Caliper brake service life prediction method and system based on machine learning

Abstract

The invention relates to the technical field of artificial intelligence prediction, and discloses a caliper brake service life prediction method and system based on machine learning, and the method comprises the steps: carrying out the pixel-level segmentation based on a corrected image set, calculating a wear rate, and obtaining a wear trend curve graph. The U-2-Net model automatically extracts multi-scale semantic features and optimizes boundary segmentation precision by nesting a U-shaped framework and a side output supervision mechanism, pixel-level accurate recognition of a wear area is realized, an improved Paris formula is combined with a temperature-stress coupling coefficient, geometric factors and crack growth rate are dynamically calibrated through machine learning, and the accuracy of the wear area is improved. A nonlinear degradation behavior under the combined action of temperature and stress is quantified, a machine learning driven rule engine can fuse dynamic data such as wear rate, crack length and hot spot area in real time, and through LOESS smoothing and fractal dimension change rate analysis, an early warning threshold value is adaptively adjusted and a three-dimensional feature map is generated. And accurate division from an initial damage stage to a later damage stage is realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence prediction technology, and more specifically, to a method and system for predicting the lifespan of caliper brakes based on machine learning. Background Technology

[0002] With the development of the times, machine learning, as an artificial intelligence technology, no longer relies on manually written fixed calculation rules. Instead, it autonomously mines the hidden patterns and mapping relationships in the data by automatically learning, extracting features, and training models on a large amount of data. Using this technology, multi-dimensional operation and degradation data of caliper brakes in actual use can be collected. Based on this, by building a suitable machine learning prediction model, the intrinsic relationship between brake performance degradation and life loss can be learned. This breaks away from the limitations of traditional empirical formulas and simplified mechanical models, and achieves accurate and dynamic prediction of the remaining service life, failure points, and reliability changes of caliper brakes. This provides intelligent technical support for the condition monitoring, preventive maintenance, and safety early warning of braking systems.

[0003] However, the caliper brake life prediction method has several shortcomings in practical applications. First, the feature extraction dimension is often relatively simple, which makes it impossible for the model to fully capture complex failure modes. This limitation means that the prediction results may lack accuracy and fail to reflect the true performance of the brake under different operating conditions. At the same time, the static nature of the dynamic evaluation model also significantly affects the real-time performance and effectiveness of the prediction. As the environment and usage conditions change, the static model is difficult to update adaptively, which may lead to the life prediction results lagging behind actual needs. In addition, existing solutions have failure issues in multi-source data fusion. Differences in data format, features, and quality from different data sources may lead to fusion failure, thereby affecting the overall predictive ability and reliability of the model. Summary of the Invention

[0004] To overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a method and system for predicting the lifespan of caliper brakes based on machine learning.

[0005] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, a machine learning-based method for predicting the lifespan of a caliper brake is provided, the method comprising: S101: Acquire a multi-angle original image set of the brake during stationary and braking processes, correct and enhance the multi-angle original image set, and generate a corrected image set; S102: Pixel-level segmentation is performed based on the calibration image set to calculate the wear rate and obtain the wear trend curve; texture features are extracted from the calibration image set, crack linear data are measured, infrared images are acquired, and the infrared images are registered with the calibration image set to obtain the hot spot marking map. S103: Collect the temperature, pressure and vibration parameters of the brake during the braking process, preprocess and align the three types of parameters, and obtain a multi-source dataset after alignment; S104: Combine wear rate, crack linear data, and hot spot marking map to draw a three-dimensional feature map, calculate the damage matrix, generate a damage report, dynamically formulate life assessment indicators, use the improved Paris formula and integrate the damage report and life assessment indicators to derive a life prediction curve and generate maintenance recommendations.

[0006] Furthermore, the calculation of the wear rate and the resulting wear trend curve include: The corrected image set is output to the trained U-2-Net model, and the boundary segmentation accuracy is optimized by combining the side output supervision mechanism to obtain a binarized segmentation map. An adaptive window is applied to perform morphological gradient operations on the binarized segmentation image. The opening operation eliminates small noise in the binarized segmentation image, while the closing operation fills in the holes in the region, and the optimized segmentation image is output. The wear rate is obtained by calculating the ratio of the pixel area of ​​the worn area to the total pixel area of ​​the brake disc. The wear rate is then plotted by correlating the number of braking operations over time and drawing a dynamic curve between the wear rate and the number of braking operations. The non-linear trend is captured by using LOESS smoothing to obtain a wear trend curve.

[0007] Furthermore, the step of extracting texture features from the corrected image set and measuring crack linear data includes: Convolutional filtering is applied to the corrected image set to generate a texture feature map. Adaptive thresholding is used to identify the transverse and longitudinal crack orientations from the texture feature map, and a crack orientation map is output. The crack centerline was extracted and refined using the Zhang-Suen skeletonization algorithm. The length, width, and number of branches of the crack were measured to obtain crack region data. The fractal dimension of the crack region was calculated using the box-difference dimension method to quantify the surface roughness change and calculate the crack length growth rate, thus obtaining the crack linear data.

[0008] Further, the registration of the infrared image with the calibration image set to obtain the hot spot marker map includes: Feature points are extracted from the visible light and infrared images in the calibration image set, respectively. The extracted feature points are matched to obtain matching point pairs. Based on the matching point pairs, registration is performed to obtain registered image pairs. The edge contour of the brake disc is extracted from the registered image pair. The edge contour is used as the boundary condition for solving the heat conduction equation. The temperature field distribution on the surface of the brake disc is inverted to generate a temperature distribution map. Hot spot regions are extracted from the temperature distribution map. Adjacent hot spot regions are merged, and the hot spot position, area and temperature peak are calculated to obtain a hot spot marking map.

[0009] Furthermore, the specific steps for drawing a three-dimensional feature map by combining wear rate, crack linear data, and hot spot marking map are as follows: Using the number of braking cycles as the Z-axis, wear rate as the X-axis, crack length as the Y-axis, and hot spot area as the color code, a three-dimensional scatter plot is constructed and converted into a three-dimensional feature map.

[0010] Furthermore, the specific steps for dynamically formulating life assessment indicators are as follows: Wear threshold setting: The upper limit of wear rate is adjusted linearly based on the number of braking events. When the number of braking events is less than M, the yellow warning threshold is a and the red warning threshold is b. When the number of braking attempts is greater than or equal to M, the yellow warning threshold is h, and the red warning threshold is k. Crack threshold setting: Calculate the rate of change of the fractal dimension. If the rate of change is greater than n, adjust the crack critical value, compare the crack length with the adjusted crack critical value, and assess the failure risk. Thermal stress index setting: Calculate the real-time difference between the highest temperature of the brake disc and the boiling point of the brake fluid to assess the probability of thermal failure.

[0011] Furthermore, the process of deriving the lifetime prediction curve includes: Calculate the stress intensity factor based on crack length and pressure parameters. Then, by applying the improved Paris formula, the crack propagation behavior under the combined effects of temperature and stress is quantified, namely:

[0012] In the formula: ， Where is the temperature-stress coupling coefficient, Q is the activation energy, R is the gas constant, and T is the absolute temperature; By combining the number of braking cycles and integrating the remaining life, a life prediction curve is obtained.

[0013] Secondly, a machine learning-based caliper brake life prediction system is provided, which is implemented based on the aforementioned machine learning-based caliper brake life prediction method. The system includes: The image correction module acquires a set of original images of the brake from multiple angles during its stationary and braking processes, corrects and enhances the original image set from multiple angles, and generates a corrected image set. The data analysis module performs pixel-level segmentation based on the calibration image set, calculates the wear rate, and obtains the wear trend curve. It also extracts texture features from the calibration image set, measures crack linear data, and then acquires infrared images. The infrared images are registered with the calibration image set to obtain hot spot marking maps. The parameter acquisition module collects the temperature, pressure, and vibration parameters of the brake during the braking process, preprocesses and aligns the three types of parameters, and obtains a multi-source dataset after alignment. The life prediction module combines wear rate, crack linear data, and hot spot marking maps to draw a three-dimensional feature map, calculate the damage matrix, generate a damage report, dynamically formulate life assessment indicators, and use the improved Paris formula and the damage report and life assessment indicators to derive a life prediction curve and generate maintenance recommendations.

[0014] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the aforementioned machine learning-based caliper brake life prediction method.

[0015] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed, implements the aforementioned machine learning-based caliper brake life prediction method.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This application discloses a machine learning-based method and system for predicting the life of caliper brakes, comprising: the U-2-Net model of this invention, through a nested U-shaped architecture and a side-output supervision mechanism, automatically extracts multi-scale semantic features and optimizes boundary segmentation accuracy, achieving pixel-level accurate identification of wear areas; the improved Paris formula, combined with the temperature-stress coupling coefficient, dynamically calibrates geometric factors and crack propagation rates through machine learning, quantifying the nonlinear degradation behavior under the combined action of temperature and stress; in the multi-index comprehensive decision-making, the machine learning-driven rule engine can fuse dynamic data such as wear rate, crack length, and hot spot area in real time, and adaptively adjust the warning threshold and generate a three-dimensional feature map through LOESS smoothing and fractal dimension change rate analysis, achieving accurate division from the initial to the later damage stages; through automatic feature learning and multi-source data fusion, the robustness and accuracy of remaining life prediction are significantly improved, supporting intelligent decision-making for preventive maintenance strategies. Attached Figure Description

[0017] Figure 1A flowchart of a machine learning-based caliper brake life prediction method provided by the present invention; Figure 2 A schematic diagram of the module structure of a caliper brake life prediction system based on machine learning provided by the present invention; Figure 3 A schematic diagram of the structure of an electronic device provided by the present invention; Figure 4 The flowchart of S104 in the caliper brake life prediction method based on machine learning provided by the present invention; Figure 5 The flowchart of S102 in the caliper brake life prediction method based on machine learning provided by the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 Please see Figure 1 As shown, this embodiment discloses a machine learning-based method for predicting the lifespan of a caliper brake, the method comprising: S101: Acquire a multi-angle original image set of the brake during stationary and braking processes, correct and enhance the multi-angle original image set, and generate a corrected image set; It should be understood that in this embodiment, an industrial-grade 8K high-resolution camera (≥33 million pixels) is used with a ring LED light source and a polarizing filter to acquire surface images from three angles: 0°, 45° and 90°, when the brake is stationary or during dynamic braking. During dynamic braking, a synchronous triggering device ensures that each image corresponds to a key stage of the braking process.

[0020] A checkerboard calibration board was used to capture and correct a set of original images from multiple angles. The camera intrinsic matrix and distortion coefficients were calculated using the OpenCV distortion correction algorithm. The correction formula was then applied to the original image set from multiple angles to eliminate radial and tangential distortion, generating a distortion-free image set.

[0021] For the distortion-free image set, histogram equalization is first applied to expand the grayscale range of the image and improve the overall contrast. Then, adaptive gamma correction is used to optimize local details to ensure that the brake disc texture is clearly visible, resulting in a contrast-enhanced image set.

[0022] The BM3D denoising algorithm is used to input the contrast-enhanced image set. Similar image block groups in the contrast-enhanced image set are found through block matching. Three-dimensional collaborative filtering is performed to eliminate Gaussian noise. In this embodiment, the block size is set to 8×8 and the similar block search window is set to 39×39 to ensure that edge details are preserved, and a corrected image set is obtained.

[0023] S102: As Figure 5 As shown, pixel-level segmentation is performed based on the calibration image set to calculate the wear rate and obtain the wear trend curve; texture features are extracted from the calibration image set, crack linear data are measured, infrared images are acquired, and the infrared images are registered with the calibration image set to obtain the hot spot marking map. To quantify the wear area, the wear rate is calculated to obtain a wear trend curve, including: The corrected image set is output to the trained U-2-Net model, and the boundary segmentation accuracy is optimized by combining the side output supervision mechanism to obtain a binary segmentation map, where pixel value 0 represents the background and 255 represents the wear area. It should be noted that the U-2-Net model extracts multi-scale features through a nested U-shaped architecture and optimizes boundary segmentation accuracy by combining a side output supervision mechanism.

[0024] Nested U-shaped architecture: The encoder consists of 6 downsampling modules. Each module contains 2 3×3 convolutional layers (with ReLU activation function) + batch normalization (BN) + max pooling (2×2). During the downsampling process, the feature map size is halved layer by layer, and the number of channels is multiplied to extract multi-scale semantic features.

[0025] Decoder: It consists of 6 upsampling modules, each containing 2 3×3 convolutional layers + BN + bilinear upsampling (2×2), which fuse encoder features of the same scale (through skip connections) to restore the spatial resolution to the input size.

[0026] Side output supervision: Add a 1×1 convolutional layer to each level of the decoder (6 layers in total) to generate a side output segmentation map, and optimize the boundary accuracy through depth supervision.

[0027] Obtain a brake wear sample library, which can be obtained from historical brake wear data and includes the initial, intermediate, and late wear stages.

[0028] The U-2-Net model is trained based on a brake wear sample library. It adopts a nested U-shaped architecture combined with a side output supervision mechanism, improves generalization ability through data augmentation, optimizes boundary accuracy using Dice loss and multi-side output cross-entropy loss, and uses early stopping to prevent overfitting.

[0029] An adaptive window is applied to perform morphological gradient operations on the binarized segmentation image. The opening operation eliminates small noise in the binarized segmentation image, while the closing operation fills in the holes in the region, and the optimized segmentation image is output. The system automatically selects the optimal structural element size based on the area, for example, 3×3 for small areas and 15×15 for large areas.

[0030] The wear rate is obtained by calculating the ratio of the pixel area of ​​the worn region to the total pixel area of ​​the brake disc. The wear rate is then plotted by correlating the number of braking operations with the caliper brake over time, creating a dynamic curve between the wear rate and the number of braking operations. The non-linear trend is captured using LOESS smoothing to obtain a wear trend curve, the formula of which is:

[0031] In the formula: The pixel area of ​​the worn area. This represents the total pixel area of ​​the brake disc.

[0032] The number of braking cycles of a caliper brake refers to the number of times the brake is activated to apply braking force within a certain period of time. The number of braking cycles is usually used to evaluate the operating frequency and performance of the brake during use. A higher number of braking cycles may mean that the brake is working under frequent use or high load conditions. Understanding the number of braking cycles can help engineers assess key characteristics of the braking system such as durability and thermal fade performance, thus providing an important basis for maintenance and design.

[0033] LOESS is a locally weighted regression scatter smoothing method, which fits a weighted regression model around each data point. In this embodiment, the smoothing coefficient is 0.2.

[0034] Furthermore, the step of extracting texture features from the corrected image set and measuring crack linear data includes: A Gabor filter bank with 8 directions (0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°) and 5 scales (σ=1.0, 1.5, 2.0, 2.5, 3.0) is applied to the corrected image set for convolutional filtering to generate a texture feature map. The transverse and longitudinal crack orientations are identified from the texture feature map through adaptive thresholding, and the crack orientation map is output. The adaptive thresholding is defined as threshold = mean of texture feature map + 2 × standard deviation. The crack direction map is input using the Zhang-Suen skeletonization algorithm. The crack centerline is extracted, the crack centerline is refined, and the length, width, and number of branches of the crack are measured to obtain crack region data. The crack pixels are marked as foreground points and the background pixels are marked as background points. Each foreground point is traversed, and the following three conditions are checked. Condition 1: The number of crack pixels in the neighborhood is between 2 and 6.

[0035] Condition 2: The number of times the pixel value changes from the background to the crack in the neighborhood is 1, to ensure the continuity of the boundary.

[0036] Condition 3: The pixel position satisfies that at least one neighboring area is the background and does not form a closed region.

[0037] If all three conditions are met, the pixel is marked as to be deleted, a binary matrix is ​​obtained, the iteration is repeated, and finally the splitting trend map is output.

[0038] For example, the steps to refine the crack centerline are as follows: A 3×3 cross-shaped structural element (center point + four neighboring areas on all four sides) is used to perform an iterative erosion operation on the centerline image. The iteration rule is: each iteration removes only pixels that simultaneously meet the following conditions: These are edge pixels, meaning they are in the neighborhood where background pixels exist. Removal does not cause the center line to break, which can be determined by the connectivity of 8 neighboring areas; The iteration count should be ≤5 times. For example, if the change rate of the number of center line pixels is <1% after the 3rd iteration, the iteration should be terminated early.

[0039] The fractal dimension of the crack region is calculated using the box-difference dimension method to quantify the change in surface roughness; that is, the higher the fractal dimension, the rougher the surface. The crack length growth rate is also calculated, and the linear data of the crack are synthesized. The formula for calculating the crack length growth rate is as follows:

[0040] In the formula: To indicate the rate at which a crack grows from its initial length during fatigue testing or actual use. Extended to final length The absolute growth rate; Indicates the length of the crack Extended to The difference in the number of cyclic loads required.

[0041] Meanwhile, in this embodiment, the difference box dimension method optimized by dynamic programming is used to calculate the fractal dimension of the crack region within the box size range of 0.5mm-5mm, and the optimal box size corresponding to the minimum mean square error is selected through residual analysis.

[0042] Further, the registration of the infrared image with the calibration image set to obtain the hot spot marker map includes: Feature points are extracted from the visible light and infrared images in the calibration image set, respectively. The extracted feature points are matched to obtain matching point pairs. Based on the matching point pairs, registration is performed to obtain registered image pairs. Infrared images are images of the thermal radiation distribution of an object captured by an infrared thermal imager, providing temperature information, while visible light images provide structural and texture information about the actuator.

[0043] The edge contour of the brake disc is extracted from the registered image pair. The edge contour is used as the boundary condition for solving the heat conduction equation. The temperature field distribution on the surface of the brake disc is inverted to generate a temperature distribution map. Hot spot regions are extracted from the temperature distribution map. Adjacent hot spot regions are merged, and the hot spot position, area and temperature peak are calculated to obtain a hot spot marking map.

[0044] It should be added that the location of the hot spot can be obtained by identifying the bounding box coordinates of each independent hot spot region and then calculating the center coordinates of the independent hot spot region. The hotspot area is determined by first counting the total number of pixels within the hotspot region, then obtaining the pixel area of ​​each hotspot region, and finally obtaining the resolution parameter: Hotspot area = pixel area × resolution. 2 ; Resolution refers to the number of pixels contained within a unit physical length (such as millimeters or inches) in an image. In hot spot detection, resolution converts the pixel area into the actual physical area, and its determination method can be obtained from the device parameters.

[0045] For merging adjacent hot spot regions, a sub-image of the corresponding region is cropped from the temperature distribution map based on its bounding box coordinates. The temperature values ​​of all pixels in the sub-image are obtained, and the index position of the maximum value is found. The index position is mapped back to the pixel coordinates in the temperature distribution map, and the temperature value at that coordinate is recorded as the temperature peak of the current hot spot.

[0046] The expression for the heat conduction equation is: ; In the formula: Let t be the temperature and t be the time. Thermal diffusivity (m) 2 / s), Laplace operator.

[0047] S103: Collect the temperature, pressure and vibration parameters of the brake during the braking process, preprocess and align the three types of parameters, and obtain a multi-source dataset after alignment; In this embodiment, the temperature parameters are captured by an infrared thermal imager through a thermal imaging lens to capture the temperature distribution on the surface of the brake disc, generating temperature matrix data with a frame rate of 50Hz and a resolution of 0.1℃; the brake fluid embedded sensor collects the brake fluid temperature through a PT100 thermal resistor and outputs a 50Hz digital signal synchronously.

[0048] The pressure parameters are collected by the hydraulic pressure sensor through the piezoresistive effect to collect the change in brake caliper oil pressure and output a 1kHz analog signal, which is then converted into a digital signal after being amplified and filtered by the signal conditioning circuit; the contact pressure sensor collects the contact pressure between the brake pad and the brake disc through a thin-film strain gauge and outputs a 1kHz digital signal synchronously.

[0049] Vibration parameters are obtained by synchronously acquiring X / Y / Z three-dimensional vibration signals through a triaxial accelerometer, passing them through an anti-aliasing filter, and outputting them at a sampling rate of 20kHz. Then, time-frequency conversion is performed to obtain the vibration parameters. Finally, using the braking start timestamp as a reference, the temperature, pressure, and vibration data are aligned along the time axis to generate a multi-source dataset.

[0050] S104: As Figure 4 As shown, a three-dimensional feature map is drawn by combining wear rate, crack linear data, and hot spot marking map, the damage matrix is ​​calculated, a damage report is generated, life assessment indicators are dynamically formulated, and a life prediction curve is obtained by using the improved Paris formula and integrating the damage report and life assessment indicators, and maintenance recommendations are generated. Furthermore, the specific steps for drawing a three-dimensional feature map by combining wear rate, crack linear data, and hot spot marking map are as follows: Using the number of braking cycles as the Z-axis, wear rate as the X-axis, crack length as the Y-axis, and hot spot area as the color code, a three-dimensional scatter plot is constructed and converted into a three-dimensional feature map.

[0051] As one specific implementation method, the formula for calculating the damage matrix is:

[0052] In the formula: , Let i be the variable observation value of the i-th braking event. , denoted as the mean wear rate and the mean crack length, and n is the sample size.

[0053] In addition, in this embodiment, it is also necessary to divide the damage stage according to the crack length growth rate. And the rate of hot spot area expansion, combined with the number of braking events, to classify the damage stages: Initial stage: Rate <0.1 mm / cycle (crack) or <1 mm 2 / time (hot spot), damage spreads slowly; Intermediate stage: 0.1-0.3mm / crack or 1-5mm 2 / time (hot spot), damage spreads faster; Later stage: >0.3mm / crack or >5mm 2 / time (hot spot), the damage deteriorates rapidly and requires emergency intervention.

[0054] As a specific example, the calculation process is as follows: The braking frequency was from the 190th to the 200th braking action; Crack length growth data (mm): [190: 2.81, 191: 2.85, 192: 2.92, 193: 3.00, 194: 3.10, 195: 3.25, 196: 3.40, 197: 3.55, 198: 3.70, 199: 3.85, 200: 3.90]; Hot spot area expansion data (mm) 2 ): [190: 122, 191: 125, 192: 130, 193: 135, 194: 140, 195: 145, 196: 150, 197: 155, 198: 160, 199: 165, 200: 178]; Crack length difference = Length of 200 cracks - Length of 190 cracks = 3.90mm - 2.81mm = 1.09mm Calculate the crack length growth rate: 1.09 mm / 10 cycles = 0.109 mm / cycle Hot spot area difference = area of ​​200 hot spots - area of ​​190 hot spots = 178 mm 2 -122mm 2 =56mm 2 ; Hot spot area expansion rate = 56mm 2 / 10 times = 5.6mm 2 / Second-rate Furthermore, the specific steps for dynamically formulating life assessment indicators are as follows: Wear threshold setting: The upper limit of wear rate is adjusted linearly based on the number of braking events. When the number of braking events is less than M, the yellow warning threshold is a and the red warning threshold is b. When the number of braking attempts is greater than or equal to M, the yellow warning threshold is h, and the red warning threshold is k. As a specific example, M is set to 1000 times, a to 20%, b to 25%, h to 15%, and k to 20%. That is, when the number of braking actions is less than 1000, the yellow warning threshold is 20% and the red warning threshold is 25%; when the number of braking actions is greater than or equal to 1000, the yellow warning threshold is 15% and the red warning threshold is 20%.

[0055] Crack threshold setting: Calculate the rate of change of the fractal dimension. If the rate of change is greater than n, and n is set to 5%, adjust the critical crack value and compare the crack length with the adjusted critical crack value to assess the failure risk. The formula for calculating the rate of change of fractal dimension needs to be supplemented as follows:

[0056] In the formula: This is the fractal dimension for this week. This is the fractal dimension of last week; The formula for calculating the fractal dimension is as follows: In the formula, Minimum grid size required to cover hot spots This represents the number of non-empty grid cells.

[0057] Thermal stress index setting: Calculate the real-time difference between the highest temperature of the brake disc and the boiling point of the brake fluid to assess the probability of thermal failure.

[0058] Based on the above indicators, a summary judgment is made, and rules are established: Rule 1: If any two indicators exceed the threshold, a brake replacement command is triggered; Rule 2: If a single indicator reaches the emergency threshold, a replacement instruction will be triggered directly; It should be noted that the triggering logic of Rule 1 is as follows: by judging through the coordinated use of multiple indicators, the overall situation is reflected as deteriorating. Even if a single indicator does not reach the emergency threshold, the risk has increased significantly when multiple indicators are abnormally superimposed, and immediate intervention is required. The triggering logic for Rule 2 is as follows: if an extreme anomaly occurs in a single indicator, the system is directly determined to be in a high-risk state and must be replaced immediately to avoid catastrophic failure.

[0059] Rule 1 enables preventative maintenance, reducing the risk of system failure due to the coupling of multiple factors, and is suitable for progressive degradation scenarios. Rule 2 enables rapid fault isolation, avoiding catastrophic consequences caused by extreme anomalies in a single indicator, and is suitable for sudden high-risk scenarios.

[0060] Rule 3: Make a comprehensive judgment based on the stage of damage. For example, if the damage is in the late stage and a single indicator is close to 80% of the threshold, then trigger an early warning.

[0061] The emergency thresholds include: The emergency threshold for hot spot area is ≥200 mm. 2 (When the number of braking cycles is less than 1000); ≥180mm 2 (When the number of braking cycles is ≥1000); The critical threshold for crack length is ≥3.5mm; The emergency threshold for wear rate is: ≥25% (when the number of braking cycles is <1000); ≥20% (when the number of braking cycles is ≥1000). Furthermore, the process of deriving the lifetime prediction curve includes: Calculate the stress intensity factor based on crack length and pressure parameters. Then, by applying the improved Paris formula, the crack propagation behavior under the combined effects of temperature and stress is quantified, namely:

[0062] In the formula: , Where is the temperature-stress coupling coefficient, Q is the activation energy, R is the gas constant, and T is the absolute temperature; Stress intensity factor The calculation formula is: ; In the formula: Geometric factor Let be the stress, and 'a' be the crack length. The value of 'Y' needs to be calibrated through finite element analysis in conjunction with the specific crack morphology.

[0063] By combining the number of braking operations, the remaining life is calculated integrally, representing the difference in the number of braking operations required for the crack length to reach the critical value from the current number of braking operations. This yields the life prediction curve, the specific calculation formula of which is as follows:

[0064] In the formula: This represents the current crack length. The maximum permissible crack length specified in the material or design specifications. That is, the crack growth rate.

[0065] In addition, maintenance recommendations can be generated based on the life prediction curve, including: Emergency replacement: If the remaining lifespan is less than 500 braking cycles, immediate replacement is recommended and is the highest priority.

[0066] Monthly inspection: 500 times ≤ remaining lifespan ≤ 1000 times. It is recommended to conduct a detailed inspection every month to monitor the spread of damage.

[0067] Normal monitoring: If the remaining lifespan is >1000 cycles, maintain routine monitoring and perform basic checks every quarter.

[0068] Example 2 Please see Figure 2 As shown, based on the same inventive concept, this embodiment discloses a caliper brake life prediction system based on machine learning. For details not covered in this embodiment, please refer to the relevant sections of Embodiment 1. The system includes: The image correction module acquires a set of original images of the brake from multiple angles during its stationary and braking processes, corrects and enhances the original image set from multiple angles, and generates a corrected image set. The data analysis module performs pixel-level segmentation based on the calibration image set, calculates the wear rate, and obtains the wear trend curve. It also extracts texture features from the calibration image set, measures crack linear data, and then acquires infrared images. The infrared images are registered with the calibration image set to obtain hot spot marking maps. The parameter acquisition module collects the temperature, pressure, and vibration parameters of the brake during the braking process, preprocesses and aligns the three types of parameters, and obtains a multi-source dataset after alignment. The life prediction module combines wear rate, crack linear data, and hot spot marking maps to draw a three-dimensional feature map, calculate the damage matrix, generate a damage report, dynamically formulate life assessment indicators, and use the improved Paris formula and the damage report and life assessment indicators to derive a life prediction curve and generate maintenance recommendations.

[0069] Example 3 Please see Figure 3 As shown, this embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the machine learning-based caliper brake life prediction method provided by the above methods.

[0070] Since the electronic device described in this embodiment is the electronic device used to implement the machine learning-based caliper brake life prediction method described in this application embodiment, those skilled in the art can understand the specific implementation and various variations of the electronic device in this embodiment based on the machine learning-based caliper brake life prediction method described in this application embodiment. Therefore, how the electronic device implements the method in this application embodiment will not be described in detail here. Any electronic device used by those skilled in the art to implement the machine learning-based caliper brake life prediction method in this application embodiment falls within the scope of protection of this application.

[0071] Example 4 This embodiment discloses a computer-readable storage medium, including a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, it implements the machine learning-based caliper brake life prediction method provided by the above methods.

[0072] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters, weights, and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0073] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0074] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0075] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0076] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only one method, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0077] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0078] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0079] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0080] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the lifespan of a caliper brake based on machine learning, characterized in that, The method includes: S101: Acquire a multi-angle original image set of the brake during stationary and braking processes, correct and enhance the multi-angle original image set, and generate a corrected image set; S102: Pixel-level segmentation is performed based on the calibration image set to calculate the wear rate and obtain the wear trend curve; texture features are extracted from the calibration image set, crack linear data are measured, infrared images are acquired, and the infrared images are registered with the calibration image set to obtain the hot spot marking map. S103: Collect the temperature, pressure and vibration parameters of the brake during the braking process, preprocess and align the three types of parameters, and obtain a multi-source dataset after alignment; S104: Combine wear rate, crack linear data, and hot spot marking map to draw a three-dimensional feature map, calculate the damage matrix, generate a damage report, dynamically formulate life assessment indicators, use the improved Paris formula and integrate the damage report and life assessment indicators to derive a life prediction curve and generate maintenance recommendations.

2. The caliper brake life prediction method based on machine learning according to claim 1, characterized in that, The calculation of the wear rate and the resulting wear trend curve include: The corrected image set is output to the trained U-2-Net model, and the boundary segmentation accuracy is optimized by combining the side output supervision mechanism to obtain a binarized segmentation map. An adaptive window is applied to perform morphological gradient operations on the binarized segmentation image. The opening operation eliminates small noise in the binarized segmentation image, while the closing operation fills in the holes in the region, and the optimized segmentation image is output. The wear rate is obtained by calculating the ratio of the pixel area of ​​the worn area to the total pixel area of ​​the brake disc. The wear rate is then plotted by correlating the number of braking operations over time and drawing a dynamic curve between the wear rate and the number of braking operations. The non-linear trend is captured by using LOESS smoothing to obtain a wear trend curve.

3. The caliper brake life prediction method based on machine learning according to claim 1, characterized in that, The step of extracting texture features from the corrected image set and measuring crack linear data includes: Convolutional filtering is applied to the corrected image set to generate a texture feature map. Adaptive thresholding is used to identify the transverse and longitudinal crack orientations from the texture feature map, and a crack orientation map is output. The crack centerline was extracted and refined using the Zhang-Suen skeletonization algorithm. The length, width, and number of branches of the crack were measured to obtain crack region data. The fractal dimension of the crack region was calculated using the box-difference dimension method to quantify the surface roughness change and calculate the crack length growth rate, thus obtaining the crack linear data.

4. The caliper brake life prediction method based on machine learning according to claim 3, characterized in that, The process of registering the infrared image with the calibration image set to obtain a hot spot marker map includes: Feature points are extracted from the visible light and infrared images in the calibration image set, respectively. The extracted feature points are matched to obtain matching point pairs. Based on the matching point pairs, registration is performed to obtain registered image pairs. The edge contour of the brake disc is extracted from the registered image pair. The edge contour is used as the boundary condition for solving the heat conduction equation. The temperature field distribution on the surface of the brake disc is inverted to generate a temperature distribution map. Hot spot regions are extracted from the temperature distribution map. Adjacent hot spot regions are merged, and the hot spot position, area and temperature peak are calculated to obtain a hot spot marking map.

5. The caliper brake life prediction method based on machine learning according to claim 1, characterized in that, The specific steps for drawing a three-dimensional feature map by combining wear rate, crack linear data, and hot spot marking map are as follows: Using the number of braking cycles as the Z-axis, wear rate as the X-axis, crack length as the Y-axis, and hot spot area as the color code, a three-dimensional scatter plot is constructed and converted into a three-dimensional feature map.

6. The caliper brake life prediction method based on machine learning according to claim 5, characterized in that, The specific steps for dynamically formulating life assessment indicators are as follows: Wear threshold setting: The upper limit of wear rate is adjusted linearly based on the number of braking events. When the number of braking events is less than M, the yellow warning threshold is a and the red warning threshold is b. When the number of braking attempts is greater than or equal to M, the yellow warning threshold is h, and the red warning threshold is k. Crack threshold setting: Calculate the rate of change of the fractal dimension. If the rate of change is greater than n, adjust the crack critical value, compare the crack length with the adjusted crack critical value, and assess the failure risk. Thermal stress index setting: Calculate the real-time difference between the highest temperature of the brake disc and the boiling point of the brake fluid to assess the probability of thermal failure.

7. The caliper brake life prediction method based on machine learning according to claim 6, characterized in that, The process of deriving the lifetime prediction curve includes: Calculate the stress intensity factor based on crack length and pressure parameters. Then, by applying the improved Paris formula, the crack propagation behavior under the combined effects of temperature and stress is quantified, namely: ； In the formula: , Where is the temperature-stress coupling coefficient, Q is the activation energy, R is the gas constant, and T is the absolute temperature; By combining the number of braking cycles and integrating the remaining life, a life prediction curve is obtained.

8. A caliper brake life prediction system based on machine learning, characterized in that, It is implemented based on the machine learning-based caliper brake life prediction method according to any one of claims 1 to 7, and the system includes: The image correction module acquires a set of original images of the brake from multiple angles during its stationary and braking processes, corrects and enhances the original image set from multiple angles, and generates a corrected image set. The data analysis module performs pixel-level segmentation based on the calibration image set, calculates the wear rate, and obtains the wear trend curve. It also extracts texture features from the calibration image set, measures crack linear data, and then acquires infrared images. The infrared images are registered with the calibration image set to obtain hot spot marking maps. The parameter acquisition module collects the temperature, pressure, and vibration parameters of the brake during the braking process, preprocesses and aligns the three types of parameters, and obtains a multi-source dataset after alignment. The life prediction module combines wear rate, crack linear data, and hot spot marking maps to draw a three-dimensional feature map, calculate the damage matrix, generate a damage report, dynamically formulate life assessment indicators, and use the improved Paris formula and the damage report and life assessment indicators to derive a life prediction curve and generate maintenance recommendations.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the machine learning-based caliper brake life prediction method as described in claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the machine learning-based caliper brake life prediction method as described in claims 1-7.

Images