Online Prediction Method, Equipment, and Medium for Brake Disc Thermal Cracks Based on Micromorphological Transfer Learning

Abstract

This invention discloses an online prediction method, device, and medium for brake disc hot cracks based on microscopic morphology transfer learning, relating to the field of brake detection technology. The method, performed before braking, uses a miniature high-resolution microscopic imaging module inside the brake caliper to acquire images of a designated area selected through historical crack high-incidence zone analysis and roughness and stress concentration coefficient screening. These images are then processed to obtain a real-time microscopic morphology image sequence. This sequence is input into a pre-trained transfer learning model to extract features such as texture, contour, micro-pits, and grain boundary deformation. Combined with the evolution law of hot cracks under multiple operating conditions, the probability and time of hot crack occurrence are predicted. If the probability exceeds a threshold, warning information including a warning level and image thumbnail is generated and stored. The model is also periodically updated incrementally using historical data. The device includes a processor, memory, microscopic imaging, and vehicle communication module. The storage medium contains the corresponding program, achieving accurate online prediction and warning.

Description

Technical Field

[0001] This invention relates to the field of brake detection technology, and more specifically, to a method, device, and medium for online prediction of brake disc thermal cracks based on micromorphological transfer learning. Background Technology

[0002] As a core component of vehicle safety control, the braking system's key component, the brake disc, is subject to the alternating effects of braking friction heat cycles and mechanical stress during long-term service. This can easily lead to the formation and gradual expansion of thermal cracks on its surface, ultimately causing brake performance degradation or even structural failure, posing a significant threat to driving safety. Traditional brake disc health monitoring relies heavily on macroscopic condition detection, such as thickness wear and radial runout. However, the formation of thermal cracks originates from microscopic crystal structure damage and stress concentration. Macroscopic monitoring methods are unable to capture early signs of defects, resulting in significant prediction lag and failing to meet the needs of real-time safety warnings.

[0003] With the development of intelligent manufacturing and machine learning technologies, image-based microscopic monitoring methods are gradually being applied to the diagnosis of defects in mechanical components. However, the microscopic morphology analysis of brake discs faces multiple technical challenges: on the one hand, the working environment of brake discs is complex, and they are affected by oil, dust, and vibration. Microscopic image acquisition is prone to noise and distortion. Moreover, the evolution path of hot cracks under different braking conditions, such as high speed and heavy load, and low temperature and low speed, is significantly different, and the generalization ability of models trained under a single condition is insufficient. On the other hand, the cost of obtaining microscopic samples of hot cracks in brake discs is high. It is necessary to simulate the damage process under multiple conditions through bench tests. The number of samples is limited, and directly training deep learning models is prone to overfitting, which leads to a decrease in the accuracy of feature extraction and the reliability of prediction.

[0004] Transfer learning technology offers a new approach to solving the challenge of predicting microscopic defects under small sample sizes and multiple operating conditions. By transferring a feature extraction network pre-trained on a large-scale general image dataset to the microscopic morphology analysis scenario of brake discs, and only fine-tuning the deep parameters of the network and the prediction module, the convergence speed and feature learning ability of the model can be improved with limited sample size. At the same time, by combining the microscopic mechanism of brake disc thermal crack evolution, such as the correlation between micro-pit formation, grain boundary deformation and crack initiation, a prediction model integrating multi-dimensional microscopic features, such as texture, contour and defect precursors, can be constructed. This enables cross-scale correlation analysis from microscopic defect precursors to macroscopic thermal crack occurrence, providing technical support for online prediction of brake disc thermal cracks. In addition, by continuously updating the model parameters through an incremental learning mechanism, it can adapt to the performance degradation law changes during the long-term service of brake discs, further improving the dynamic adaptability and long-term reliability of the prediction model.

[0005] Therefore, existing technologies suffer from problems such as limited microscopic samples, poor generalization, difficulty in capturing early defects, and prediction lag without dynamic adaptation. Summary of the Invention

[0006] In order to overcome the problems of limited microscopic samples, poor generalization, difficulty in capturing early defects, and prediction lag without dynamic adaptation in existing technologies, this invention discloses an online prediction method, equipment and medium for brake disc hot cracks based on microscopic morphology transfer learning, which can effectively solve the above-mentioned technical problems.

[0007] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0008] The online prediction method for brake disc hot cracks based on micromorphological transfer learning includes the following steps:

[0009] Microscopic morphology images of a designated monitoring area on the surface of the brake disc are acquired to obtain a real-time microscopic morphology image sequence of the brake disc;

[0010] The real-time micro-morphology image sequence is input into the pre-trained transfer learning model. The transfer learning model is based on the micro-morphology sample set of brake disc hot cracks under multiple working conditions. It is obtained by fine-tuning the pre-trained image feature extraction network through the transfer learning algorithm. The transfer learning model includes a micro-morphology feature extraction module and a hot crack prediction module.

[0011] The micromorphology feature extraction module extracts features from the real-time micromorphology image sequence to obtain the texture features, contour features, and defect precursor features of the real-time micromorphology of the brake disc. The defect precursor features include micro-pit distribution features and grain boundary deformation features.

[0012] The hot crack prediction module predicts whether the brake disc will develop hot cracks based on the extracted real-time micromorphological features and the hot crack evolution law data under multiple working conditions stored in the transfer learning model, and outputs the hot crack prediction result, which includes the probability of hot crack occurrence and the predicted occurrence time.

[0013] If the probability of thermal cracking is greater than a preset probability threshold, a brake disc thermal cracking warning message is generated, and the real-time microscopic morphology image sequence and the thermal cracking prediction result are stored in the associated database.

[0014] Preferably, the obtained transfer learning model includes:

[0015] A sample set of microscopic morphology of brake disc thermal cracks under multiple working conditions was obtained. The sample set includes microscopic morphology images of brake discs from no cracks to the presence of thermal cracks of different degrees under different temperature loads and different braking frequencies. Each microscopic morphology image is labeled with a corresponding thermal crack state label.

[0016] The sample set is divided into a training set, a validation set, and a test set, wherein the training set is used for model parameter tuning, the validation set is used for model hyperparameter optimization, and the test set is used for model performance verification.

[0017] A pre-trained convolutional neural network is selected as the basic image feature extraction network, and the pre-trained convolutional neural network is trained on a large-scale general image dataset.

[0018] The transfer learning algorithm is used to freeze some of the shallow network parameters of the basic image feature extraction network, and the micro-morphology images in the training set are input into the basic image feature extraction network to train the deep network parameters and the parameters of the newly added thermal crack prediction module.

[0019] The model's performance during training is evaluated using the validation set, and the model's hyperparameters are adjusted until the model's prediction accuracy on the validation set reaches a preset accuracy threshold.

[0020] The adjusted model is then tested using the test set. If the test passes, the trained transfer learning model is obtained.

[0021] Preferably, obtaining the real-time microscopic morphology image sequence of the brake disc includes:

[0022] When the braking system is not in braking operation, the miniature high-resolution microscopic imaging module installed inside the brake caliper is controlled to move to the designated monitoring area on the surface of the brake disc.

[0023] The miniature high-resolution microscopic imaging module continuously acquires images of the designated monitoring area at a preset acquisition frequency, and simultaneously records the brake disc temperature data and braking condition parameters at each acquisition time.

[0024] The acquired original microscopic morphology images are subjected to noise reduction, distortion correction, and image alignment processing to obtain a real-time microscopic morphology image sequence of the brake disc.

[0025] Preferably, the method further includes:

[0026] Historical micromorphological image sequences and corresponding thermal crack prediction results are periodically extracted from the associated database;

[0027] The extracted historical data is input into the transfer learning model according to a preset update cycle, and the parameters of the hot crack prediction module of the transfer learning model are incrementally updated.

[0028] The updated transfer learning model is used to predict thermal cracks in newly acquired real-time microscopic morphology image sequences. The accuracy of the prediction results before and after the update is compared. If the improvement in accuracy is less than the preset improvement threshold, the incremental update strategy is adjusted.

[0029] Preferably, the method for determining the designated monitoring area includes:

[0030] Analyze the distribution data of high-incidence areas of thermal cracks in brake discs during historical use to identify several potential high-risk areas;

[0031] Preliminary microscopic morphology detection was performed on each potentially high-risk area, and the surface roughness and stress concentration factor of each area were calculated.

[0032] The area where both surface roughness and stress concentration coefficient exceed the corresponding preset threshold is determined as the designated monitoring area. If there are multiple areas that meet the conditions, they are sorted from high to low according to the risk coefficient, and the first N areas are selected as the designated monitoring areas, where N is an integer greater than or equal to 1.

[0033] Preferably, the hot crack prediction module predicts whether the brake disc will develop hot cracks based on the extracted real-time microscopic morphology features and the hot crack evolution data under multiple working conditions stored in the transfer learning model, including:

[0034] The similarity between the extracted real-time micromorphological features and the standard micromorphological features of different hot crack stages under multiple working conditions stored in the transfer learning model is calculated.

[0035] The current micromorphological evolution stage of the brake disc is determined based on the similarity calculation results;

[0036] By combining the current braking condition parameters and brake disc temperature data, and calling the corresponding hot crack evolution law data in the transfer learning model, a hot crack development prediction model is established.

[0037] The probability of hot cracks occurring in the brake disc and the predicted occurrence time are calculated using the hot crack development prediction model, and the result is used as the hot crack prediction result.

[0038] Preferably, the generation of brake disc thermal crack early warning information includes:

[0039] Based on the probability of hot cracking and the predicted occurrence time, an early warning level is determined, wherein the early warning level includes a first-level early warning, a second-level early warning and a third-level early warning, and the early warning level increases as the probability of hot cracking increases and the predicted occurrence time decreases.

[0040] According to the preset warning information format, the warning level, the probability of hot crack occurrence, the predicted occurrence time and the corresponding real-time microscopic morphology image thumbnail are integrated into the brake disc hot crack warning information.

[0041] The brake disc thermal crack warning information is sent to the vehicle control system and driver terminal via the vehicle communication module, and a warning prompt is displayed on the vehicle instrument panel at the same time.

[0042] Preferably, an online prediction device for brake disc hot cracks based on micromorphological transfer learning includes a processor, an input device, an output device, and a memory. The processor, the input device, the output device, and the memory are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to call the program instructions to execute the prediction method described above.

[0043] Preferably, it also includes a miniature high-resolution microscopic imaging module and an on-board communication module. The miniature high-resolution microscopic imaging module is installed inside the brake caliper and is used to acquire microscopic morphology images of a designated monitoring area on the surface of the brake disc and transmit the acquired image data to the processor. The on-board communication module is connected to the processor and is used to send the brake disc thermal crack warning information generated by the processor to the vehicle control system and the driver terminal.

[0044] Preferably, a computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the prediction method as described above.

[0045] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention addresses the problem of poor generalization due to the scarcity of microscopic samples. Traditional models rely on training with specific microscopic samples, but the cost and quantity of collecting crack samples from brake discs under various operating conditions are high, resulting in models being unable to adapt to different braking scenarios. The solution constructs a microscopic morphology sample set of brake disc thermal cracks under various operating conditions (covering images of all stages from no cracks to different degrees of cracks under different temperature loads and braking frequencies), and uses a transfer learning algorithm to fine-tune a convolutional neural network pre-trained on a large-scale general image dataset—freezing the shallow general feature extraction parameters of the network (to avoid parameter shifts due to insufficient samples), and training only the deep parameters related to the microscopic features of the brake disc and the newly added thermal crack parameters. The crack prediction module parameters leverage the mature feature extraction capabilities of pre-trained networks to reduce reliance on a limited number of specific samples. Furthermore, by covering the entire crack evolution scenario with multi-condition samples, the model can accurately adapt to different braking conditions such as high-speed heavy loads and low-temperature low speeds, improving generalization ability and solving the problem of poor adaptability caused by the limited sample size in traditional models. Addressing the difficulty in capturing early defects, traditional methods often focus on macroscopic wear or visible cracks. However, thermal crack initiation begins with precursors such as grain boundary deformation and micro-pit aggregation at the microscale, which are difficult to capture with conventional detection. This solution uses a micromorphological feature extraction module to specifically extract texture features, contour features, and defect precursor features such as micro-pit distribution and grain boundary deformation. Simultaneously, it filters by specifying a monitoring area (first dividing...). By analyzing historical hot cracking high-incidence areas to identify potential high-risk regions, and then calculating surface roughness and stress concentration coefficients to screen out areas with dual-parameter exceeding thresholds, precise monitoring of microstructural changes in high-risk areas is achieved. This allows for the capture of microscopic deterioration signs before crack formation, avoiding the problem of missed early defects caused by neglecting microscopic precursors in traditional methods. Addressing the issue of predictive lag and lack of dynamic adaptation, traditional detection methods are mostly offline and periodic, unable to reflect the brake disc status in real time. Furthermore, the parameters are fixed after model training, making it difficult to adapt to the performance degradation patterns of brake discs during long-term service. This solution utilizes a miniature high-resolution microscopic imaging module (installed inside the brake caliper, without affecting braking operation) to monitor the microstructure changes in high-risk areas during non-braking states. Images are continuously acquired at a preset frequency, and after noise reduction and correction, a real-time microscopic morphology image sequence is formed. Combined with the multi-condition thermal crack evolution data stored in the transfer learning model, the current evolution stage is determined through similarity calculation. Then, a prediction model is established by combining real-time temperature and operating parameters to output the probability of thermal crack occurrence and prediction time, realizing real-time monitoring and early prediction and solving the lag problem. On the other hand, by periodically extracting historical data from the associated database to incrementally update the parameters of the model prediction module, and comparing the accuracy before and after the update to adjust the update strategy, the model can be continuously optimized with the accumulation of long-term service data of the brake disc, dynamically adapting to its performance degradation law, avoiding the problem of prediction accuracy decay caused by the inability to iterate in traditional fixed models. Attached Figure Description

[0046] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other embodiments can be derived from the provided drawings without creative effort.

[0047] Figure 1 This is a diagram illustrating the steps of the method of the present invention. Detailed Implementation

[0048] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.

[0049] To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions.

[0050] It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.

[0051] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0052] Example

[0053] Please see Figure 1 The online prediction method for brake disc hot cracks based on micromorphological transfer learning includes the following steps:

[0054] Microscopic morphology images of a designated monitoring area on the surface of the brake disc are acquired to obtain a real-time microscopic morphology image sequence of the brake disc;

[0055] The real-time micro-morphology image sequence is input into the pre-trained transfer learning model. The transfer learning model is based on the micro-morphology sample set of brake disc hot cracks under multiple working conditions. It is obtained by fine-tuning the pre-trained image feature extraction network through the transfer learning algorithm. The transfer learning model includes a micro-morphology feature extraction module and a hot crack prediction module.

[0056] The micromorphology feature extraction module extracts features from the real-time micromorphology image sequence to obtain the texture features, contour features, and defect precursor features of the real-time micromorphology of the brake disc. The defect precursor features include micro-pit distribution features and grain boundary deformation features.

[0057] The hot crack prediction module predicts whether the brake disc will develop hot cracks based on the extracted real-time micromorphological features and the hot crack evolution law data under multiple working conditions stored in the transfer learning model, and outputs the hot crack prediction result, which includes the probability of hot crack occurrence and the predicted occurrence time.

[0058] If the probability of thermal cracking is greater than a preset probability threshold, a brake disc thermal cracking warning message is generated, and the real-time microscopic morphology image sequence and the thermal cracking prediction result are stored in the associated database.

[0059] The obtained transfer learning model includes:

[0060] A sample set of microscopic morphology of brake disc thermal cracks under multiple working conditions was obtained. The sample set includes microscopic morphology images of brake discs from no cracks to the presence of thermal cracks of different degrees under different temperature loads and different braking frequencies. Each microscopic morphology image is labeled with a corresponding thermal crack state label.

[0061] The sample set is divided into a training set, a validation set, and a test set, wherein the training set is used for model parameter tuning, the validation set is used for model hyperparameter optimization, and the test set is used for model performance verification.

[0062] A pre-trained convolutional neural network is selected as the basic image feature extraction network, and the pre-trained convolutional neural network is trained on a large-scale general image dataset.

[0063] The transfer learning algorithm is used to freeze some of the shallow network parameters of the basic image feature extraction network, and the micro-morphology images in the training set are input into the basic image feature extraction network to train the deep network parameters and the parameters of the newly added thermal crack prediction module.

[0064] The model's performance during training is evaluated using the validation set, and the model's hyperparameters are adjusted until the model's prediction accuracy on the validation set reaches a preset accuracy threshold.

[0065] The adjusted model is then tested using the test set. If the test passes, the trained transfer learning model is obtained.

[0066] The real-time microscopic morphology image sequence of the brake disc includes:

[0067] When the braking system is not in braking operation, the miniature high-resolution microscopic imaging module installed inside the brake caliper is controlled to move to the designated monitoring area on the surface of the brake disc.

[0068] The miniature high-resolution microscopic imaging module continuously acquires images of the designated monitoring area at a preset acquisition frequency, and simultaneously records the brake disc temperature data and braking condition parameters at each acquisition time.

[0069] The acquired original microscopic morphology images are subjected to noise reduction, distortion correction, and image alignment processing to obtain a real-time microscopic morphology image sequence of the brake disc.

[0070] The method further includes:

[0071] Historical micromorphological image sequences and corresponding thermal crack prediction results are periodically extracted from the associated database;

[0072] The extracted historical data is input into the transfer learning model according to a preset update cycle, and the parameters of the hot crack prediction module of the transfer learning model are incrementally updated.

[0073] The updated transfer learning model is used to predict thermal cracks in newly acquired real-time microscopic morphology image sequences. The accuracy of the prediction results before and after the update is compared. If the improvement in accuracy is less than the preset improvement threshold, the incremental update strategy is adjusted.

[0074] The method for determining the designated monitoring area includes:

[0075] Analyze the distribution data of high-incidence areas of thermal cracks in brake discs during historical use to identify several potential high-risk areas;

[0076] Preliminary microscopic morphology detection was performed on each potentially high-risk area, and the surface roughness and stress concentration factor of each area were calculated.

[0077] The area where both surface roughness and stress concentration coefficient exceed the corresponding preset threshold is determined as the designated monitoring area. If there are multiple areas that meet the conditions, they are sorted from high to low according to the risk coefficient, and the first N areas are selected as the designated monitoring areas, where N is an integer greater than or equal to 1.

[0078] The hot crack prediction module predicts whether the brake disc will develop hot cracks based on the extracted real-time micromorphological features and the hot crack evolution data under multiple working conditions stored in the transfer learning model.

[0079] The similarity between the extracted real-time micromorphological features and the standard micromorphological features of different hot crack stages under multiple working conditions stored in the transfer learning model is calculated.

[0080] The current micromorphological evolution stage of the brake disc is determined based on the similarity calculation results;

[0081] By combining the current braking condition parameters and brake disc temperature data, and calling the corresponding hot crack evolution law data in the transfer learning model, a hot crack development prediction model is established.

[0082] The probability of hot cracks occurring in the brake disc and the predicted occurrence time are calculated using the hot crack development prediction model, and the result is used as the hot crack prediction result.

[0083] The generated brake disc thermal crack early warning information includes:

[0084] Based on the probability of hot cracking and the predicted occurrence time, an early warning level is determined, wherein the early warning level includes a first-level early warning, a second-level early warning and a third-level early warning, and the early warning level increases as the probability of hot cracking increases and the predicted occurrence time decreases.

[0085] According to the preset warning information format, the warning level, the probability of hot crack occurrence, the predicted occurrence time and the corresponding real-time microscopic morphology image thumbnail are integrated into the brake disc hot crack warning information.

[0086] The brake disc thermal crack warning information is sent to the vehicle control system and driver terminal via the vehicle communication module, and a warning prompt is displayed on the vehicle instrument panel at the same time.

[0087] An online prediction device for brake disc hot cracks based on micromorphological transfer learning includes a processor, an input device, an output device, and a memory. The processor, the input device, the output device, and the memory are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to call the program instructions to execute the prediction method described above.

[0088] It also includes a miniature high-resolution microscopic imaging module and an on-board communication module. The miniature high-resolution microscopic imaging module is installed inside the brake caliper and is used to acquire microscopic morphology images of a designated monitoring area on the surface of the brake disc and transmit the acquired image data to the processor. The on-board communication module is connected to the processor and is used to send the brake disc thermal crack warning information generated by the processor to the vehicle control system and the driver terminal.

[0089] A computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the prediction method as described above.

[0090] Brake disc thermal cracking is a significant factor affecting vehicle braking performance and driving safety. Traditional methods for detecting thermal cracks primarily rely on periodic inspections of macroscopic cracks on the brake disc surface. However, this approach is time-consuming and struggles to detect cracks in their early stages. With advancements in computer vision and transfer learning technologies, online prediction of thermal cracks based on microscopic morphology has become possible. This paper details an online prediction method and device for brake disc thermal cracks based on microscopic morphology transfer learning, aiming to achieve early and accurate prediction of brake disc thermal cracks and improve the reliability and safety of automotive braking systems.

[0091] First, we analyzed the distribution data of high-incidence areas of hot cracks in brake discs during historical use. By collecting a large amount of brake disc usage data, including the location of hot cracks under different vehicle models, road conditions, and braking frequencies, we determined the high-incidence areas of hot cracks. For example, for a common family car brake disc, data analysis revealed that hot cracks mostly occurred in the outer edge area and the middle part near the brake caliper contact surface of the brake disc.

[0092] Preliminary microscopic morphology inspection is performed on each potentially high-risk area using portable microscopes and other equipment. The surface roughness and stress concentration factor of each area are calculated. Surface roughness can be determined by measuring the surface micro-irregularity, while the stress concentration factor can be calculated using methods such as finite element analysis. For example, a preset threshold of 0.5 μm for surface roughness and 1.5 for stress concentration factor are set. Areas where both surface roughness and stress concentration factor exceed the corresponding preset thresholds are identified as designated monitoring areas. If multiple areas meet the criteria, they are sorted from high to low risk factor, and the top N (N≥1) areas are selected as designated monitoring areas. The risk factor can be calculated by comprehensively considering factors such as surface roughness, stress concentration factor, and the frequency of hot cracking in the area in historical data.

[0093] When the braking system is not in braking operation, the miniature high-resolution microscopic imaging module installed inside the brake caliper is moved to the designated monitoring area on the surface of the brake disc. This miniature high-resolution microscopic imaging module has the characteristics of high resolution and high contrast, and can clearly acquire microscopic morphology images of the brake disc surface. For example, its resolution can reach 1μm or even higher, and it can capture microscopic features such as micro-pits and grain boundaries.

[0094] The miniature high-resolution microscopic imaging module continuously acquires images of a designated monitoring area at a preset acquisition frequency. This frequency can be set based on the brake disc's condition and the rate of thermal crack formation. For example, for vehicles that frequently travel at high speeds and brake frequently, the acquisition frequency can be set to acquire images every 10 minutes, while simultaneously recording brake disc temperature data and braking condition parameters at each acquisition. Brake disc temperature can be measured using temperature sensors such as thermocouples installed near the brake disc. Braking condition parameters, including braking frequency and braking intensity, can be obtained through the vehicle's control system.

[0095] The acquired raw microscopic morphology images are subjected to noise reduction, distortion correction, and image alignment. Due to potential noise interference, lens distortion, and inconsistent image shooting angles during the acquisition process, image preprocessing is required. Noise reduction can be achieved by removing random noise from the image through methods such as median filtering and Gaussian filtering. Distortion correction can be performed using a lens distortion model. Image alignment is achieved by aligning consecutively acquired images through methods such as feature point matching to ensure the consistency of the image sequence, thereby obtaining a real-time microscopic morphology image sequence of the brake disc.

[0096] Construction of transfer learning models

[0097] Sample Set Acquisition and Division: A sample set of microscopic morphology images of brake disc thermal cracks under multiple operating conditions was acquired. This sample set includes microscopic morphology images of brake discs from no cracks to varying degrees of thermal cracks under different temperature loads, such as 100℃, 200℃, and 300℃, and different braking frequencies, such as 100, 200, and 300 braking cycles per hour. Each microscopic morphology image is labeled with a corresponding thermal crack state label, such as no crack, slight crack, and severe crack. The sample set is divided into a training set, a validation set, and a test set, for example, in a 6:2:2 ratio. The training set is used for model parameter adjustment, the validation set is used for model hyperparameter optimization, and the test set is used for model performance verification.

[0098] Basic Image Feature Extraction Network Selection and Training: Pre-trained convolutional neural networks, such as ResNet and VGG, are selected as the basic image feature extraction networks. These pre-trained convolutional neural networks are trained on large-scale general image datasets, such as ImageNet, and have powerful image feature extraction capabilities. A transfer learning algorithm is used to freeze some shallow network parameters of the basic image feature extraction network. The shallow network parameters are mainly responsible for extracting low-level features of the image, such as edges and textures. These features have certain universality in different types of images, so they can be frozen. Microscopic morphology images from the training set are input into the basic image feature extraction network to train the deep network parameters and the parameters of the newly added thermal crack prediction module. The deep parameters are responsible for extracting high-level features of the image, such as the shape and structure of objects. These features are more critical for the prediction of thermal cracks. The newly added thermal crack prediction module can predict thermal cracks based on the extracted features.

[0099] Model performance evaluation and optimization: The model's performance during training is evaluated using the validation set. Evaluation metrics may include prediction accuracy, recall, F1 score, etc. Based on the evaluation results, the model's hyperparameters, such as learning rate and batch size, are adjusted until the model's prediction accuracy on the validation set reaches a preset accuracy threshold, for example, a preset accuracy threshold of 90%. The adjusted model is then tested using the test set. If the test passes, the trained transfer learning model is obtained.

[0100] The real-time micro-morphology image sequence is input into a pre-trained transfer learning model. The micro-morphology feature extraction module in the transfer learning model extracts features from the real-time micro-morphology image sequence. This module can extract texture features, contour features, and defect precursor features of the real-time micro-morphology of the brake disc. Texture features reflect the arrangement pattern of the microstructure on the surface of the brake disc; contour features describe the shape features of the surface of the brake disc; defect precursor features include micro-pit distribution features, grain boundary deformation features, etc. These features are early signs of hot crack formation. For example, the dense distribution of micro-pits may indicate the initiation of hot cracks, and the deformation of grain boundaries may indicate abnormal internal stress of the material. These features are crucial for the prediction of hot cracks.

[0101] The hot crack prediction module predicts whether the brake disc will develop hot cracks based on the extracted real-time microscopic morphology features and the hot crack evolution data under multiple working conditions stored in the transfer learning model. The specific process is as follows:

[0102] The similarity between the extracted real-time micro-morphological features and the standard micro-morphological features of different hot cracking stages under multiple working conditions stored in the transfer learning model is calculated. For example, the degree of similarity between the real-time micro-morphology and the standard micro-morphological features is determined by calculating the cosine similarity between feature vectors.

[0103] The current micromorphological evolution stage of the brake disc is determined based on the similarity calculation results. If the real-time micromorphology has the highest similarity to the standard micromorphological features of the minor crack stage, then the brake disc is judged to be in the minor crack stage.

[0104] By combining current braking parameters, such as braking frequency, braking intensity, and brake disc temperature data, and calling the corresponding hot crack evolution law data in the transfer learning model, a hot crack development prediction model is established. This model can predict the future development trend of hot cracks based on the current operating conditions and temperature conditions.

[0105] The probability of hot cracks occurring in the brake disc within a preset time period and the predicted occurrence time are calculated using a hot crack development prediction model. This is used as the hot crack prediction result. For example, the probability of hot cracks occurring in the brake disc is predicted to be 80% within the next 100 hours, and the predicted occurrence time is 85 hours later.

[0106] If the probability of hot cracking is greater than a preset probability threshold, such as 70%, a brake disc hot cracking warning is generated. The warning level is determined based on the probability of hot cracking and the predicted occurrence time. For example, when the probability of hot cracking is greater than 90% and the predicted occurrence time is less than 50 hours, it is a Level 1 warning; when the probability of hot cracking is between 70% and 90% and the predicted occurrence time is between 50 and 100 hours, it is a Level 2 warning; and in other cases, it is a Level 3 warning. The warning level increases as the probability of hot cracking increases and the predicted occurrence time decreases.

[0107] According to the preset warning information format, the warning level, the probability of hot cracking, the predicted occurrence time, and the corresponding real-time microscopic morphology image thumbnail are integrated into the brake disc hot crack warning information. For example, the warning information format can be: warning level: Level 1 warning; probability of hot cracking: 95%; predicted occurrence time: after 48 hours; real-time microscopic morphology image thumbnail.

[0108] The brake disc thermal crack warning information is sent to the vehicle control system and driver terminal via the vehicle communication module. The vehicle control system can take corresponding measures based on the warning information, such as reminding the driver to pay attention to the brake disc condition and limiting the vehicle speed. The driver terminal can be the vehicle's central control screen or the driver's mobile device, which can promptly remind the driver of potential problems with the brake disc and display warning prompts on the vehicle's instrument panel, such as displaying a brake disc thermal crack warning and suggesting that it be checked as soon as possible.

[0109] Historical micromorphological image sequences and corresponding thermal crack prediction results are periodically extracted from the associated database. The extracted historical data is then input into the transfer learning model according to a preset update cycle, such as every 3 months. The parameters of the thermal crack prediction module of the transfer learning model are incrementally updated. The updated transfer learning model is then used to predict thermal cracks in newly acquired real-time micromorphological image sequences. The accuracy of the prediction results before and after the update is compared. If the improvement in accuracy is less than a preset improvement threshold, the incremental update strategy is adjusted. For example, the number of training samples can be increased, or the hyperparameters of the model can be adjusted to improve the model's prediction performance.

[0110] Brake disc hot crack online prediction equipment

[0111] The processor is used to perform various computing tasks, including image processing, feature extraction, model training, and thermal crack prediction. It can be a high-performance central processing unit (CPU) or a graphics processing unit (GPU), such as NVIDIA GPUs, which have powerful parallel computing capabilities and can quickly process large amounts of image data and complex model calculations.

[0112] The input device is used to input the collected microscopic morphology image data of the brake disc and other relevant information, such as brake disc temperature data and braking condition parameters, into the device. The main input devices include a miniature high-resolution microscopic imaging module and various sensors, such as temperature sensors and braking frequency sensors. The miniature high-resolution microscopic imaging module is installed inside the brake caliper and can acquire microscopic morphology images of a specified monitoring area on the surface of the brake disc in real time and transmit the acquired image data to the processor. Temperature sensors and braking frequency sensors are used to collect the temperature and braking condition parameters of the brake disc, which are very important for the prediction of thermal cracks.

[0113] The output devices are used to output the prediction results and warning information to the vehicle control system, driver terminal, etc. The main output devices include the vehicle communication module and the vehicle instrument panel. The vehicle communication module is connected to the processor and can send the brake disc hot crack warning information generated by the processor to the vehicle control system and driver terminal. The vehicle instrument panel is used to display the warning prompts and remind the driver of possible problems with the brake disc.

[0114] The memory is used to store computer programs, model parameters, acquired image data, historical data, etc. The computer program includes program instructions, and the processor is configured to call these program instructions to execute the above prediction method. The memory can be a hard disk, solid-state drive, etc., with a large storage capacity, capable of storing a large amount of data.

[0115] The miniature high-resolution microscopic imaging module is a key device for image acquisition. It is installed inside the brake caliper and features high resolution, high contrast, and miniaturization. It can acquire microscopic images of the surface of the brake disc. The module can move to the designated monitoring area according to the control signal and continuously acquire images according to the preset acquisition frequency. The acquired image data is transmitted to the processor for subsequent processing through the data transmission line.

[0116] The vehicle communication module serves as a bridge for communication between the device and the vehicle control system and driver terminal. It supports multiple communication protocols, such as CAN bus and Ethernet, and can quickly and accurately send warning information to the vehicle control system and driver terminal. At the same time, it can also receive feedback information from the vehicle control system and driver terminal, realizing two-way communication between the device and the vehicle system.

[0117] Equipment Workflow: When the vehicle is in a non-braking state, the equipment controls the miniature high-resolution microscopic imaging module to move to the designated monitoring area on the surface of the brake disc. The microscopic imaging module begins to acquire microscopic morphology images of the brake disc surface according to the preset acquisition frequency, and simultaneously records the brake disc temperature data and braking condition parameters. The acquired raw microscopic morphology images are transmitted to the processor through the data transmission line. After receiving the image data, the processor first performs noise reduction, distortion correction and image alignment processing on the image to obtain a real-time microscopic morphology image sequence of the brake disc.

[0118] The processor inputs the real-time micro-morphology image sequence into the pre-trained transfer learning model. The micro-morphology feature extraction module in the transfer learning model extracts features from the image sequence, extracting the texture features, contour features, and defect precursor features of the real-time micro-morphology of the brake disc. The hot crack prediction module, based on the extracted features and combined with the hot crack evolution law data stored in the model, predicts whether the brake disc will develop hot cracks and outputs the hot crack prediction results, including the probability of hot crack occurrence and the predicted occurrence time.

[0119] If the probability of hot cracking is greater than the preset probability threshold, the processor determines the warning level based on the probability of hot cracking and the predicted occurrence time. Then, according to the preset warning information format, the warning level, the probability of hot cracking, the predicted occurrence time, and the corresponding real-time microscopic morphology image thumbnail are integrated into brake disc hot cracking warning information. The warning information is then sent to the vehicle control system and the driver terminal through the vehicle communication module, and the warning prompt is displayed on the vehicle instrument panel.

[0120] The processor periodically extracts historical micromorphological image sequences and corresponding thermal crack prediction results from the associated database. It inputs these historical data into the transfer learning model according to a preset update cycle, incrementally updates the parameters of the thermal crack prediction module of the model, and uses the updated model to predict thermal cracks in newly acquired real-time micromorphological image sequences. The accuracy of the prediction results before and after the update is compared. If the accuracy improvement is less than a preset improvement threshold, the processor adjusts the incremental update strategy to optimize the model performance.

[0121] The computer-readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, the processor performs the above-mentioned online prediction method for brake disc thermal cracks. The computer-readable storage medium can be an optical disc, a hard disk, a USB flash drive, etc., which facilitates the storage and dissemination of the program. By installing the computer program into the online prediction device for brake disc thermal cracks, the function of the device can be realized, and the automation and accuracy of brake disc thermal crack prediction can be improved.

[0122] This embodiment acquires microscopic morphology images of a designated monitoring area on the surface of the brake disc, and uses a transfer learning model to extract features and predict thermal cracks in the images. This enables early and accurate prediction of brake disc thermal cracks. The device includes a processor, input devices, output devices, a memory, a miniature high-resolution microscopic imaging module, and an in-vehicle communication module. It can realize functions such as image acquisition, processing, prediction, and early warning information output. By regularly updating the transfer learning model, the predictive performance of the model can be further improved. The application of this method and device is of great significance for improving the reliability and safety of automotive braking systems, and is expected to reduce traffic accidents caused by brake disc thermal cracks, thus protecting people's lives and property.

[0123] The same or similar labels correspond to the same or similar parts;

[0124] The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0125] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all implementation methods here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A brake disc thermal crack online prediction method based on micro-topography transfer learning, characterized in that, The method comprises the following steps: Collecting micro-morphology images of a specified monitoring area on the surface of a brake disc to obtain a real-time micro-morphology image sequence of the brake disc; Inputting the real-time micro-morphology image sequence into a pre-trained transfer learning model, wherein the transfer learning model is obtained by fine-tuning a pre-trained image feature extraction network based on a brake disc thermal crack micro-morphology sample set under multiple working conditions through a transfer learning algorithm, and the transfer learning model comprises a micro-morphology feature extraction module and a thermal crack prediction module; Extracting features of the real-time micro-morphology image sequence through the micro-morphology feature extraction module to obtain texture features, contour features and defect precursor features of the real-time micro-morphology of the brake disc, wherein the defect precursor features include micro-pit distribution features and grain boundary deformation features; The thermal crack prediction module predicts whether the brake disc will produce thermal cracks according to the extracted real-time micro-morphology features and the thermal crack evolution law data stored in the transfer learning model under multiple working conditions, and outputs a thermal crack prediction result, wherein the thermal crack prediction result includes a thermal crack occurrence probability and a predicted occurrence time; If the thermal crack occurrence probability is greater than a preset probability threshold, generating brake disc thermal crack warning information, and storing the real-time micro-morphology image sequence and the thermal crack prediction result in an associated database; The method for obtaining the real-time micro-morphology image sequence of the brake disc comprises the following steps: When the brake system is in a non-braking working state, moving a miniature high-resolution microscopic imaging module installed on the inner side of a brake caliper to a specified monitoring area on the surface of a brake disc; The miniature high-resolution microscopic imaging module continuously collects images of the specified monitoring area according to a preset collection frequency, and records brake disc temperature data and brake working condition parameters at each collection time; Performing noise reduction, distortion correction and image alignment processing on the collected original micro-morphology images to obtain the real-time micro-morphology image sequence of the brake disc; The method for determining the specified monitoring area comprises the following steps: Analyzing the distribution data of high-risk areas of thermal cracks of the brake disc in the historical use process to determine multiple potential high-risk areas; Performing preliminary micro-morphology detection on each potential high-risk area to calculate the surface roughness and stress concentration coefficient of each area; Determining the areas with surface roughness and stress concentration coefficient both exceeding the corresponding preset threshold as the specified monitoring area, and if there are multiple areas meeting the conditions, ranking the risk coefficients from high to low, and selecting the top N areas as the specified monitoring area, wherein N is an integer greater than or equal to 1.

2. The prediction method of claim 1, wherein, The method for obtaining the transfer learning model comprises the following steps: Obtaining a brake disc thermal crack micro-morphology sample set under multiple working conditions, wherein the sample set comprises micro-morphology images of the brake disc from no cracks to different degrees of thermal cracks under different temperature loads and different braking frequencies, and each micro-morphology image is labeled with a corresponding thermal crack state label; Dividing the sample set into a training set, a validation set and a test set, wherein the training set is used for model parameter adjustment, the validation set is used for model hyperparameter optimization, and the test set is used for model performance verification; The pre-trained convolutional neural network is selected as a basic image feature extraction network, and the pre-trained convolutional neural network is trained on a large-scale general image dataset; A transfer learning algorithm is used to freeze part of the shallow network parameters of the basic image feature extraction network, input the micro-morphology images in the training set into the basic image feature extraction network, and train the deep network parameters and the newly added hot crack prediction module parameters; The performance of the model in the training process is evaluated by using the verification set, and the model hyperparameters are adjusted until the prediction accuracy of the model on the verification set reaches a preset accuracy threshold; The final performance test of the adjusted model is performed using the test set, and if the test passes, a trained transfer learning model is obtained.

3. The prediction method of claim 1, wherein, The method further comprises: Periodically extracting historical micro-morphology image sequences and corresponding hot crack prediction result data from the association database; The extracted historical data is input into the transfer learning model according to a preset update period, and the hot crack prediction module parameters of the transfer learning model are incrementally updated; The real-time micro-morphology image sequences newly collected are predicted by the updated transfer learning model, and the prediction result accuracy before and after the update is compared, and if the accuracy improvement is less than a preset improvement threshold, the incremental update strategy is adjusted.

4. The prediction method of claim 1, wherein, The hot crack prediction module predicts whether the brake disc will produce hot cracks according to the extracted real-time micro-morphology features and the hot crack evolution law data stored in the transfer learning model under multiple working conditions, comprising: Calculating the similarity between the extracted real-time micro-morphology features and the standard micro-morphology features of different hot crack stages under multiple working conditions stored in the transfer learning model; Determine the current micro-morphology evolution stage of the brake disc according to the similarity calculation result; Combine the current braking condition parameters and brake disc temperature data to call the corresponding hot crack evolution law data in the transfer learning model to establish a hot crack development prediction model; The hot crack development prediction model is used to calculate the hot crack occurrence probability and predicted occurrence time of the brake disc in a future preset time period as the hot crack prediction result.

5. The prediction method of claim 1, wherein, The generation of brake disc hot crack warning information comprises: According to the hot crack occurrence probability and the predicted occurrence time, determine the warning level, wherein the warning level includes first warning, second warning and third warning, and the warning level increases with the increase of hot crack occurrence probability and the shortening of predicted occurrence time; According to a preset warning information format, the warning level, hot crack occurrence probability, predicted occurrence time and corresponding real-time micro-morphology image thumbnail are integrated into brake disc hot crack warning information; The brake disc hot crack warning information is sent to the vehicle control system and the driver terminal through the vehicle communication module, and the warning prompt is displayed on the vehicle instrument panel.

6. A brake disc thermal crack online prediction device based on micro-topography transfer learning, characterized in that, The application further discloses a computer readable storage medium, which stores a computer program, and the computer program comprises program instructions.

7. The apparatus of claim 6, wherein, The application further discloses a computer readable storage medium, which stores a computer program, and the computer program comprises program instructions.

8. A computer-readable storage medium, characterized in that, The application further discloses a computer readable storage medium, which stores a computer program, and the computer program comprises program instructions.

Images