A new energy automobile accessory surface defect detection system

Abstract

A kind of be applied to new energy automobile accessory surface defect detection system, it is related to new energy automobile accessory detection technical field, obtains the multi-source heterogeneous data in the production process of automobile accessory;The multi-source heterogeneous data of each process subsequence is analyzed by feature extraction, obtains the noise level corresponding to surface image, and the surface image is preprocessed by denoising;The texture feature, surface flatness and surface temperature of each process subsequence automobile accessory are extracted, and real-time defect detection is carried out;Dynamic threshold adjustment model is constructed, and dynamic threshold interval is set according to the process unit characteristics and production environment data of process subsequence;Automobile accessory for generating defect detection qualified signal is artificially sampled, and adjustment correction operation of dynamic threshold adjustment model is carried out according to sampling result, significantly improve defect detection efficiency and accuracy.

Description

TECHNICAL FIELD

[0001] The application relates to the field of new energy automobile accessory detection technology, and particularly relates to a new energy automobile accessory surface defect detection system. BACKGROUND

[0002] CN118446970A "Automobile accessory defect detection method based on image features", S1: an image acquisition system is used to acquire an automobile accessory surface image; S2: image preprocessing: the acquired image is first decomposed using Laplace, to obtain a low-frequency band and a high-frequency band, the high-frequency band is amplified, and the low-frequency band is improved in image quality through a Retinex image enhancement algorithm.

[0003] CN117197141A "Automobile part surface defect detection method" acquires a gray scale image of an automobile part, determines an initial threshold value according to a gray scale value of a pixel point, and then obtains a to-be-detected threshold value and a preliminary defect region and a normal region; the trend and positional relationship between the final defect regions are used to screen defect discontinuous regions; a light influence value is obtained based on the gray scale value of the pixel point, the gray scale value of the pixel point is adjusted, and a gray scale adjustment value is obtained in combination with the gray scale value of the pixel point in the normal region; a comparison threshold value is obtained according to the adjusted gray scale value of the pixel point and compared with the to-be-detected threshold value, to obtain an optimal threshold value for defect detection of the gray scale image.

[0004] With the rapid development of the new energy automobile industry, the quality requirements for new energy automobile accessories are increasingly improved. The defects on the surface of the accessories may affect the performance, safety and service life of the automobile. Traditional new energy automobile accessory surface defect detection methods, such as manual visual inspection, not only have low efficiency, but also are easily affected by subjective factors of the inspectors, and it is difficult to ensure the accuracy and consistency of the detection. Some existing automatic detection systems often have problems such as insufficient detection accuracy and limited recognition ability for complex defect types, and cannot meet the needs of high-quality production of new energy automobile accessories. SUMMARY

[0005] In order to solve the above technical problems, the purpose of the present application is to provide a new energy automobile accessory surface defect detection system, which comprises a monitoring center, and the monitoring center is communicatively connected with a data acquisition module, a feature extraction module, a defect detection module, a threshold value adjustment module and a threshold value feedback module.

[0006] The data acquisition module is used to acquire multi-source heterogeneous data in the production process of automobile accessories, and the multi-source heterogeneous data comprises surface images, three-dimensional topographic data, temperature data and production environment data (such as temperature and humidity in the workshop).

[0007] The feature extraction module is configured to perform feature extraction analysis on the multi-source heterogeneous data of each process sub-sequence, obtain the noise level corresponding to the surface image, and perform denoising preprocessing on the surface image.

[0008] The defect detection module is configured to extract the texture features, surface flatness and surface temperature of the automobile parts in each process sub-sequence, and perform real-time defect detection.

[0009] The threshold adjustment module is configured to construct a dynamic threshold adjustment model, and set a dynamic threshold interval according to the process unit characteristics and production environment data of the process sub-sequence.

[0010] The threshold feedback module is configured to perform artificial sampling inspection on the automobile parts generating the defect detection qualified signal, and perform adjustment and correction operation of the dynamic threshold adjustment model according to the sampling inspection result.

[0011] It should be further explained that, in the specific implementation process, the process of acquiring the multi-source heterogeneous data in the automobile part production process by the data acquisition module includes:

[0012] Acquiring the process flow characteristics (such as casting, boring, milling plane, drilling, stamping, etc.) of the automobile part production, dividing the automobile part production process according to the process flow characteristics into a plurality of process sub-sequences;

[0013] Setting a defect detection point in each process sub-sequence, the defect detection point is configured to collect multi-source heterogeneous data and mark the collection time, and set a collection period.

[0014] Further, the process of performing feature extraction analysis on the multi-source heterogeneous data of each process sub-sequence by the feature extraction module to obtain the noise level corresponding to the surface image includes:

[0015] Performing grayscale processing on the surface image, converting the surface image into a grayscale image, performing statistical analysis on the gray values in the neighborhood of each pixel point in the grayscale image, obtaining the local variance in the neighborhood of each pixel point in the grayscale image, performing feature extraction on the three-dimensional topographic data, obtaining the surface flatness of the automobile part, and simultaneously obtaining the surface temperature of the automobile part according to the temperature data;

[0016] The calculation formula for obtaining the local variance in the neighborhood of the pixel point is:

[0017]

[0018] Wherein, fc represents the local variance, d represents the gray mean value, I(i, j) represents the gray value at the pixel point (i, j), W represents the set of all pixel points in the neighborhood range, and n1 represents the total number of pixel points in the neighborhood range.

[0019] According to the surface flatness and surface temperature of the automobile parts, the surface flatness and local temperature variance in the neighborhood of each pixel point in the gray image are obtained, and the feature energy value in the neighborhood of each pixel point is obtained according to the surface flatness and local temperature variance in the neighborhood of each pixel point.

[0020] The feature energy value in the neighborhood of each pixel point is compared with a preset feature energy threshold value, and if the feature energy value in the neighborhood of the pixel point is less than the feature energy threshold value, the neighborhood of the pixel point is marked as a first pixel point neighborhood.

[0021] The local variances of the first pixel point neighborhoods are summed and averaged to obtain an average local variance, and the average local variance is marked as a noise level.

[0022] Further, the process of denoising preprocessing of the surface image includes:

[0023] A filter parameter reference table is constructed, the filter parameter reference table including filter parameters corresponding to different noise difference levels, the local variances in the neighborhoods of the respective pixel points are compared with the noise level to obtain noise difference levels of the respective pixel points, noise difference level = noise level - local variance, filter parameters of the respective pixel points are obtained according to the noise difference levels of the respective pixel points and the filter parameter reference table, and the respective pixel points of the gray image are subjected to Gaussian filtering processing using a Gaussian filtering method according to the filter parameters of the respective pixel points to obtain a gray image after noise removal, the filter parameters including a Gaussian kernel size and a standard deviation, when the local variance is close to the average local variance, it indicates that this area is mainly composed of noise, and then the standard deviation is increased to enhance the smoothing effect, and vice versa, if the local variance is much greater than the average local variance, it indicates that there may be important image features such as edges in this area, and then the standard deviation is reduced to reduce the influence on these features.

[0024] Further, the process of the defect detection module extracting texture features, surface flatness and surface temperature of the automobile parts in each process subsequence and performing real-time defect detection includes:

[0025] A gray level co-occurrence matrix in the neighborhood of each pixel point in the gray image after noise removal is obtained using a gray level co-occurrence matrix technique, and texture features in the neighborhood of each pixel point are obtained according to the gray level co-occurrence matrix in the neighborhood of each pixel point.

[0026] Dynamic threshold intervals are set according to the process unit characteristics and production environment data of the process subsequence, the dynamic threshold intervals including threshold intervals corresponding to the texture features, threshold intervals corresponding to the surface flatness and threshold intervals corresponding to the surface temperature, and the texture features, surface flatness and surface temperature of the automobile parts are compared with the corresponding threshold intervals, and if the texture features, surface flatness or surface temperature of the automobile parts are not located in the corresponding threshold intervals, a defect alarm signal of the automobile parts is generated.

[0027] If the texture feature, surface flatness and surface temperature of the automobile part are within the corresponding threshold intervals, a defect detection qualified signal of the automobile part is generated.

[0028] Further, the process of constructing the dynamic threshold adjustment model by the threshold adjustment module includes:

[0029] Surface images, three-dimensional topographic data and temperature data of a plurality of automobile parts under different process unit characteristics and different production environment data conditions are collected, and a plurality of automobile parts are labeled as having defects or not having defects. Feature extraction is performed on the surface images, three-dimensional topographic data and temperature data of the plurality of automobile parts with defects and the plurality of automobile parts without defects, to obtain texture features, surface flatness and surface temperature of the plurality of automobile parts with defects and the plurality of automobile parts without defects under different process unit characteristics and different production environment data conditions.

[0030] A dynamic threshold adjustment model is constructed based on deep learning, and the texture features, surface flatness and surface temperature of the plurality of automobile parts with defects and the plurality of automobile parts without defects under different process unit characteristics and different production environment data conditions are used as a training set and a test set. The training set is input into the dynamic threshold adjustment model for training until the loss function training is stable, and the model parameters are saved. The dynamic threshold adjustment model is tested by the test set until it meets the preset requirements, and the dynamic threshold adjustment model is output.

[0031] Constructing a dynamic threshold adjustment model based on deep learning is a complex process that involves multiple steps such as model selection, training, validation and testing. The following is a detailed supplementary explanation of this process:

[0032] Support Vector Machine (SVM) is selected as the deep learning architecture. After determining the model architecture, the next step is to define the loss function, and Binary Cross-Entropy Loss is selected as the optimization objective. Then, the prepared training set is input into the selected deep learning model to start training. During the training process, the weights are constantly updated through the backpropagation algorithm, so that the loss function gradually decreases until it reaches a stable state. During this period, techniques such as EarlyStopping are used to avoid overfitting. In addition to the basic training process, various parameters of the model are also optimized through GridSearch, including learning rate, batch size, regularization coefficient, etc.

[0033] When the model training is completed and the parameters are adjusted, the final evaluation is performed through the test set to obtain the evaluation results of the model, including classification indicators such as accuracy, recall rate, F1 score, etc. According to the evaluation results on the test set, it is judged whether the model meets the expected standard. If it meets the requirements, the model parameters are saved and prepared for deployment; if it is not ideal, it needs to return to a certain stage to re-examine the data quality, model structure or training strategy and other problems.

[0034] Further, the process of setting a dynamic threshold interval according to the process unit characteristics and production environment data of the process subsequence includes:

[0035] The process unit characteristics and production environment data of the process subsequence are input into the dynamic threshold adjustment model, and the threshold interval corresponding to the texture feature, the threshold interval corresponding to the surface flatness and the threshold interval corresponding to the surface temperature of the non-defective automobile parts are output according to the dynamic threshold adjustment model.

[0036] Further, the process of the threshold feedback module artificially sampling the automobile parts generating the defect detection qualified signal and adjusting and correcting the operation of the dynamic threshold adjustment model according to the sampling result includes:

[0037] The automobile parts generating the defect detection qualified signal are artificially sampled, and it is judged whether the automobile parts generating the defect detection qualified signal have defects. If there are defects, the process unit characteristics, production environment data, texture features, surface flatness and surface temperature of the automobile parts are used as training data to retrain the dynamic threshold adjustment model.

[0038] Compared with the prior art, the beneficial effects of the present application are:

[0039] 1. Multi-source data fusion: The data acquisition module of the system comprehensively acquires surface images, three-dimensional topographic data, temperature data and production environment data, covering key information from the surface visual features of the parts to the internal structure and external production environment. The comprehensive data collection can more accurately reflect the state of the parts and avoid the limitations of single data type, providing rich and comprehensive data support for accurate defect detection.

[0040] 2. Process subdivision collection: According to the characteristics of the automobile parts production process, the process subsequence is divided, and special defect detection points are set in each subsequence. The data is collected in a targeted manner. This not only improves the data collection efficiency, but also captures potential defects at different production stages in time, making the detection more timely and targeted, which helps to find problems early in production and reduce repair costs later.

[0041] 3. Accurate noise evaluation: After the surface image is grayed, the feature extraction module analyzes the surface flatness and local temperature variance from the neighborhood of the pixel points, and obtains the characteristic energy value to determine the noise level. This multi-dimensional analysis method can more accurately evaluate the noise situation, and compared with the traditional noise evaluation method based on the image itself, it considers more comprehensive factors and provides more accurate evaluation results.

[0042] 4. Adaptive noise reduction: A filter parameter table is constructed, and the corresponding filter parameters are obtained for Gaussian filtering according to the difference between the local variance and the noise level in the neighborhood of each pixel point. This adaptive noise reduction method for different pixel noise differences can effectively remove noise and maximize the preservation of image details, providing high-quality image data for subsequent defect detection and improving detection accuracy.

[0043] 5. Multi-feature comprehensive detection: The defect detection module detects the automobile parts from three dimensions of texture features, surface flatness and surface temperature. Multi-dimensional feature extraction and analysis can more comprehensively describe the surface state of the parts, effectively identify various types of defects such as texture abnormalities, surface unevenness or temperature abnormalities, and greatly improve the reliability and accuracy of detection compared with single feature detection.

[0044] 6. Dynamic threshold adaptation: A dynamic threshold adjustment model is constructed based on deep learning, and a dynamic threshold interval is set according to the process unit characteristics and production environment data of the process subsequence. Process changes and environmental fluctuations in the production process will not affect the accuracy of detection. The system can adaptively adjust the threshold to avoid false positives or false negatives caused by fixed thresholds, ensuring accurate detection of defects under various complex and changing production conditions.

[0045] 7. Model optimization training: The threshold adjustment module collects a large amount of data of defective and non-defective automobile parts under different conditions to train the dynamic threshold adjustment model, so that it has good generalization ability. The threshold feedback module artificially inspects the qualified parts, and if a false judgment is found, the relevant data is used to retrain the model. This closed-loop feedback mechanism enables the system to continuously learn and adapt to changes in actual production, continuously optimize detection results, and as time passes and data accumulates, the accuracy and reliability of detection will continuously improve. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A schematic diagram of a surface defect detection system for new energy automobile parts. DETAILED DESCRIPTION

[0047] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of the present application.

[0048] As shown in Figure 1 A surface defect detection system applied to new energy automobile accessories, comprising a monitoring center, the monitoring center is in communication with a data acquisition module, a feature extraction module, a defect detection module, a threshold adjustment module and a threshold feedback module;

[0049] The data acquisition module is used to acquire multi-source heterogeneous data in the production process of automobile accessories, and the multi-source heterogeneous data includes surface images, three-dimensional topographic data, temperature data and production environment data (such as temperature, humidity, etc. in the workshop);

[0050] The feature extraction module is used to perform feature extraction analysis on the multi-source heterogeneous data of each process subsequence, acquire the noise level corresponding to the surface image, and perform denoising preprocessing on the surface image;

[0051] The defect detection module is used to extract the texture features, surface flatness and surface temperature of each process subsequence automobile accessory, and perform real-time defect detection;

[0052] The threshold adjustment module is used to construct a dynamic threshold adjustment model, and set a dynamic threshold interval according to the process unit characteristics and production environment data of the process subsequence;

[0053] The threshold feedback module is used to artificially inspect the automobile accessories that generate defect detection qualified signals, and adjust and correct the dynamic threshold adjustment model according to the inspection results.

[0054] It needs to be further explained that, in the specific implementation process, the process of acquiring multi-source heterogeneous data in the production process of automobile accessories by the data acquisition module includes:

[0055] Acquiring the process flow characteristics (such as casting, boring, milling, drilling, stamping, etc.) of automobile accessory production, dividing the automobile accessory production process according to the process flow characteristics into several process subsequences;

[0056] Setting defect detection points in each process subsequence, the defect detection points are used to collect multi-source heterogeneous data and mark the collection time, and set the collection period.

[0057] It needs to be further explained that, in the specific implementation process, the process of performing feature extraction analysis on the multi-source heterogeneous data of each process subsequence by the feature extraction module to acquire the noise level corresponding to the surface image includes:

[0058] The surface image is converted to grayscale. The grayscale value in the neighborhood of each pixel in the grayscale image is statistically analyzed to obtain the local variance in the neighborhood of each pixel in the grayscale image. The features of the three-dimensional topography data are extracted to obtain the surface flatness of the automotive parts. At the same time, the surface temperature of the automotive parts is obtained based on the temperature data.

[0059] The formula for calculating the local variance within the neighborhood of a pixel is as follows:

[0060]

[0061] Where fc represents the local variance, d represents the gray mean, I(i,j) represents the gray value at pixel (i,j), W represents the set of all pixels in the neighborhood, and n1 represents the total number of pixels in the neighborhood.

[0062] Based on the surface smoothness and surface temperature of the automotive parts, the surface smoothness and local temperature variance in the neighborhood of each pixel in the grayscale image are obtained, and the feature energy value in the neighborhood of each pixel is obtained based on the surface smoothness and local temperature variance in the neighborhood of each pixel.

[0063] The feature energy value in the neighborhood of each pixel is compared with a preset feature energy threshold. If the feature energy value in the neighborhood of a pixel is less than the feature energy threshold, the neighborhood of the pixel is marked as the first pixel neighborhood.

[0064] The local variances of the neighborhood of each first pixel are summed and averaged to obtain the average local variance, which is then marked as the noise level.

[0065] It should be further explained that, in the specific implementation process, the denoising preprocessing of the surface image includes:

[0066] A filter parameter lookup table is constructed, which includes filter parameters corresponding to different noise difference levels. The local variance in the neighborhood of each pixel is compared with the noise level to obtain the noise difference level of each pixel. Noise difference level = noise level - local variance. Based on the noise difference level of each pixel and the filter parameter lookup table, the filter parameters of each pixel are obtained. Gaussian filtering is then applied to each pixel of the grayscale image using the filter parameters of each pixel to obtain the noise-removed grayscale image. The filter parameters include the Gaussian kernel size and standard deviation. When the local variance is close to the average local variance, it indicates that this area is mainly composed of noise, so the standard deviation is increased to enhance the smoothing effect; conversely, if the local variance is much greater than the average local variance, it indicates that there may be important image features such as edges, so the standard deviation is decreased to reduce the influence of these features.

[0067] It should be further explained that, in the specific implementation process, the defect detection module extracts the texture features, surface flatness, and surface temperature of automotive parts from each process sub-sequence and performs real-time defect detection, including:

[0068] The gray-level co-occurrence matrix technique is used to obtain the gray-level co-occurrence matrix in the neighborhood of each pixel in the noise-removed gray-level image, and the texture features in the neighborhood of each pixel are obtained based on the gray-level co-occurrence matrix in the neighborhood of each pixel.

[0069] Dynamic threshold ranges are set based on the process unit characteristics and production environment data of the process subsequence. The dynamic threshold ranges include threshold ranges corresponding to texture features, surface flatness, and surface temperature. The texture features, surface flatness, and surface temperature of the automotive parts are compared with the corresponding threshold ranges. If the texture features, surface flatness, or surface temperature of the automotive parts are not within the corresponding threshold ranges, a defect alarm signal for the automotive parts is generated.

[0070] If the texture features, surface flatness, and surface temperature of the automotive parts are all within the corresponding threshold range, a defect detection pass signal for the automotive parts will be generated.

[0071] It should be further explained that, in the specific implementation process, the process of the threshold adjustment module constructing the dynamic threshold adjustment model includes:

[0072] The surface images, three-dimensional morphology data, and temperature data of several automotive parts under different process unit characteristics and production environment data conditions were collected. Defective and non-defective parts were marked, and feature extraction was performed on the surface images, three-dimensional morphology data, and temperature data of several defective and non-defective automotive parts. The texture features, surface flatness, and surface temperature of several defective and non-defective automotive parts under different process unit characteristics and production environment data conditions were obtained.

[0073] A dynamic threshold adjustment model is constructed based on deep learning. The texture features, surface flatness, and surface temperature of several defective and non-defective auto parts under different process unit characteristics and production environment data conditions are used as training and testing sets. The training set is input into the dynamic threshold adjustment model for training until the loss function is stable and the model parameters are saved. The dynamic threshold adjustment model is tested through the test set until it meets the preset requirements, and then the dynamic threshold adjustment model is output.

[0074] Building a deep learning-based dynamic threshold adjustment model is a complex process involving multiple steps such as model selection, training, validation, and testing. The following is a detailed supplementary explanation of this process:

[0075] After selecting Support Vector Machine (SVM) as the deep learning architecture and determining the model architecture, the next step is to define the loss function. Binary Cross-Entropy Loss is chosen as the optimization objective. The prepared training set is then input into the selected deep learning model to begin training. During training, the weights are continuously updated using the backpropagation algorithm, causing the loss function to gradually decrease until a stable state is reached. During this process, techniques such as Early Stopping are used to avoid overfitting. In addition to the basic training procedures, various parameters of the model, including the learning rate, batch size, and regularization coefficient, are fine-tuned using GridSearch.

[0076] Once the model training is complete and the parameters have been tuned, a final evaluation is performed using a test set to obtain the model's evaluation results. These results include classification metrics such as accuracy, recall, and F1 score. Based on the evaluation results on the test set, it is determined whether the model has met the expected standards. If the requirements are met, the model parameters are saved and deployment is prepared; otherwise, it is necessary to return to a previous stage to re-examine issues such as data quality, model structure, or training strategy.

[0077] It should be further explained that, in the specific implementation process, the process of setting dynamic threshold ranges based on the characteristics of the process unit and production environment data of the process sub-sequence includes:

[0078] The process unit characteristics of the process subsequence and the production environment data are input into the dynamic threshold adjustment model. Based on the dynamic threshold adjustment model, the threshold ranges corresponding to the texture features, surface flatness, and surface temperature of the defect-free automotive parts are output.

[0079] It should be further explained that, in the specific implementation process, the threshold feedback module performs manual random inspections of automotive parts that have generated defect detection qualified signals. The process of adjusting and correcting the dynamic threshold adjustment model based on the random inspection results includes:

[0080] Manual sampling inspection is conducted on automotive parts that have generated defect detection pass signals to determine whether there are defects. If defects are found, the process unit characteristics, production environment data, texture features, surface flatness, and surface temperature of the automotive parts are used as training data to retrain the dynamic threshold adjustment model.

[0081] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A surface defect detection system for new energy vehicle parts, characterized in that, It includes a monitoring center, which is communicatively connected to a data acquisition module, a feature extraction module, a defect detection module, a threshold adjustment module, and a threshold feedback module; The data acquisition module is used to acquire multi-source heterogeneous data during the production process of automotive parts. The multi-source heterogeneous data includes surface images, three-dimensional morphology data, temperature data, and production environment data. The feature extraction module is used to perform feature extraction and analysis on multi-source heterogeneous data of each process subsequence, obtain the noise level corresponding to the surface image, and perform noise reduction preprocessing on the surface image; The defect detection module is used to extract the texture features, surface flatness, and surface temperature of automotive parts in each process subsequence and to perform real-time defect detection. The threshold adjustment module is used to build a dynamic threshold adjustment model and set dynamic threshold ranges based on the process unit characteristics and production environment data of the process subsequence. The threshold feedback module is used to perform manual sampling inspections on automotive parts that have generated defect detection qualified signals, and to adjust and correct the dynamic threshold adjustment model based on the sampling inspection results. The feature extraction module performs feature extraction and analysis on the multi-source heterogeneous data of each process sub-sequence to obtain the noise level corresponding to the surface image. The process includes: The surface image is converted to grayscale. The grayscale value in the neighborhood of each pixel in the grayscale image is statistically analyzed to obtain the local variance in the neighborhood of each pixel in the grayscale image. The features of the three-dimensional topography data are extracted to obtain the surface flatness of the automotive parts. At the same time, the surface temperature of the automotive parts is obtained based on the temperature data. Based on the surface smoothness and surface temperature of the automotive parts, the surface smoothness and local temperature variance in the neighborhood of each pixel in the grayscale image are obtained, and the feature energy value in the neighborhood of each pixel is obtained based on the surface smoothness and local temperature variance in the neighborhood of each pixel. The feature energy value in the neighborhood of each pixel is compared with a preset feature energy threshold. If the feature energy value in the neighborhood of a pixel is less than the feature energy threshold, the neighborhood of the pixel is marked as the first pixel neighborhood. The local variances of the neighborhood of each first pixel are summed and averaged to obtain the average local variance, which is then marked as the noise level.

2. The surface defect detection system for new energy vehicle parts according to claim 1, characterized in that, The process by which the data acquisition module acquires multi-source heterogeneous data during the automotive parts production process includes: The process characteristics of automotive parts production are obtained, and the automotive parts production process is divided into several process sub-sequences according to the process characteristics. Defect detection points are set in each process subsequence. These defect detection points are used to collect multi-source heterogeneous data and mark the collection time, and the collection cycle is set.

3. The surface defect detection system for new energy vehicle parts according to claim 2, characterized in that, The process of denoising preprocessing for surface images includes: A filter parameter lookup table is constructed, which includes filter parameters corresponding to different noise difference levels. The local variance in the neighborhood of each pixel is compared with the noise level to obtain the noise difference level of each pixel. The filter parameters of each pixel are obtained according to the noise difference level of each pixel and the filter parameter lookup table. The Gaussian filtering method is used to perform Gaussian filtering on each pixel of the grayscale image according to the filter parameters of each pixel to obtain the grayscale image after noise removal.

4. The surface defect detection system for new energy vehicle parts according to claim 3, characterized in that, The defect detection module extracts texture features, surface smoothness, and surface temperature of automotive parts from each process sub-sequence and performs real-time defect detection. This process includes: The gray-level co-occurrence matrix technique is used to obtain the gray-level co-occurrence matrix in the neighborhood of each pixel in the noise-removed gray-level image, and the texture features in the neighborhood of each pixel are obtained based on the gray-level co-occurrence matrix in the neighborhood of each pixel. Dynamic threshold ranges are set based on the process unit characteristics and production environment data of the process subsequence. The dynamic threshold ranges include threshold ranges corresponding to texture features, surface flatness, and surface temperature. The texture features, surface flatness, and surface temperature of the automotive parts are compared with the corresponding threshold ranges. If the texture features, surface flatness, or surface temperature of the automotive parts are not within the corresponding threshold ranges, a defect alarm signal for the automotive parts is generated. If the texture features, surface flatness, and surface temperature of the automotive parts are all within the corresponding threshold range, a defect detection pass signal for the automotive parts will be generated.

5. The surface defect detection system for new energy vehicle parts according to claim 4, characterized in that, The process of building a dynamic threshold adjustment model by the threshold adjustment module includes: The surface images, three-dimensional morphology data, and temperature data of several automotive parts under different process unit characteristics and production environment data conditions were collected. Defective and non-defective parts were marked, and feature extraction was performed on the surface images, three-dimensional morphology data, and temperature data of several defective and non-defective automotive parts. The texture features, surface flatness, and surface temperature of several defective and non-defective automotive parts under different process unit characteristics and production environment data conditions were obtained. A dynamic threshold adjustment model is constructed based on deep learning. The texture features, surface flatness, and surface temperature of several defective and non-defective auto parts under different process unit characteristics and production environment data conditions are used as training and testing sets. The training set is input into the dynamic threshold adjustment model for training until the loss function is stable and the model parameters are saved. The dynamic threshold adjustment model is tested through the test set until it meets the preset requirements, and then the dynamic threshold adjustment model is output.

6. The surface defect detection system for new energy vehicle parts according to claim 5, characterized in that, The process of setting dynamic threshold ranges based on the characteristics of process units and production environment data of process sub-sequences includes: The process unit characteristics of the process subsequence and the production environment data are input into the dynamic threshold adjustment model. Based on the dynamic threshold adjustment model, the threshold ranges corresponding to texture features, surface flatness, and surface temperature are output.

7. The surface defect detection system for new energy vehicle parts according to claim 6, characterized in that, The threshold feedback module performs manual random inspections on automotive parts that have generated defect detection pass signals. The process of adjusting and correcting the dynamic threshold adjustment model based on the inspection results includes: Manual sampling inspection is conducted on automotive parts that have generated defect detection pass signals to determine whether there are defects. If defects are found, the process unit characteristics, production environment data, texture features, surface flatness, and surface temperature of the automotive parts are used as training data to retrain the dynamic threshold adjustment model.

Images