Brake disc surface wear degree detection method

Abstract

The invention relates to the technical field of automobile part detection, and discloses a brake disc surface wear degree detection method, which comprises the following steps of: acquiring hyperspectral images of brake disc surfaces with different wear degrees by using a hyperspectral camera, and preprocessing the acquired hyperspectral images; according to the method, preprocessing and spectral feature extraction are performed on a hyperspectral image, the extracted features are fused to form a fusion vector, multi-dimensional spectral information is effectively integrated, a data set is rich in information and clear in structure, learning and processing of a subsequent model are facilitated, and through reasonable division of a training set, a test set and a verification set, the multi-dimensional spectral information is obtained. According to the method, the generalization ability of the model and the evaluation accuracy are ensured, automatic detection and classification of the wear degree of the brake disc are realized, the detection efficiency and accuracy are greatly improved, and intelligent processing and analysis of complex spectral information are realized through a deep learning technology of the convolutional neural network.

Description

Technical Field

[0001] The present invention relates to the technical field of automotive part detection, and particularly to a method for detecting the surface wear degree of a brake disc. Background Technique

[0002] With the rapid development of modern industrial technology, the brake disc, as a key safety component in vehicles such as automobiles, the stability and reliability of its performance are directly related to the driving safety of the vehicle. During the long-term use of the brake disc, due to friction and wear, the surface characteristics will change, and these changes directly affect the braking effect and the vehicle braking performance. Therefore, the accurate detection of the wear degree of the brake disc is particularly important. Traditionally, the detection of the wear degree of the brake disc mainly relies on manual visual inspection or simple physical measurement. These methods are not only time-consuming and laborious, but also easily affected by human factors, resulting in difficulties in ensuring the accuracy and consistency of the detection results. In recent years, with the continuous progress of optical technology and image processing technology, hyperspectral imaging technology has gradually been applied to material detection and surface characteristic analysis.

[0003] Hyperspectral imaging technology is a new type of detection technology that combines imaging technology and spectral analysis technology. It can image the target object in the visible light to near-infrared spectral range with hundreds of continuous and narrow spectral bands. This technology can not only obtain the spatial information of the target object, but also obtain its rich spectral information, thereby realizing the precise analysis and identification of the surface characteristics of the object. In the field of brake disc wear detection, hyperspectral imaging technology has significant advantages. First of all, hyperspectral images can capture the subtle spectral changes on the surface of the brake disc under different wear degrees, and these changes are often difficult to observe with the naked eye or traditional imaging technology. Secondly, through the extraction and analysis of the spectral features of hyperspectral images, the quantitative evaluation of the wear degree of the brake disc can be realized, improving the accuracy and objectivity of the detection. Finally, combined with advanced image processing technology and machine learning algorithms, the automatic detection and intelligent identification of the wear degree of the brake disc can be realized, greatly improving the detection efficiency and reliability. Therefore, a method for detecting the surface wear degree of a brake disc is proposed to solve the above problems. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a method for detecting the surface wear degree of a brake disc, which has the advantages of high detection effect and high accuracy, and solves the problems mentioned in the above background technique.

[0005] To achieve the above object, the present invention provides the following technical solution: A method for detecting the surface wear degree of a brake disc, characterized by comprising the following steps: S1: Use a hyperspectral camera to obtain hyperspectral images of the surfaces of brake discs with different wear degrees; S2: Preprocess the acquired hyperspectral image, extract spectral features from the preprocessed hyperspectral image, and fuse the extracted features to obtain the fusion vector F fused , multiple fusion vectors F fused That is, they form the dataset D, and the dataset D is divided into the training set D 1 , the test set D 2 and the validation set D 3 ; S3: Construct the initial model of the convolutional neural network, input the training set D 1 into the initial model of the convolutional neural network, and train it with the cross-entropy loss function as the optimization objective to obtain the intermediate model of the convolutional neural network; S4: Use the validation set D 3 to validate the intermediate model of the convolutional neural network, tune it, and repeat the training until the performance of the intermediate model of the convolutional neural network on the validation set D 3 no longer improves significantly, and obtain the optimized model of the convolutional neural network; S5: Input the test set D 2 into the optimized model of the convolutional neural network, obtain the prediction result of the optimized model of the convolutional neural network on the wear degree of the brake disc, and analyze the difference between the prediction result and the real result; S6: Classify the wear degree of the brake disc according to the prediction result.

[0006] Preferably, the surface detection area of the brake disc is divided into the braking surface, the edge and the groove.

[0007] Preferably, the detailed steps in step S2 are as follows: S2.1: Use Gaussian filtering to smooth the hyperspectral image and reduce noise; S2.2: For each pixel (x, y) in the hyperspectral image, extract a spectral feature vector F spectral (x, y). Let S(x, y, λ) be the spectral reflectance of the pixel (x, y) at the wavelength λ, and the expression is: S2.3: Use the Canny edge detection image processing algorithm to extract the texture feature T(b, c) and edge and other features E(b, c) in the hyperspectral image to obtain the image feature vector F image (a, b), and the expression is: where K is the number of texture features, M is the number of edge features, and both a and b are the position coordinates of the pixel points; S2.4: Combine the spectral feature vector F spectral (x, y) and the image feature vector F imageFuse (a, b) to obtain the fused vector F fused = {F spectral (x, y), F image (a, b)}; S2.5: Multiple fused vectors F fused That is, they form the dataset D.

[0008] Preferably, the specific steps of S3 are as follows: S3.1: Randomly sort the fused vectors F 1 in the training set D fused and divide them into multiple batches according to a preset number of batches; S3.2: Initialize the parameters of the constructed initial convolutional neural network model. Input the fused vectors F 1 in the training set D fused batch by batch into the convolutional neural network model after parameter initialization, and perform iterative training with the cross-entropy loss function as the optimization objective. After the iterative training is completed, the final convolutional neural network model is obtained.

[0009] Preferably, the specific steps of S4 are as follows: S4.1: Use the validation set D 3 to validate the intermediate convolutional neural network model and evaluate the performance of the intermediate convolutional neural network model; S4.2: Optimize the intermediate convolutional neural network model according to its performance on the validation set D 3 ; S4.3: Repeat steps S3 and S4.1 until the performance of the intermediate convolutional neural network model on the validation set D 3 no longer improves significantly, and the optimized convolutional neural network model is obtained.

[0010] Preferably, the specific steps of S5 are as follows: S5.1: Input the samples in the test set D 2 into the optimized convolutional neural network model to obtain the prediction result of the optimized convolutional neural network model for the wear degree of the brake disc. The expression is: where O i is the i-th sample in the test set D 2 , J α is the prediction result, G is the function representation of the model, which maps the input O i to the prediction result J, and θ is the set of model parameters; S5.2: Quantitatively evaluate the prediction result and calculate the prediction result accuracy rate. The expression is: where N testFor the test set D 2 the number of samples in, argmax(J α ) is the index of the class with the highest probability in the prediction result, G α is the true class label of the α-th sample, and 1(·) is the indicator function; S5.3: Analyze the difference between the prediction result and the true result, and the expression is: where C is the number of classes, and J αc indicates whether the α-th sample belongs to class C, and G αc is the probability that the α-th sample predicted by the model belongs to class C.

[0011] Preferably, the step S6 is specifically: According to the prediction result J α obtain the wear degree γ of the brake disc, and the expression is: where when the wear degree γ ≤ β 1 , the wear degree is slight wear, when the wear degree β 1＜ γ ≤ β 2 , the wear degree is slightly medium wear, when the wear degree β 2＜ γ ≤ β 3 , the wear degree is medium wear, when the wear degree β 3＜ γ ≤ β 4 , the wear degree is medium-heavy wear, when the wear degree γ > β 4 , the wear degree is heavy wear, and β 1 , β 2 , β 3 and β 4 are 0.18, 0.43, 0.62 and 0.8 in sequence.

[0012] Compared with the prior art, the present invention provides a method for detecting the surface wear degree of a brake disc, which has the following beneficial effects: This method for detecting the surface wear degree of a brake disc preprocesses the hyperspectral image and extracts spectral features, and fuses the extracted features to form a fusion vector, effectively integrating multi-dimensional spectral information. Such a data set is not only rich in information but also has a clear structure, facilitating the learning and processing of subsequent models. By reasonably dividing the training set, test set and validation set, the generalization ability of the model and the accuracy of evaluation are ensured, realizing the automatic detection and classification of the wear degree of the brake disc, greatly improving the detection efficiency and accuracy. At the same time, through the deep learning technology of the convolutional neural network, the intelligent processing and analysis of complex spectral information are realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1Structural schematic diagram of a method for detecting the surface wear degree of a brake disc proposed by the present invention. Detailed implementation manners

[0014] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0015] Please refer to Figure 1 , a method for detecting the surface wear degree of a brake disc, comprising the following steps: S1: Use a hyperspectral camera to obtain hyperspectral images of the surfaces of brake discs with different wear degrees; S2: Preprocess the obtained hyperspectral images, extract spectral features from the preprocessed hyperspectral images, and fuse the extracted features to obtain a fusion vector F fused , multiple fusion vectors F fused That is, a data set D is formed. The data set D is divided into a training set D 1 , a test set D 2 and a validation set D 3 ; S3: Construct an initial convolutional neural network model, input the training set D 1 into the initial convolutional neural network model, and train it with the cross-entropy loss function as the optimization objective to obtain an intermediate convolutional neural network model; S4: Use the validation set D 3 to validate the intermediate convolutional neural network model, optimize it, and repeat the training until the performance of the intermediate convolutional neural network model on the validation set D 3 no longer improves significantly, and an optimized convolutional neural network model is obtained; Input the test set into the optimized convolutional neural network model, and the prediction result of the brake disc wear degree can be obtained quickly and accurately. By analyzing the difference between the prediction result and the real result, the accuracy and reliability of the model can be further verified. At the same time, classify the brake disc wear degree according to the prediction result, providing a scientific basis for subsequent maintenance or replacement.

[0016] S5: Input the test set D 2 into the optimized convolutional neural network model, obtain the prediction result of the optimized convolutional neural network model for the brake disc wear degree, and analyze the difference between the prediction result and the real result; S6: Classify the wear degree of the brake disc according to the prediction result.

[0017] The surface inspection area of the brake disc is divided into the braking surface, the edge and the grooves; Inspection of the braking surface: The braking surface is the area where the brake disc directly contacts the brake pads and generates frictional force. Inspecting this area can directly reflect the main wear condition and usage status of the brake disc. By analyzing the wear degree, scratches, cracks, etc. on the braking surface, the performance and service life of the brake disc can be evaluated.

[0018] Inspection of the edge: Although the edge part of the brake disc does not directly participate in the braking process, its condition is equally important. Wear, deformation or cracks on the edge may affect the installation stability and braking effect of the brake disc.

[0019] Inspecting the edge helps to detect potential safety hazards in a timely manner and ensure the overall performance of the brake disc.

[0020] Inspection of the grooves: The grooves on the brake disc are usually formed due to long-term use and wear, and they may affect the contact area and frictional force between the brake disc and the brake pads. Inspecting the grooves can understand the depth, width and distribution of the grooves, so as to evaluate their impact on the braking performance. The detailed steps in step S2 are as follows: S2.1: Use Gaussian filtering to smooth the hyperspectral image and reduce noise; S2.2: For each pixel (x, y) in the hyperspectral image, extract a spectral feature vector F spectral (x, y), let S(x, y, λ) be the spectral reflectance of pixel (x, y) at wavelength λ, and the expression is: S2.3: Use the Canny edge detection image processing algorithm to extract the texture feature T(b, c) and edge features such as E(b, c) in the hyperspectral image, and obtain the image feature vector F image (a, b), and the expression is: Among them, K is the number of texture features, M is the number of edge features, and both a and b are the position coordinates of the pixel points; S2.4: Fuse the spectral feature vector F spectral (x, y) and the image feature vector F image (a, b) to obtain the fusion vector F fused = {F spectral (x, y), F image (a, b)}; S2.5: Multiple fusion vectors F fused That is, they form the data set D.

[0021] Step S3 is specifically: S3.1: Randomly sort the fusion vectors F 1 in the training set D fused and divide them into multiple batches according to a preset number of batches; S3.2: Initialize the parameters of the constructed initial convolutional neural network model, and input the fusion vectors F 1 in the training set D fused batch by batch into the convolutional neural network model after parameter initialization, and perform iterative training with the cross-entropy loss function as the optimization objective. After the iterative training is completed, the final convolutional neural network model is obtained.

[0022] Step S4 is specifically as follows: S4.1: Use the validation set D 3 to validate the intermediate convolutional neural network model and evaluate the performance of the intermediate convolutional neural network model. The model structure and hyperparameters can be adjusted to achieve better generalization ability; S4.2: Tune the intermediate convolutional neural network model according to its performance on the validation set D 3 ; S4.3: Repeat Step S3 and Step S4.1 until the performance of the intermediate convolutional neural network model on the validation set D 3 no longer improves significantly, and the optimized convolutional neural network model is obtained.

[0023] Step S5 is specifically as follows: S5.1: Input the samples in the test set D 2 into the optimized convolutional neural network model to obtain the prediction result of the optimized convolutional neural network model for the wear degree of the brake disc. The expression is: where O i is the i-th sample in the test set D 2 , J α is the prediction result, G is the function representation of the model, which maps the input O i to the prediction result J, and θ is the set of model parameters; S5.2: Quantitatively evaluate the prediction result and calculate the prediction result accuracy rate. The expression is: where N test is the number of samples in the test set D 2 , argmax(J α ) is the index of the category with the highest probability in the prediction result, G α is the true category label of the α-th sample, and 1(·) is the indicator function; S5.3: Analyze the differences between the prediction results and the true labels, identify possible misjudgment situations in the model. The difference between the prediction result and the true result is expressed as: where C is the number of categories, J αc indicates whether the α-th sample belongs to category C, and G αc is the probability that the model predicts the α-th sample belongs to category C.

[0024] Step S6 is specifically as follows: The output range of the model is [0, 1], and the closer the output value is to 1, the more severe the wear degree is. According to the prediction result J α obtain the wear degree γ of the brake disc, and the expression is: where, when the wear degree γ ≤ β 1 , the wear degree is slight wear. When the wear degree β 1＜ γ ≤ β 2 , the wear degree is slightly medium wear. When the wear degree β 2＜ γ ≤ β 3 , the wear degree is medium wear. When the wear degree β 3＜ γ ≤ β 4 , the wear degree is medium-heavy wear. When the wear degree γ > β 4 , the wear degree is severe wear. β 1 , β 2 , β 3 and β 4 are 0.18, 0.43, 0.62 and 0.8 in sequence.

[0025] In summary, for this method of detecting the surface wear degree of the brake disc, by preprocessing the hyperspectral image and extracting spectral features, and fusing the extracted features to form a fusion vector, it effectively integrates multi-dimensional spectral information. Such a data set is not only rich in information but also has a clear structure, facilitating the subsequent learning and processing of the model. By reasonably dividing the training set, test set and validation set, it ensures the generalization ability of the model and the accuracy of evaluation, realizes the automatic detection and classification of the wear degree of the brake disc, greatly improves the detection efficiency and accuracy, and at the same time, through the deep learning technology of the convolutional neural network, it realizes the intelligent processing and analysis of complex spectral information.

[0026] It should be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or apparatus. Without further limitation, an element defined by the phrase "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

[0027] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting the degree of wear on the surface of a brake disc, characterized in that: The following steps are involved: S1: Use a hyperspectral camera to obtain hyperspectral images of the brake disc surface with different wear degrees; S2: Preprocess the acquired hyperspectral image, extract spectral features from the preprocessed hyperspectral image, and fuse the extracted features to obtain the fusion vector F fused , multiple fusion vectors F fused That is, a data set D is formed, and the data set D is divided into a training set D1, a test set D2, and a validation set D3; S3: Construct the initial model of the convolutional neural network, input the training set D1 into the initial model of the convolutional neural network, train it with the cross entropy loss function as the optimization target, and obtain the intermediate model of the convolutional neural network; S4: Use the validation set D3 to validate the intermediate model of the convolutional neural network, tune it, and repeat the training until the performance of the intermediate model of the convolutional neural network on the validation set D3 is no longer significantly improved, and the optimized model of the convolutional neural network is obtained; S5: input the test set D2 into the convolutional neural network optimization model, obtain the prediction result of the convolutional neural network optimization model on the wear degree of the brake disc, and analyze the difference between the prediction result and the actual result; S6: Classify the wear degree of the brake disc according to the prediction result.

2. A method for detecting the degree of wear on the surface of a brake disc according to claim 1, characterized in that: The brake disc surface detection area is divided into a braking surface, an edge and a groove.

3. A method for detecting the degree of wear on the surface of a brake disc according to claim 1, characterized in that: The detailed steps in step S2 are: S2.1: Use Gaussian filtering to smooth the hyperspectral image and reduce noise; S2.2: For each pixel (x, y) in the hyperspectral image, extract a spectral feature vector F of length L spectral (x, y), let S (x, y, λ) be the spectral reflectance of pixel (x, y) at wavelength λ, the expression is: S2.3: Use the Canny edge detection image processing algorithm to extract the texture features T(b,c) and edge features E(b,c) in the hyperspectral image and obtain the image feature vector F image (a, b), the expression is: Among them, K is the number of texture features, M is the number of edge features, and a and b are the position coordinates of pixel points; S2.4: The spectral feature vector F spectral (x, y) and image feature vector F image (a, b) are fused to obtain the fusion vector F fused ={F spectral (x,y),F image (a, b)}; S2.5: Multiple fusion vectors F fused That constitutes the data set D.

4. A method for detecting the degree of wear on the surface of a brake disc according to claim 1, characterized in that: The step S3 is specifically: S3.1: The fusion vector F in the training set D1 fused Randomly sort and divide into multiple batches according to the preset number of batches; S3.2: Initialize the parameters of the constructed convolutional neural network initial model and convert the fusion vector F in the training set D1 fused In the convolutional neural network model initialized with this input parameter in batches, iterative training is performed with the cross entropy loss function as the optimization target, and the final convolutional neural network model is obtained after the iterative training is completed.

5. The method for detecting the degree of wear on the surface of a brake disc according to claim 1, characterized in that: The step S4 is specifically as follows: S4.1: Use the validation set D3 to validate the intermediate model of the convolutional neural network and evaluate the performance of the intermediate model of the convolutional neural network; S4.2: Tune the intermediate model of the convolutional neural network based on the performance on the validation set D3; S4.3: Repeat steps S3 and S4.1 until the performance of the convolutional neural network intermediate model on the validation set D3 is no longer significantly improved, and the convolutional neural network optimized model is obtained.

6. A method for detecting the degree of wear on the surface of a brake disc according to claim 1, characterized in that: The step S5 is specifically as follows: S5.1: Input the samples in the test set D2 into the convolutional neural network optimization model to obtain the prediction result of the convolutional neural network optimization model on the wear degree of the brake disc, which is expressed as: Among them, O i is the i-th sample in the test set D2, J α To predict the result, G is the function representation of the model, and the input O i Mapped to the prediction result J, θ is the parameter set of the model; S5.2: Quantitatively evaluate the prediction results and calculate the accuracy of the prediction results. The expression is: Among them, N test is the number of samples in the test set D2, argmax(J α ) is the index of the category with the highest probability in the prediction result, G α is the true category label of the αth sample, 1(·) is the indicator function; S5.3: Analyze the difference between the predicted results and the actual results, expressed as: Where C is the number of categories, J αc Whether the αth sample belongs to category C, G αc The probability that the αth sample belongs to category C is predicted by the model.

7. A method for detecting the degree of wear on the surface of a brake disc according to claim 1, characterized in that: The step S6 is specifically as follows: According to the prediction results J α The wear degree γ of the brake disc is obtained, and the expression is: Among them, when the wear degree γ≤β1, the wear degree is slight wear, and when the wear degree β 1＜ γ≤β2, the wear degree is slightly to medium wear. 2＜ γ≤β3, the wear degree is moderate wear. 3＜ When γ≤β4, the wear degree is moderate to heavy wear. When the wear degree γ>β4, the wear degree is heavy wear. β1, β2, β3 and β4 are 0.18, 0.43, 0.62 and 0.8 respectively.

Images