectric material (response time 0.1ms) and uses pulse width modulation (PWM) technology to control the opening degree. For example, when the spray duration is set to 3 seconds, the module generates a 60% duty cycle PWM signal (frequency 1kHz), causing the microvalve to open for a cumulative 1.8 seconds within 3 seconds, accurately regulating the water flow to 8L / min.

[0150] Ultrasonic atomizer tuning: The atomizer's resonant frequency (typically 20-100kHz) determines the mist particle size (D50 = 10-50μm). The system dynamically adjusts the frequency based on the turbulence intensity within the three-dimensional coverage area: In high-turbulence areas (such as the brake pad edge), a high frequency (80kHz) is used to produce small-size mist (15μm) for enhanced penetration; in low-turbulence areas, a low frequency (30kHz) is used to generate large-size mist (40μm) for enhanced impingement cooling.

[0151] Nanoscale water film formation: Through the superposition effect of water mist, a continuous water film approximately 200nm thick forms on the brake pad surface. This process relies on the synergy between the microvalve and the atomizer: the rapid opening and closing of the microvalve (cycle time 1ms) enables precise control of water volume, while the frequency modulation of the atomizer ensures uniform water mist distribution (spatial coefficient of variation <5%).

[0152] For example, in a truck emergency braking scenario, if the system detects that the local temperature of the brake pad reaches 900°C, it will activate the "enhanced mode": the microvalve opening is increased to 80%, and the atomization frequency is increased to 120kHz (exceeding the conventional range), allowing ultra-fine water mist (8μm) to quickly penetrate the high-temperature area.

[0153] LiDAR is used to monitor the truck's position in real time. When the truck enters the spraying area, the nozzle deflection angle is dynamically corrected according to the wheel speed. Through distributed collaborative control of the microvalve array, sub-millisecond synchronization of the water jet direction and the vehicle's motion trajectory is achieved.

[0154] LiDAR (scanning frequency 100Hz) tracks the truck's position in real time (with a positioning accuracy of ±2cm). It calculates vehicle displacement (0.5m per second) based on wheel speed (e.g., 200rpm) and dynamically adjusts the nozzle deflection angle (e.g., 0.5° per second). The microvalve array is synchronized via timestamps (error ≤ 0.1ms), ensuring that the water flow direction deviates from the vehicle's trajectory by ≤5cm. This achieves high-precision spatiotemporal synchronization of dynamic spraying (synchronization error ≤ 1ms), ensuring full coverage and cooling of the brake pads on moving vehicles, adapting to complex traffic scenarios.

[0155] To achieve precise synchronization between the sprinkler system and the mobile truck, a real-time feedback control chain with multi-sensor fusion is required: LiDAR positioning: A 32-line LiDAR (100Hz scanning frequency) captures the truck's contours in real time and combines it with point cloud registration algorithms (such as ICP Iterative Closest Point) to calculate the vehicle's center coordinates with an accuracy of ±2cm. For example, when a truck enters a spray zone at 72km / h, the system updates its position data every 10ms and predicts its trajectory for the next 50ms.

[0156] Printhead deflection control: The printhead is equipped with a high-torque servo motor (5ms response time) that calculates the vehicle's lateral displacement based on wheel speed (obtained via OBD). For example, if the left wheel speed is higher than the right wheel speed (indicating a right turn), the printhead must deflect to the left by Δθ = arctan(v_y / v_x) to compensate for the deviation, where v_x is the longitudinal velocity and v_y is the lateral velocity.

[0157] Microvalve coordinated control: The distributed microvalve array uses CAN bus communication (transmission latency <0.1ms), with each microvalve receiving independent control commands. For example, if a truck drifts to the right, the right microvalve opens 5ms earlier and increases its opening, creating a right-sloping water curtain.

[0158] Ultimately, the system uses spatiotemporal synchronization algorithms (such as clock synchronization based on the PTP protocol) to achieve sub-millisecond matching of the spray direction and vehicle trajectory. Tests have shown that at a speed of 100 km / h, the deviation between the center of the water flow and the actual position of the brake pad can be controlled to within 3 cm, ensuring maximum cooling efficiency.

[0159] S204, based on the temperature decay rate of the brake pad after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and spraying instructions are corrected through the Bayesian optimization algorithm. At the same time, the high temperature event data is uploaded to the traffic management platform, triggering the roadside LED warning screen and the on-board terminal to synchronize warnings, forming a closed-loop safety control.

[0160] After spraying, the temperature decay curve is monitored in real time using an infrared thermal imager. Bayesian optimization is used to update the safety threshold parameters with posterior probabilities. Gaussian process regression is used to build a model linking the parameters with the cooling effect, generating a Pareto optimal solution (e.g., balancing cooling rate with water consumption). High-temperature event data (such as vehicle ID, peak temperature, and spray response time) is encrypted with AES-256 and uploaded to the platform, where data integrity is ensured through blockchain hash verification. Roadside LED screens adjust their flashing frequency based on the warning level (e.g., 1Hz yellow flash for a level 1 warning, 2Hz red flash for a level 2 warning), and on-board terminals receive commands synchronously via 5G-V2X. This enables dynamic self-optimization of system parameters and multi-party collaborative warning, forming a closed loop of "perception-decision-execution-feedback," improving safety management capabilities on long downhill sections (e.g., reducing accident rates by 60%). Furthermore, data encryption and blockchain technology ensure information security and meet the compliance requirements of the traffic management platform.

[0161] Specifically, the temperature decay rate curve can be extracted based on the infrared thermal image feedback data of the brake pad after spraying, and a thermal relaxation model can be constructed based on the fractional-order differential equation to quantify the deviation between the cooling effect and the theoretical prediction value. Post-spray temperature data was collected using an infrared thermal imager and fitted to a temperature decay curve. The root mean square deviation (RMSE) between the measured curve and the theoretical prediction was calculated (e.g., RMSE = 8°C) to quantify the degree of deviation. This revealed the nonlinear characteristics of the actual cooling process (goodness of fit R² ≥ 0.98), providing a quantitative basis for parameter optimization and improving model prediction accuracy.

[0162] After spray cooling, a high-precision infrared thermal imager collects brake pad surface temperature distribution data at a sampling rate of 10 frames per second. A spatiotemporal registration algorithm is used to eliminate image jitter caused by vehicle vibration. The temperature decay rate curve is extracted using a sliding window difference method with a window width of 3 seconds (corresponding to 30 frames of data) and a step size of 1 second. The average temperature change rate of pixels within the brake pad outline mask is calculated for each frame. For example, if the temperature in a certain area drops from 300°C to 280°C in 5 seconds, the instantaneous decay rate is 4°C / s.

[0163] To establish a thermal relaxation model, a fractional-order differential equation (FODE) was used to describe the unsteady heat transfer process in brake pads. Compared to traditional integer-order models, FODE introduces fractional-order derivatives (e.g., 0.5-order) to more accurately characterize the thermal memory effect within the material (i.e., the influence of historical temperature on current heat dissipation). The model's input parameters include the thermal diffusivity of the brake pad material (e.g., 2.3×10 for cast iron), -5m² / s), the latent heat of evaporation of the spray water film (approximately 2260 kJ / kg), and the surface convective heat transfer coefficient inverted from thermal imaging data. The Grünwald-Letnikov discretization method was used to solve the FODE and obtain the theoretical temperature decay curve.

[0164] Deviation quantification utilizes a dynamic time warping (DTW) algorithm to nonlinearly align the measured temperature curve with the theoretically predicted curve, calculating the cumulative distance between the two along the optimal path. For example, if the measured curve after a spraying event reaches the same temperature point two seconds later than the theoretical curve, the deviation is labeled "time lag." If the temperature value consistently exceeds the predicted value, it is labeled "amplitude exceedance." This deviation serves as a key metric for subsequent optimization, with a threshold set at 15% (exceeding this threshold triggers a correction mechanism).

[0165] The deviation is input into the Bayesian optimization algorithm, and a nonlinear mapping relationship between the temperature safety threshold curve parameters and the spray instruction set is established through Gaussian process regression. The Pareto optimal solution set is generated by combining Monte Carlo sampling to correct the temperature safety threshold and the spray instruction. A Gaussian process regression model was constructed with input variables including the threshold slope and spray duration, and the output was the deviation. Monte Carlo sampling (1000 iterations) was used to generate the Pareto frontier, and the optimal solution (for example, a threshold slope of 0.58 and a spray duration of 3.2 seconds) was selected, reducing the deviation to below 4%. This adaptive parameter optimization (increasing optimization efficiency by 50%) balanced cooling effectiveness with resource consumption, improving system robustness.

[0166] The core of Bayesian optimization is to use Gaussian Process Regression (GPR) to build a probabilistic model between the objective function (i.e., cooling effect) and decision variables (such as spray duration, water pressure, and atomization particle size). The specific process is as follows: Parameter space definition: Encode the spray instruction set into a 5-dimensional vector, including: Spray duration (0.5~5 seconds, step size 0.1 second); Piezoelectric ceramic microvalve opening (20%~100%, corresponding to water flow rate 2~10 L / min); Ultrasonic atomizer frequency (1.5~3 MHz, affecting water mist particle size between 10~50 microns); Nozzle elevation angle (15°~45°, controlling the parabolic trajectory of the water flow); Trigger delay (0~200 ms, to compensate for vehicle inertia).

[0167] Gaussian process modeling: A radial basis function (RBF) is used as the covariance kernel, and its hyperparameters are automatically optimized through maximum likelihood estimation (MLE). For example, when historical data indicates a nonlinear saturation relationship between spray duration and cooling rate, the RBF kernel can capture this "diminishing returns" characteristic.

[0168] Monte Carlo sampling and Pareto optimization: 100 initial parameter combinations were generated using Latin Hypercube Sampling (LHS). The mean and variance of the deviations in the cooling effect predicted by the GPR model were then input. The next set of parameters to be tested was selected based on a modified Expected Improvement (EI) acquisition function. Furthermore, the NSGA-III multi-objective optimization algorithm was employed to find Pareto frontier solutions under the conflicting objectives of maximizing the cooling rate and minimizing water consumption. For example, a single optimization run might yield two optimal solutions: Solution A: Spray for 3 seconds + water pressure 80%, the deviation is reduced by 12% but the water consumption is 6L; Solution B: Spray for 2 seconds + water pressure 60% + atomization particle size 30μm, the deviation is reduced by 9% but the water consumption is only 3L.

[0169] The final decision is made by the traffic management platform based on real-time water resource reserves. The revised spraying instructions are sent to roadside equipment via the 5G-V2X link, and the critical values of the temperature safety threshold curve are updated simultaneously (for example, adjusting the first-level warning threshold from 280°C to 275°C).

[0170] The vehicle ID, temperature peak, and spray response time in the high-temperature event data are feature-encoded, compressed using a lightweight encryption algorithm, and uploaded to the traffic management platform. The platform verifies the data integrity through blockchain technology and triggers an upgrade in the flashing frequency of the roadside LED warning screen. It also synchronizes warnings with the on-board terminal through the 5G-V2X communication link between the on-board terminal and the roadside equipment, achieving closed-loop safety regulation of the road section.

[0171] Data (such as vehicle ID hash values and peak temperatures exceeding 380°C) is compressed and encoded using the AES-128 encryption algorithm (compression ratio ≥ 70%). Blockchain nodes verify the data hash using SHA-256. Roadside LED screens adjust their flashing frequency based on the warning level (1Hz for level 1 and 2Hz for level 2). 5G-V2X ensures end-to-end latency ≤ 10ms. This builds a trusted, real-time closed-loop control network (reducing the risk of data leakage by 99%), enabling coordinated safety responses across all road sections and improving the overall reliability of the transportation system.

[0172] The processing of high temperature event data is divided into three stages: Feature encoding and compression: The vehicle ID is converted into a hash value (such as SHA-256 truncated to 64 bits) to avoid privacy leakage; The temperature peak and spray response time are quantized into 16-bit floating point numbers, and Delta coding is used to compress the timing differences between adjacent events; Data packets are encrypted using a lightweight encryption algorithm (such as ChaCha20-Poly1305). The key is rotated every 30 minutes and is managed by the hardware security module (HSM) of the roadside equipment.

[0173] Blockchain evidence storage and verification: The traffic management platform deploys a private blockchain network, with each roadside device participating as a node in the consensus (using the PBFT algorithm, which can tolerate 1 / 3 node failures). When a data packet is uploaded, it is accompanied by a digital signature (ECDSA-secp256k1 curve). After the blockchain smart contract verifies the validity of the signature, it writes the event data into an immutable distributed ledger. For example, the block structure of a high temperature event includes: Block #7821: - Vehicle ID hash: 0x3a7d...f2c1; - Peak temperature: 320℃; - Spray response delay: 1.2 seconds; - Previous block hash: 0x5b9e...d4a3.

[0174] Multi-terminal linkage warning: The roadside LED warning screen switches the flashing frequency according to the severity of the incident (level one warning 1Hz yellow, level two warning 2Hz red); The vehicle terminal receives the warning message through the PC5 interface of 5G-V2X, triggering the sound and light alarm in the cockpit (such as 80dB buzzer + HUD red flashing icon); After the platform aggregates data from all road sections, it dynamically adjusts the spraying strategy through reinforcement learning. For example, if 10 vehicles in a row trigger a high-temperature event at the same location, the system automatically marks the area as a "high-risk section" and initiates pre-spraying 50 meters in advance.

[0175] The real-time nature of closed-loop control relies on 5G URLLC (Ultra-Reliable Low-Latency Communication) technology, keeping end-to-end latency below 20ms. All operation logs are stored on-chain for audit by regulatory authorities.

[0176] It can be seen that according to the high-precision infrared temperature sensor array and thermal imager deployed on both sides of the long downhill section, the surface temperature data of the truck's brake pads are collected in real time to generate a time-space correlated temperature field distribution map; based on the temperature field distribution map, the temperature safety threshold curve is dynamically adjusted through reinforcement learning, and a dynamic warning trigger signal is output; according to the dynamic warning trigger signal, a corresponding spray instruction is generated to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pads; according to the temperature decay rate of the brake pads after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and the spray instruction are corrected through the Bayesian optimization algorithm to form a closed-loop safety control, so that the temperature of the brake pads can be monitored in real time, and precise spray cooling control can be achieved, ensuring that the truck maintains safe and stable braking performance during the downhill process.

[0177] Another embodiment of the present invention provides a high temperature detection and cooling system for truck brake pads on a long downhill section. Figure 3 , the system may include: Acquisition module 301 is used to collect real-time surface temperature data of truck brake pads using high-precision infrared temperature sensor arrays and thermal imagers deployed on both sides of the long downhill section. It also uses the on-board OBD system to obtain vehicle speed, load, and slope angle parameters. It then generates a spatiotemporally correlated temperature field distribution map using a multimodal data fusion algorithm that uses an adaptive Kalman filter to eliminate environmental noise and compensates for temperature measurement errors based on the vehicle's motion trajectory. A locking module 302 is configured to identify the brake pad contour based on the temperature field distribution map using a deep convolutional neural network, lock the target area using a spatiotemporal attention mechanism, dynamically adjust the temperature safety threshold curve through reinforcement learning based on real-time meteorological data and a historical brake failure case library, and output a dynamic warning trigger signal; A generation module 303 is configured to invoke a fluid mechanics simulation model based on the dynamic warning trigger signal, predict the water flow coverage range through vortex flow field simulation, adjust the spray duration based on the heat capacity characteristics of the brake pad material, and generate a corresponding spray instruction to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pad; Correction module 304 is used to correct the temperature safety threshold curve and spraying instructions based on the temperature decay rate of the brake pad after spraying and the infrared thermal image feedback data through the Bayesian optimization algorithm. At the same time, it uploads the high temperature event data to the traffic management platform, triggering the roadside LED warning screen and the on-board terminal to issue synchronous warnings, forming a closed-loop safety control.

[0178] It can be seen that according to the high-precision infrared temperature sensor array and thermal imager deployed on both sides of the long downhill section, the surface temperature data of the truck's brake pads are collected in real time to generate a time-space correlated temperature field distribution map; based on the temperature field distribution map, the temperature safety threshold curve is dynamically adjusted through reinforcement learning, and a dynamic warning trigger signal is output; according to the dynamic warning trigger signal, a corresponding spray instruction is generated to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pads; according to the temperature decay rate of the brake pads after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and the spray instruction are corrected through the Bayesian optimization algorithm to form a closed-loop safety control, so that the temperature of the brake pads can be monitored in real time, and precise spray cooling control can be achieved, ensuring that the truck maintains safe and stable braking performance during the downhill process.

[0179] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps of any one of the above method embodiments when running.

[0180] Specifically, in this embodiment, the above-mentioned storage medium may be configured to store a computer program for performing the following steps: S201: Using high-precision infrared temperature sensor arrays and thermal imagers deployed on both sides of the long downhill section, real-time surface temperature data of the truck's brake pads is collected. Vehicle speed, load, and slope angle parameters are acquired through the onboard OBD system. A spatiotemporal temperature field distribution map is generated using a multimodal data fusion algorithm. The multimodal data fusion algorithm employs an adaptive Kalman filter to eliminate ambient noise and compensates for temperature measurement errors based on the vehicle's motion trajectory. S202, based on the temperature field distribution map, using a deep convolutional neural network to identify the brake pad contour, combining a spatiotemporal attention mechanism to lock the target area, and dynamically adjusting the temperature safety threshold curve through reinforcement learning based on real-time meteorological data and a historical brake failure case library, and outputting a dynamic warning trigger signal; S203: Based on the dynamic warning trigger signal, a fluid dynamics simulation model is invoked to predict the water flow coverage range through vortex flow field simulation. The spray duration is adjusted based on the heat capacity characteristics of the brake pad material, and a corresponding spray instruction is generated to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pad; S204, based on the temperature decay rate of the brake pad after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and spraying instructions are corrected through the Bayesian optimization algorithm. At the same time, the high temperature event data is uploaded to the traffic management platform, triggering the roadside LED warning screen and the on-board terminal to synchronize warnings, forming a closed-loop safety control.

[0181] It can be seen that according to the high-precision infrared temperature sensor array and thermal imager deployed on both sides of the long downhill section, the surface temperature data of the truck's brake pads are collected in real time to generate a time-space correlated temperature field distribution map; based on the temperature field distribution map, the temperature safety threshold curve is dynamically adjusted through reinforcement learning, and a dynamic warning trigger signal is output; according to the dynamic warning trigger signal, a corresponding spray instruction is generated to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pads; according to the temperature decay rate of the brake pads after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and the spray instruction are corrected through the Bayesian optimization algorithm to form a closed-loop safety control, so that the temperature of the brake pads can be monitored in real time, and precise spray cooling control can be achieved, ensuring that the truck maintains safe and stable braking performance during the downhill process.

[0182] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any one of the above method embodiments.

[0183] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0184] Specifically, in this embodiment, the processor may be configured to execute the following steps through a computer program: S201: Using high-precision infrared temperature sensor arrays and thermal imagers deployed on both sides of the long downhill section, real-time surface temperature data of the truck's brake pads is collected. Vehicle speed, load, and slope angle parameters are acquired through the onboard OBD system. A spatiotemporal temperature field distribution map is generated using a multimodal data fusion algorithm. The multimodal data fusion algorithm employs an adaptive Kalman filter to eliminate ambient noise and compensates for temperature measurement errors based on the vehicle's motion trajectory. S202, based on the temperature field distribution map, using a deep convolutional neural network to identify the brake pad contour, combining a spatiotemporal attention mechanism to lock the target area, and dynamically adjusting the temperature safety threshold curve through reinforcement learning based on real-time meteorological data and a historical brake failure case library, and outputting a dynamic warning trigger signal; S203: Based on the dynamic warning trigger signal, a fluid dynamics simulation model is invoked to predict the water flow coverage range through vortex flow field simulation. The spray duration is adjusted based on the heat capacity characteristics of the brake pad material, and a corresponding spray instruction is generated to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pad; S204, based on the temperature decay rate of the brake pad after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and spraying instructions are corrected through the Bayesian optimization algorithm. At the same time, the high temperature event data is uploaded to the traffic management platform, triggering the roadside LED warning screen and the on-board terminal to synchronize warnings, forming a closed-loop safety control.

[0185] It can be seen that according to the high-precision infrared temperature sensor array and thermal imager deployed on both sides of the long downhill section, the surface temperature data of the truck's brake pads are collected in real time to generate a time-space correlated temperature field distribution map; based on the temperature field distribution map, the temperature safety threshold curve is dynamically adjusted through reinforcement learning, and a dynamic warning trigger signal is output; according to the dynamic warning trigger signal, a corresponding spray instruction is generated to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pads; according to the temperature decay rate of the brake pads after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and the spray instruction are corrected through the Bayesian optimization algorithm to form a closed-loop safety control, so that the temperature of the brake pads can be monitored in real time, and precise spray cooling control can be achieved, ensuring that the truck maintains safe and stable braking performance during the downhill process.

[0186] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the scope of implementation of the present invention is not limited to what is shown in the drawings. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which do not exceed the spirit covered by the description and drawings, should be within the scope of protection of the present invention.

Claims

1. A method for detecting and cooling high temperature of truck brake pads on a long downhill section, characterized in that: The method comprises: High-precision infrared temperature sensor arrays and thermal imagers deployed on both sides of the long downhill section collect real-time surface temperature data of the truck's brake pads. Combined with the vehicle's onboard optical borehole detection (OBD) system to obtain vehicle speed, load, and slope angle parameters, a multimodal data fusion algorithm is used to generate a spatiotemporally correlated temperature field distribution map. This multimodal data fusion algorithm uses an adaptive Kalman filter to eliminate environmental noise and compensates for temperature measurement errors based on the vehicle's motion trajectory. Based on the temperature field distribution map, a deep convolutional neural network is used to identify the brake pad contour. The target area is locked in combination with a spatiotemporal attention mechanism. Based on real-time meteorological data and a historical brake failure case library, the temperature safety threshold curve is dynamically adjusted through reinforcement learning to output a dynamic warning trigger signal. Based on the dynamic warning trigger signal, a fluid mechanics simulation model is called to predict the water flow coverage range through vortex flow field simulation. The spray duration is adjusted based on the heat capacity characteristics of the brake pad material, and a corresponding spray instruction is generated to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pad; Based on the temperature decay rate of the brake pad after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and spray instructions are corrected through the Bayesian optimization algorithm. At the same time, the high temperature event data is uploaded to the traffic management platform, triggering the roadside LED warning screen and the on-board terminal to synchronize warnings, forming a closed-loop safety control.

2. The method according to claim 1, characterized in that The system uses high-precision infrared temperature sensor arrays and thermal imagers deployed on both sides of the long downhill section to collect surface temperature data of truck brake pads in real time. It also uses the on-board OBD system to obtain vehicle speed, load, and slope angle parameters. A multimodal data fusion algorithm is used to generate a spatiotemporally correlated temperature field distribution map. The multimodal data fusion algorithm uses an adaptive Kalman filter to eliminate environmental noise and compensates for temperature measurement errors based on the vehicle's motion trajectory. The algorithm includes the following steps: Based on the raw temperature data collected by a high-precision infrared temperature sensor array and a thermal imager, a nonlinear phase synchronization algorithm is used to dynamically compensate for multi-sensor clock bias and generate a temporally and spatially aligned temperature data stream. The synchronization algorithm uses millimeter-wave radar signals reflected from the edge of the vehicle's wheel hub as a time reference, achieving sub-millisecond synchronization accuracy. The vehicle speed, load, and slope angle parameters acquired by the onboard OBD system are input into the multi-physics field coupling model. A real-time calculation function for the heat generation power of the brake pad is constructed based on the vehicle dynamics equation. This function is then jointly calibrated with the spatiotemporally aligned temperature data stream to generate a temperature-mechanical correlation feature matrix. The temperature-mechanical correlation characteristic matrix is processed to eliminate environmental interference. An adaptive Kalman filter algorithm is used to model rain and fog scattering noise as a time-varying Gaussian mixture distribution. The noise covariance matrix is iteratively updated through expectation maximization. Based on the continuous Bezier curve fitting results of the vehicle motion trajectory, the sensor measurement position deviation is dynamically compensated to output environmentally corrected brake pad temperature field distribution data. The corrected brake pad temperature field distribution data is input into a spatiotemporal correlation encoder. The spatiotemporal evolution pattern of the brake pad surface temperature gradient is extracted through tensor decomposition technology, and the vehicle load and slope angle parameters are integrated to generate a three-dimensional temperature field distribution map. The encoder adopts a dynamic weight allocation strategy to perform nonlinear enhancement mapping on the high-temperature area.

3. The method according to claim 2, characterized in that Based on the temperature field distribution map, a deep convolutional neural network is used to identify the brake pad contour, and a spatiotemporal attention mechanism is combined to lock the target area. According to real-time meteorological data and a historical brake failure case library, the temperature safety threshold curve is dynamically adjusted through reinforcement learning to output a dynamic warning trigger signal, including: Based on the three-dimensional temperature field distribution map, a multi-scale feature fusion module of a deep convolutional neural network is used to extract the gradient mutation features of the brake pad edge. Geometric constraints are established in combination with the thermal expansion coefficient of the brake pad material to generate a pixel-level segmentation mask of the brake pad contour. The segmentation mask is input into the spatiotemporal attention mechanism. Based on the historical temperature change rate of the brake pad and the real-time wind speed data, the temperature sensitivity weight of each area is calculated. The gated recurrent unit is used to predict the heat diffusion path in the next few seconds and target high-risk target areas. The system uses a historical brake failure case library and transfer learning technology to map the temperature-failure time series in the case into equivalent feature vectors under the current vehicle load and slope angle conditions. Combined with humidity and air density parameters from real-time meteorological data, the Q-learning algorithm in reinforcement learning dynamically adjusts the slope and intercept of the temperature safety threshold curve. The dynamically adjusted temperature safety threshold curve is compared frame by frame with the real-time temperature field distribution. When the temperature in the target area continuously crosses the threshold and the rate of change exceeds the preset critical value, a graded warning signal is triggered. Among them, the first-level warning triggers the vibration prompt of the vehicle terminal, and the second-level warning simultaneously activates the roadside LED warning screen.

4. The method according to claim 3, characterized in that The method calls a fluid mechanics simulation model based on the dynamic warning trigger signal, predicts the water flow coverage range through vortex flow field simulation, adjusts the spraying duration based on the heat capacity characteristics of the brake pad material, generates a corresponding spraying instruction, and drives the piezoelectric ceramic microvalve array of the roadside spraying device to spray and cool the brake pad, including: Based on the dynamic warning trigger signal, a fluid dynamics simulation engine based on the Lattice Boltzmann method is invoked to perform vortex flow field topological decomposition on the water flow of the roadside sprinkler. The three-dimensional coverage of the water flow under the influence of air resistance and gravity is predicted by combining the truck's driving speed and the spatial coordinates of the brake pads. Based on the heat capacity characteristics and phase change latent heat parameters of the brake pad material, a heat conduction-convection coupling equation was constructed. The temperature decay curves for different spraying times were solved through finite element discretization, and a greedy algorithm was used to select the spraying time that met the cooling rate requirements. The three-dimensional coverage area and the spray duration are input into the digital twin control module, which generates high-frequency switching instructions through pulse width modulation technology of the piezoelectric ceramic microvalve. At the same time, the resonant frequency of the ultrasonic atomizer is adjusted to control the water mist particle size, so that the nanoscale water film evenly covers the brake pad surface; LiDAR is used to monitor the truck's position in real time. When the truck enters the spraying area, the nozzle deflection angle is dynamically corrected according to the wheel speed. Through distributed collaborative control of the microvalve array, sub-millisecond synchronization of the water jet direction and the vehicle's motion trajectory is achieved.

5. The method according to claim 4, characterized in that Based on the temperature decay rate of the brake pad after spraying and the infrared thermal image feedback data, the temperature safety threshold curve and spraying instructions are corrected through the Bayesian optimization algorithm. At the same time, the high temperature event data is uploaded to the traffic management platform, triggering the roadside LED warning screen and the vehicle terminal to synchronize warnings, forming a closed-loop safety control, including: Based on the infrared thermal imaging feedback data of the brake pad after spraying, the temperature decay rate curve is extracted, and a thermal relaxation model is constructed based on fractional differential equations to quantify the deviation between the cooling effect and the theoretical prediction value; The deviation is input into the Bayesian optimization algorithm, and a nonlinear mapping relationship between the temperature safety threshold curve parameters and the spray instruction set is established through Gaussian process regression. The Pareto optimal solution set is generated by combining Monte Carlo sampling to correct the temperature safety threshold and the spray instruction. The vehicle ID, temperature peak, and spray response time in the high-temperature event data are feature-encoded, compressed using a lightweight encryption algorithm, and uploaded to the traffic management platform. The platform verifies the data integrity through blockchain technology and triggers an upgrade in the flashing frequency of the roadside LED warning screen. It also synchronizes warnings with the on-board terminal through the 5G-V2X communication link between the on-board terminal and the roadside equipment, achieving closed-loop safety regulation of the road section.

6. The method according to claim 2, characterized in that The vehicle speed, load, and slope angle parameters acquired by the onboard OBD system are input into the multi-physics field coupling model, and a real-time calculation function for the heat generation power of the brake pad is constructed based on the vehicle dynamics equation. The function is then jointly calibrated with the time-space aligned temperature data stream to generate a temperature-mechanical correlation feature matrix, including: Based on vehicle speed, load, and slope angle parameters acquired by the on-board OBD system, a nonlinear vehicle dynamics model is used to construct a differential equation for the brake pad force. The implicit Runge-Kutta algorithm is used to dynamically solve the brake pad friction torque, generating a high-frequency mechanical characteristic sequence containing transient friction work and slip rate. The nonlinear vehicle dynamics model introduces a random roughness parameter of the tire-road contact surface as a disturbance term. The high-frequency mechanical characteristic sequence is input into the thermal-mechanical coupling field analysis module. Based on the thermal conductivity coefficient of the brake pad material and the micromorphology data of the friction interface, the non-Fourier heat conduction equation is solved by fractional order discretization to generate the spatiotemporal distribution function of the heat flux density on the brake pad surface. The solution process uses adaptive mesh encryption technology to capture the nonlinear mutation of the heat flux boundary layer. The spatiotemporal distribution function is subjected to multi-scale tensor decomposition to extract the coupling characteristics of the brake pad surface temperature gradient and friction torque. Simultaneously, the spatiotemporally aligned infrared temperature data streams are fused, and a temperature-mechanical calibration mapping relationship is constructed through a generative adversarial network to generate a physically consistent three-dimensional joint feature tensor. The three-dimensional joint feature tensor is input into the evolutionary multi-objective optimization module constrained by physical information. The Pareto frontier search space is constructed based on the thermodynamic entropy increase principle and the friction power dissipation equation. The NSGA-III algorithm is used to simultaneously optimize the three indicators of temperature gradient distribution uniformity, friction work equivalent error and computational efficiency. The population diversity is dynamically adjusted through the adaptive mutation operator, and finally the temperature-mechanical correlation feature matrix that satisfies the multi-physical field coupling constraints is output.

7. A high temperature detection and cooling system for truck brake pads on a long downhill section, characterized in that: The system comprises: An acquisition module collects real-time surface temperature data of truck brake pads using high-precision infrared temperature sensor arrays and thermal imagers deployed on both sides of the long downhill section. It also uses the onboard OBD system to obtain vehicle speed, load, and slope angle parameters. A multimodal data fusion algorithm is used to generate a spatiotemporally correlated temperature field distribution map. The algorithm uses an adaptive Kalman filter to eliminate ambient noise and compensates for temperature measurement errors based on the vehicle's motion trajectory. A locking module is configured to identify the brake pad contour based on the temperature field distribution map using a deep convolutional neural network, lock the target area using a spatiotemporal attention mechanism, dynamically adjust the temperature safety threshold curve through reinforcement learning based on real-time meteorological data and a historical brake failure case library, and output a dynamic warning trigger signal; a generation module for invoking a fluid mechanics simulation model based on the dynamic warning trigger signal, predicting the water flow coverage range through vortex flow field simulation, adjusting the spray duration based on the heat capacity characteristics of the brake pad material, and generating a corresponding spray instruction to drive the piezoelectric ceramic microvalve array of the roadside spray device to spray and cool the brake pad; The correction module is used to correct the temperature safety threshold curve and spraying instructions based on the temperature decay rate of the brake pad after spraying and the infrared thermal image feedback data through the Bayesian optimization algorithm. At the same time, the high temperature event data is uploaded to the traffic management platform, triggering the roadside LED warning screen and the on-board terminal to issue synchronous warnings, forming a closed-loop safety control.

8. The system according to claim 7, characterized in that The acquisition module is specifically used to: Based on the raw temperature data collected by a high-precision infrared temperature sensor array and a thermal imager, a nonlinear phase synchronization algorithm is used to dynamically compensate for multi-sensor clock bias and generate a temporally and spatially aligned temperature data stream. The synchronization algorithm uses millimeter-wave radar signals reflected from the edge of the vehicle's wheel hub as a time reference, achieving sub-millisecond synchronization accuracy. The vehicle speed, load, and slope angle parameters acquired by the onboard OBD system are input into the multi-physics field coupling model. A real-time calculation function for the heat generation power of the brake pad is constructed based on the vehicle dynamics equation. This function is then jointly calibrated with the spatiotemporally aligned temperature data stream to generate a temperature-mechanical correlation feature matrix. The temperature-mechanical correlation characteristic matrix is processed to eliminate environmental interference. An adaptive Kalman filter algorithm is used to model rain and fog scattering noise as a time-varying Gaussian mixture distribution. The noise covariance matrix is iteratively updated through expectation maximization. Based on the continuous Bezier curve fitting results of the vehicle motion trajectory, the sensor measurement position deviation is dynamically compensated to output environmentally corrected brake pad temperature field distribution data. The corrected brake pad temperature field distribution data is input into a spatiotemporal correlation encoder. The spatiotemporal evolution pattern of the brake pad surface temperature gradient is extracted through tensor decomposition technology, and the vehicle load and slope angle parameters are integrated to generate a three-dimensional temperature field distribution map. The encoder adopts a dynamic weight allocation strategy to perform nonlinear enhancement mapping on the high-temperature area.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 6 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 6.

Images