Case structure defect identification method, system, equipment and medium

Abstract

The invention discloses a case structure defect identification method, system and equipment and a medium. The method comprises the following steps: acquiring an image sequence of a to-be-detected case at multiple angles and under different illumination conditions through a preset image acquisition device; performing illumination normalization, reflection suppression processing and data enhancement processing on the image sequence; inputting the preprocessed image sequence into a pre-trained defect identification model, and calculating and outputting a defect identification result by the defect identification model according to the input image; associating the defect identification result with the production batch information of the case to be detected, and storing the defect identification result in a defect result database; according to the technical scheme, on the basis of data in the defect result database, statistical analysis is conducted on the defects according to production batches, when the defect rate of a certain batch exceeds a preset threshold value, an early warning signal and a process optimization suggestion are generated, the detection precision is remarkably improved, the batches can be traced, and the production quality management efficiency is greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep learning, in particular to a case structure defect identification method, system, device and medium. BACKGROUND

[0002] With the rapid development of information technology and the Internet industry, as one of the basic hardware of data centers, cloud computing platforms and various information equipment, the manufacturing quality of the server case directly affects the reliability and stability of the system. Case structure defects, such as scratches, screw hole misplacement, plate warping, and poor riveting, not only affect the appearance of the product, but also may cause a decrease in structural strength, electromagnetic shielding failure, and reduced heat dissipation efficiency during long-term use, thereby causing system failure.

[0003] In the prior art, the main case structure defect detection methods are manual visual inspection, sampling detection, and traditional image processing-based detection. These detection methods have many shortcomings, such as insufficient detection accuracy, especially small scratches and subtle misplacement that are difficult to accurately identify, lack of automated classification of defect types in the detection process, difficulty in implementing targeted process improvement, inability to implement full-quantity detection and batch tracing, and difficulty in timely discovering and locating manufacturing process problems.

[0004] Therefore, it is urgent to study a case structure defect identification method that can not only accurately identify multiple defect types, but also trace the production batch data to guide manufacturing process optimization and reduce the occurrence of repeated defects. SUMMARY

[0005] The present application provides a case structure defect identification method, system, device and medium to overcome at least one technical problem in the related art.

[0006] According to a first aspect of the embodiments of the present application, a case structure defect identification method is provided, comprising: Obtaining production batch information of a case to be detected, and obtaining an image sequence of the case to be detected under multiple angles and different illumination conditions through a pre-set image acquisition device; Preprocessing the image sequence, the preprocessing including illumination normalization, anti-reflection suppression processing and data enhancement processing; Inputting the preprocessed image sequence into a pre-trained defect identification model, calculating and outputting a defect identification result from the defect identification model according to the input image, the defect identification result including a defect type, a position coordinate and a confidence level; Associating the defect identification result with the production batch information of the case to be detected, and storing it in a defect result database; Based on the data in the defect result database, the defects are statistically analyzed by production batch, and when the defect rate of a batch exceeds the preset threshold, a warning signal and process optimization suggestions are generated.

[0007] Optionally, the pre-set image acquisition device comprises an industrial camera, a light supplement system and a motion platform, wherein the light supplement system adopts a combination of a ring-shaped LED light source and a polarization filter, and the acquisition step is: Placing the to-be-detected case in the shooting area, controlling the industrial camera to move on the motion platform at a preset speed and path, acquiring images of the six outer surfaces and key interface areas of the to-be-detected case, and changing the incident angle of the LED light source at each acquisition point to obtain multiple images of the same position under different lighting conditions to obtain an image sequence.

[0008] Optionally, the pre-processing of the image sequence comprises the steps of illumination normalization, anti-reflection suppression processing and data enhancement processing. The Retinex image enhancement algorithm is used for illumination normalization processing, and the formula is wherein, is the original pixel value, is a low-pass filter kernel, is the pixel value after normalization processing; The polarization filter is combined with the image polarization decomposition algorithm to extract the non-reflective component in the image for anti-reflection suppression. Data enhancement is performed by means of random rotation, translation, contrast adjustment and Gaussian noise addition.

[0009] Optionally, the defect recognition model is a convolutional neural network model based on the ResNet-50 architecture, and a Softmax classifier is connected after the last fully connected layer, and the loss function formula is as follows: wherein, N is the number of samples, C is the number of categories, is the true label, is the model prediction probability.

[0010] Optionally, after the step of calculating and outputting the defect recognition result by the defect recognition model according to the input image, the following steps are further included: The gradient weighted class activation mapping is used to locate the defect area through the defect position coordinates, and a class activation heat map identifying the defect position is generated, and on the basis of CNN classification, Grad-CAM (gradient weighted class activation mapping) is introduced to realize the defect positioning step. The gradient of the target class is obtained ; Calculate the weight: wherein Z is the total number of categories; Generate a heat map: to achieve defect positioning.

[0011] Optionally, the step of associating the defect identification result with the production batch information of the to-be-detected case includes: By scanning the product serial number on the to-be-detected case, the defect identification result is associated and mapped with the corresponding production batch information, which includes batch number, production line number, equipment number and process parameters; The mapping relationship is wherein, is the i-th defect, is the j-th batch, and SN is the product serial number on the case.

[0012] Optionally, the step of statistically analyzing defects by production batch based on the data in the defect result database, generating a warning signal and process optimization suggestion when the defect rate of a batch exceeds a preset threshold includes: Calculate the defect rate of each defect category in each production batch, compare the defect rate with the preset threshold, and generate a warning signal according to the corresponding defect category for the batch with a defect rate greater than the preset threshold; Analyze the defect type distribution in the batch in the warning signal, and analyze the defect type distribution and the main process parameters used in the batch, including stamping pressure, riveting force and spraying temperature; When the defect rate of a certain category of defects in several batches is higher than the preset threshold, and there is a strong correlation with the value range of a certain process parameter, an optimization report containing specific suggestions for adjusting the process parameter is generated.

[0013] According to the second aspect of the embodiments of the present specification, a case structure defect identification system is provided, which includes an image acquisition module, an image processing module, a defect identification module and a defect analysis module, wherein The image acquisition module is configured to acquire the production batch information of the to-be-detected case, and acquire image sequences of the to-be-detected case under multiple angles and different lighting conditions through a pre-set image acquisition device; The image processing module is configured to pre-process the image sequence, and the pre-processing includes illumination normalization, anti-reflection suppression processing and data enhancement processing; The defect identification module is configured to input the preprocessed image sequence into a pre-trained defect identification model, and the defect identification model performs calculation according to the input image and outputs a defect identification result, the defect identification result including a defect type, a position coordinate and a confidence level. The defect analysis module is configured to associate the defect identification result with production batch information of the machine case to be detected, and store the information into a defect result database; based on data in the defect result database, the defect is statistically analyzed according to the production batch, and when the defect rate of a batch exceeds a preset threshold, a warning signal and a process optimization suggestion are generated.

[0014] According to a third aspect of the embodiments of the present specification, a computing device is provided, including a storage device for storing a computer program, and a processor for running the computer program to enable the computing device to perform the steps of the machine case structure defect identification method.

[0015] According to a fourth aspect of the embodiments of the present specification, a storage medium is provided, which stores a computer program used in the computing device, and the computer program is executed by a processor to implement the steps of the machine case structure defect identification method.

[0016] The beneficial effects of the embodiments of the present specification are as follows: The embodiments of the present specification provide a machine case structure defect identification method, system, device and medium, which can automatically detect and classify various defects such as scratches, misplacement, hole displacement, plate warping, etc. by introducing a convolutional neural network for image recognition of the surface and key parts of the machine case, and associate the defect detection result with the manufacturing batch information to realize defect production batch traceability and process improvement suggestion output, solve the deficiencies of traditional methods in precision, speed and traceability, thereby significantly improve the detection efficiency and accuracy, reduce the labor cost, and improve the production quality management level.

[0017] The innovations of the embodiments of the present specification include: 1. In the present specification, the image sequence is subjected to illumination normalization, anti-reflection suppression processing and data enhancement processing, which solves the problem of sensitivity to illumination and low detection precision of traditional detection methods, which is one of the innovations of the embodiments of the present specification.

[0018] 2. In the present specification, the defect identification result is associated with the production batch information of the machine case to be detected, which realizes the association of defects with batches and production processes, so that the defect identification result can be analyzed according to the production batch, and the process causes of defects can be traced and analyzed, which not only ensures the precision of defect detection, but also improves the process to reduce defects, which is one of the innovations of the embodiments of the present specification.

[0019] 3、In the specification, the combination scheme of convolutional neural network deep feature extraction and batch trace analysis is introduced in the chassis defect detection, the convolutional neural network model is used for multi-defect type recognition and classification, and the mapping relationship between the defect record and the production batch is realized. The automatic early warning of the abnormal production process parameters is realized. This combination not only improves the accuracy and automation of detection, but also directly shortens the positioning time of production problems, which is one of the innovations of the embodiments of the specification. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the specification or the related art, the drawings needed to be used in the embodiment or related art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0021] Figure 1 A flowchart of a chassis structure defect recognition method provided by an embodiment of the specification is shown in the figure. Figure 2 A structural diagram of a chassis structure defect recognition system provided by an embodiment of the specification is shown in the figure. Figure 3 A structural diagram of a computing device provided by an embodiment of the specification is shown in the figure. Figure 4 A structural diagram of a storage medium provided by an embodiment of the specification is shown in the figure. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the specification will be described clearly and completely below with reference to the drawings in the embodiments of the specification. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0023] It should be noted that the terms "include" and "have" and any variations thereof in the embodiments of the specification and the drawings are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but optionally includes other steps or units not listed or optionally includes other steps or units inherent to the process, method, product or device.

[0024] With the rapid development of deep learning, especially Convolutional Neural Network (CNN), image recognition technology based on deep learning has shown high accuracy and robustness in industrial quality inspection. However, existing deep learning quality inspection solutions are mostly focused on flat products (such as PCB, cloth, steel plate, etc.), and there are still challenges in the detection of complex three-dimensional structures and reflective metal surfaces.

[0025] For example, surface reflection and illumination variation interference, metal cases reflect strongly under different illumination angles, affecting the extraction of defect texture features by the model; structural diversity and detection position complexity, the structure of different cases is quite different, and a single detection angle is difficult to cover all key areas; defect types are diversified, including surface scratches, assembly misalignment, deformation, screw hole offset, uneven coating, etc., which requires the model to have multi-class feature recognition capability.

[0026] The embodiments of the present specification disclose a case structure defect recognition method, system, device and medium, which are described in detail below.

[0027] Figure 1 A flowchart of a case structure defect recognition method provided by an embodiment of the present specification is shown in FIG. 1. Figure 1 As shown in FIG. 1, a case structure defect recognition method includes: S110, obtaining production batch information of a case to be detected, and obtaining image sequences of the case to be detected under multiple angles and different illumination conditions through a pre-set image acquisition device.

[0028] The pre-set image acquisition device includes an industrial camera, a light supplement system and a motion platform, wherein the light supplement system adopts a combination of a ring-shaped LED light source and a polarizing filter, and the acquisition step is: Place the case to be detected in the shooting area, control the industrial camera to move on the motion platform according to the preset speed and path, acquire images of the six outer surfaces and key interface areas of the case to be detected, and change the incident angle of the LED light source at each acquisition point to obtain multiple images under different illumination conditions at the same position, thereby obtaining an image sequence.

[0029] Industrial camera: resolution ≥ 12 million pixels, frame rate ≥ 30 FPS.

[0030] Lens: focal length 25mm, distortion rate <0.5%.

[0031] The light supplement system adopts a combination of a ring-shaped LED light source and a polarizing filter to suppress metal surface reflection.

[0032] Motion platform: three-axis electric sliding platform, control the camera to move in X / Y / Z direction, ensure to cover each detection area of the case.

[0033] Case detection adopts multi-angle acquisition: 8 angle images are collected for each case, corresponding to the four sides, top, bottom and key interface area of the case respectively.

[0034] Light diversification strategy is adopted during acquisition: by changing the incident angle of LED light source (0°, 30°, 60°), the defect features under different light are captured, so as to enhance the robustness of the model.

[0035] S120, pre-processing the image sequence, the pre-processing including illumination normalization, anti-reflection suppression processing and data enhancement processing.

[0036] The pre-processing of the image sequence includes illumination normalization, anti-reflection suppression processing and data enhancement processing, which includes: Retinex image enhancement algorithm is used for illumination normalization processing, the formula is Wherein, is the original pixel value, is the low-pass filter kernel, is the pixel value after normalization processing; Polarization filter combined with image polarization decomposition algorithm is used to extract non-reflective components in the image for anti-reflection suppression. Through random rotation, translation, contrast adjustment and Gaussian noise addition, data enhancement is carried out.

[0037] S130, input the pre-processed image sequence into the pre-trained defect recognition model, calculate and output the defect recognition result according to the input image by the defect recognition model, the defect recognition result including defect type, position coordinates and confidence.

[0038] The defect recognition model is a convolutional neural network model based on ResNet-50 architecture, and a Softmax classifier is connected after the last fully connected layer.

[0039] The model structure includes: input layer (224×224×3), convolution layer + BN layer + ReLU, residual module (4 stages, 50 layers deep), global average pooling layer, fully connected layer (512 units), Softmax output layer (defect type number + 1 class "no defect").

[0040] The loss function formula is as follows: wherein N is the number of samples, C is the number of categories, is the true label, is the model prediction probability.

[0041] After the step of calculating and outputting the defect identification result by the defect identification model according to the input image, the method further comprises: The defect area is located by the defect position coordinates through the gradient weighted class activation mapping, and a class activation heat map identifying the defect position is generated, and on the basis of the CNN classification, the step of realizing the defect positioning comprises: obtaining the gradient of the target category ; calculating the weight: wherein Z is the total number of categories; generating a heat map: , realizing the defect positioning.

[0042] S140, associating the defect identification result with the production batch information of the to-be-detected case, and storing to a defect result database.

[0043] The step of associating the defect identification result with the production batch information of the to-be-detected case comprises: By scanning the product serial number on the to-be-detected case, the defect identification result is associated and mapped with the corresponding production batch information, and the production batch information comprises a batch number, a production line number, an equipment number and a process parameter; The mapping relationship is wherein, is the i-th defect, is the j-th batch, and SN is the product serial number on the case.

[0044] The defect detection result comprises a defect type, a position coordinate and a confidence. The production batch information comprises a batch number, a production date, a production line number, an equipment number, an operator and a process parameter (such as stamping pressure, riveting force, spraying temperature, etc.). The mapping relationship between the defect data and the batch information can be realized through the product serial number in the production execution system (MES). Through this mapping, all detected defects can be grouped by batch.

[0045] S150, based on the data in the defect result database, statistically analyzing the defects according to the production batch, when the defect rate of a batch exceeds a preset threshold, generating an early warning signal and a process optimization suggestion.

[0046] In specific embodiments, the step of statistically analyzing the defect identification result comprises: calculating a defect rate of each defect category in each production batch, comparing the defect rate with a preset threshold, and generating a warning signal according to the corresponding defect category for a batch with a defect rate greater than the preset threshold; analyzing the defect type distribution in the batch in the warning signal, and performing correlation analysis on the defect type distribution and main process parameters used in the batch, the process parameters including stamping pressure, riveting force, and spraying temperature; When the defect rate of a certain category of defects in several batches is higher than the preset threshold, and there is a strong correlation with the value range of a certain process parameter, an optimization report containing specific suggestions for adjusting the process parameter is generated.

[0047] The defect rate of each defect category in each production batch is calculated: wherein, is the number of cases in the batch with defects, is the total number of cases in the batch .

[0048] When exceeds the preset threshold, for example, 3%, a process abnormality warning is triggered.

[0049] In combination with the defect type distribution and the process parameters, optimization suggestions are given, for example: If scratch defects are concentrated on the same spraying production line, and the spraying temperature is too high, it is recommended to adjust the spraying temperature.

[0050] If the screw hole offset defect is concentrated in a certain stamping equipment, it is recommended to check the mold positioning accuracy of the equipment.

[0051] Figure 2 A structural diagram of a case structure defect identification system according to an embodiment of the present specification. As shown in Figure 2 A case structure defect identification system 200 includes an image acquisition module 210, an image processing module 220, a defect identification module 230, and a defect analysis module 240, wherein The image acquisition module 210 is configured to acquire production batch information of a case to be detected, and acquire image sequences of the case to be detected at multiple angles and under different lighting conditions through a pre-set image acquisition device.

[0052] The image processing module 220 is configured to pre-process the image sequences, and the pre-processing includes illumination normalization, anti-reflection suppression processing, and data enhancement processing.

[0053] The defect identification module 230 is configured to input the preprocessed image sequence into a pre-trained defect identification model, calculate and output a defect identification result according to the input image by the defect identification model, and the defect identification result includes a defect type, a position coordinate and a confidence.

[0054] The defect analysis module 240 is configured to associate the defect identification result with production batch information of the to-be-detected case and store it into a defect result database; based on data in the defect result database, statistically analyze defects according to production batches, and when a defect rate of a certain batch exceeds a preset threshold, generate a warning signal and a process optimization suggestion.

[0055] Figure 3 A structural schematic diagram of a computing device is provided for an embodiment of the present specification. As shown in the figure, a computing device 300 includes a storage device 310 for storing a computer program and a processor 320 for running the computer program to make the computing device 300 execute the steps of the case structure defect identification method. Figure 3

[0056] Figure 4 A structural schematic diagram of a storage medium is provided for an embodiment of the present specification. As shown in the figure, a storage medium 400 stores a computer program 410 used in the computing device, and the computer program 410 is executed by a processor to implement the steps of the case structure defect identification method. Figure 4

[0057] In summary, the embodiments of the present specification provide a case structure defect identification method, system, device and medium, which not only solves the deficiencies of traditional methods in precision, speed and traceability, but also organically combines deep learning detection with manufacturing process optimization to form a complete closed-loop quality management system from defect detection to production improvement, which has significant industrial application value and promotion prospect.

[0058] Those skilled in the art can understand that the drawings are only schematic diagrams of an embodiment, and the modules or flows in the drawings are not necessarily necessary for implementing the present application.

[0059] Those skilled in the art can understand that the modules in the device in the embodiments can be distributed in the device in the embodiments as described in the embodiments, or can be changed and located in one or more devices different from the present embodiments. The modules of the above embodiments can be combined into one module, or can be further split into multiple sub-modules.

[0060] ​​Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the same; although the present application has been described in detail with reference to the foregoing examples, those of ordinary skill in the art should understand that the technical solutions recorded in the foregoing examples can still be modified, or some of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A chassis structure defect recognition method characterized by comprising: The method comprises the following steps: Obtain the production batch information of the to-be-detected machine case, and acquire image sequences of the to-be-detected machine case under multiple angles and different illumination conditions through a pre-set image acquisition device; Preprocess the image sequences, wherein the preprocessing comprises illumination normalization, reflection suppression processing and data enhancement processing; Input the preprocessed image sequences into a pre-trained defect recognition model, calculate and output a defect recognition result according to the input image by the defect recognition model, wherein the defect recognition result comprises a defect type, a position coordinate and a confidence level; Associate the defect recognition result with the production batch information of the to-be-detected machine case, and store the result in a defect result database; Based on the data in the defect result database, statistically analyze the defects according to the production batch, and when the defect rate of a certain batch exceeds a pre-set threshold, generate a warning signal and a process optimization suggestion.

2. The method of claim 1, wherein, The pre-set image acquisition device comprises an industrial camera, a light supplement system and a motion platform, wherein the light supplement system adopts a ring-shaped LED light source combined with a polarization filter, and the acquisition step is as follows: Place the to-be-detected machine case in a shooting area, control the industrial camera to move on the motion platform at a pre-set speed and path, acquire images of six outer surfaces and key interface areas of the to-be-detected machine case, and change the incident angle of the LED light source at each acquisition point to acquire multiple images of the same position under different illumination conditions, thereby obtaining the image sequences.

3. The method of claim 1, wherein, The preprocessing of the image sequences comprises the following steps: Perform illumination normalization processing by using a Retinex image enhancement algorithm, and the formula is as follows: wherein, is the original pixel value, is a low-pass filter kernel, is the normalized pixel value; Extract the non-reflection component in the image by using a polarization filter combined with an image polarization decomposition algorithm to suppress reflection; Perform data enhancement by means of random rotation, translation, contrast adjustment and Gaussian noise addition.

4. The method of claim 1, wherein, The defect recognition model is a convolutional neural network model based on a ResNet-50 architecture, and a Softmax classifier is connected after the last fully connected layer, and the loss function formula is as follows: where N is the number of samples, C is the number of classes, is the true label, is the model prediction probability.

5. The method of claim 1, wherein, After the step of calculating and outputting the defect recognition result by the defect recognition model according to the input image, the following steps are further included: Generate a class activation heat map for identifying the defect position by positioning the defect area through gradient weighted class activation mapping, and the steps of realizing defect positioning based on CNN classification and introducing Grad-CAM (gradient weighted class activation mapping) are as follows: obtaining a gradient of the target class ; Computing the weights: where Z is the total number of classes; Generate a heat map: , to realize defect positioning.

6. The method of claim 1, wherein, The step of associating the defect recognition result with the production batch information of the to-be-detected machine case comprises the following steps: Map the defect recognition result to the corresponding production batch information by scanning the product serial number on the to-be-detected machine case, wherein the production batch information comprises a batch number, a production line number, an equipment number and process parameters; The mapping relationship is as follows: wherein, is the ith defect, is the jth batch, and SN is the product serial number on the chassis.

7. The method of claim 1, wherein, The steps of statistically analyzing the defects according to the production batch based on the data in the defect result database, and generating a warning signal and a process optimization suggestion when the defect rate of a certain batch exceeds a pre-set threshold comprise the following steps: Calculate the defect rate of each defect category in each production batch, compare the defect rate with the preset threshold, and generate a warning signal according to the corresponding defect category for the batch with a defect rate greater than the preset threshold; Analyze the defect type distribution in the batch in the warning signal, and perform correlation analysis on the defect type distribution and the main process parameters used in the batch, including stamping pressure, riveting force, and spraying temperature; When the defect rate of a certain category of defects in several batches is higher than the preset threshold, and there is a strong correlation with the value range of a certain process parameter, an optimization report containing specific suggestions for adjusting the process parameter is generated.

8. A chassis structure defect recognition system characterized by comprising: It comprises an image acquisition module, an image processing module, a defect identification module and a defect analysis module, wherein The image acquisition module is configured to acquire the production batch information of the machine case to be detected, and acquire image sequences of the machine case to be detected under multiple angles and different lighting conditions through a pre-set image acquisition device; The image processing module is configured to preprocess the image sequences, and the preprocessing includes illumination normalization, anti-reflection suppression processing and data enhancement processing; The defect identification module is configured to input the preprocessed image sequences into a pre-trained defect identification model, and calculate and output defect identification results according to the input images by the defect identification model, wherein the defect identification results include defect type, position coordinates and confidence; The defect analysis module is configured to associate the defect identification results with the production batch information of the machine case to be detected, and store them in a defect result database; Based on the data in the defect result database, the defects are statistically analyzed according to the production batch, and when the defect rate of a certain batch exceeds the preset threshold, a warning signal and process optimization suggestions are generated.

9. A computing device, comprising: The storage device is used to store a computer program, and the processor runs the computer program to make the computing device execute the steps of the method of any one of claims 1-7.

10. A storage medium, characterized by It stores the computer program used in the computing device of claim 9, which is executed by the processor to realize the steps of the method of any one of claims 1-7.

Images