Welding spot failure identification method and device

Abstract

The invention provides a welding spot failure identification method, which belongs to the field of finished automobile part quality testing, and comprises the following steps: firstly, carrying out a welding spot stripping test on a preset part by utilizing a stretcher until a welding spot in the preset part fails, establishing an identification model by taking an image of the failed welding spot as a reference image, and carrying out model learning and iteration; according to the method, the welding spot stripping experiment and the model learning are utilized, so that the recognition model capable of finally recognizing the welding spot failure is obtained, and after the to-be-recognized image containing the to-be-judged welding spot is input into the recognition model, whether the second welding spot fails or not can be judged according to the output recognition result. Therefore, the efficiency and accuracy of welding spot failure identification can be improved.

Description

Technical Field

[0001] The present application relates to the field of quality testing of vehicle parts, and in particular to a method and device for identifying weld failure. Background Art

[0002] Resistance spot welding is a widely used production process and the main connection method for automobile bodies and other components. It is an efficient, economical and easy-to-automate metal connection technology, playing an irreplaceable and important role in the automotive manufacturing industry.

[0003] The safety performance of a car depends not only on the strength of the material itself, but also on the quality of the connections between different materials, which also affects the rigidity and safety of the entire vehicle structure. Therefore, it is of great practical significance to detect and ensure the connection quality of the welds produced by electric welding.

[0004] Traditional solder joint failure assessment relies on visual evaluation by quality inspectors, but manual quality inspection can easily lead to loss of inspection quality or efficiency due to fatigue and other factors. Summary of the Invention

[0005] In view of this, the present application provides a method and device for identifying solder joint failures, which can improve detection quality and efficiency.

[0006] In one aspect, the present application provides a method for identifying solder joint failure, the method comprising:

[0007] A weld peeling test is performed on a preset part using a stretching machine until the first weld in the preset part fails and becomes a failed weld.

[0008] A first reference image containing the failed solder joint is acquired.

[0009] Perform image processing on the first reference image to obtain a second reference image.

[0010] A recognition model is established based on the plurality of second reference images.

[0011] A first image to be identified including the second welding point is acquired.

[0012] Perform image processing on the first image to be recognized to obtain a second image to be recognized.

[0013] The second image to be recognized is input into the recognition model to obtain an output recognition result, which indicates whether the second welding point is invalid.

[0014] Optionally, performing a failure test on a predetermined part by using a stretching machine until a first weld point in the predetermined part fails and becomes a failed weld point includes:

[0015] A tensile test machine is used to perform one of a shear test, a cross tensile test, and a peel test on the preset part until a first weld point in the preset part fails and becomes a failed weld point.

[0016] Optionally, performing image processing on the first reference image to obtain the second reference image includes:

[0017] The first reference image is sequentially subjected to erosion processing, dilation processing, median filtering, and debinarization processing to obtain a first intermediate image.

[0018] The minimum bounding rectangle of the second welding point in the first intermediate image is determined, and the image within the minimum bounding rectangle is used as the second intermediate image.

[0019] The size of the second intermediate image is stretched to a preset size to obtain a second reference image.

[0020] Optionally, after taking the image in the minimum bounding rectangle as the second intermediate image, the method further comprises:

[0021] The size of the second intermediate image is stretched to a preset size to obtain a second original reference image.

[0022] The second original reference image is subjected to horizontal mirror transformation, vertical mirror transformation, and flip transformation to obtain a plurality of transformed expanded images.

[0023] The second original reference and the plurality of augmented images are used as second reference images.

[0024] Optionally, establishing the recognition model according to the second reference image includes:

[0025] A plurality of second reference images are acquired.

[0026] The plurality of second reference images are divided into training images and validation images.

[0027] Build a neural network model.

[0028] The training image and verification image are used as the input of the neural network model, and the neural network model is iteratively trained to obtain a set of model parameters with the highest recognition accuracy.

[0029] Substitute the model parameters into the neural network model to obtain the recognition model.

[0030] On the other hand, the present application also provides a device for identifying solder joint failure, the device comprising:

[0031] The acquisition module is configured to acquire a first reference image including a failed solder joint, wherein the first reference image is obtained after a failure test is performed on a preset part using a stretching machine until a first solder joint in the preset part fails and becomes a failed solder joint.

[0032] The processing module is configured to perform image processing on the first reference image to obtain a second reference image.

[0033] The establishing module is configured to establish a recognition model according to a plurality of second reference images.

[0034] The acquisition module is further configured to acquire a first image to be identified that includes the second welding point.

[0035] The processing module is further configured to perform image processing on the first image to be identified to obtain a second image to be identified.

[0036] The recognition module is configured to input the second image to be recognized into the recognition model to obtain an output recognition result, and the recognition result indicates whether the second welding point is invalid.

[0037] Optionally, performing a failure test on a predetermined part by using a stretching machine until a first weld point in the predetermined part fails and becomes a failed weld point includes:

[0038] A tensile test machine is used to perform one of a shear test, a cross tensile test, and a peel test on the preset part until a first weld point in the preset part fails and becomes a failed weld point.

[0039] Optionally, performing image processing on the first reference image to obtain the second reference image includes:

[0040] The first reference image is sequentially subjected to erosion processing, dilation processing, median filtering, and debinarization processing to obtain a first intermediate image.

[0041] The minimum bounding rectangle of the second welding point in the first intermediate image is determined, and the image within the minimum bounding rectangle is used as the second intermediate image.

[0042] The size of the second intermediate image is stretched to a preset size to obtain a second reference image.

[0043] Optionally, performing image processing on the first reference image to obtain the second reference image further includes:

[0044] After the image in the minimum circumscribed rectangle is used as the second intermediate image, the size of the second intermediate image is stretched to a preset size to obtain a second original reference image.

[0045] The second original reference image is subjected to horizontal mirror transformation, vertical mirror transformation, and flip transformation to obtain a plurality of transformed expanded images.

[0046] The second original reference and the plurality of augmented images are used as second reference images.

[0047] Optionally, the build module is configured to:

[0048] A plurality of second reference images are acquired.

[0049] The plurality of second reference images are divided into training images and validation images.

[0050] Build a neural network model.

[0051] The training image and verification image are used as the input of the neural network model, and the neural network model is iteratively trained to obtain a set of model parameters with the highest recognition accuracy.

[0052] Substitute the model parameters into the neural network model to obtain the recognition model.

[0053] The solder joint failure identification method provided in the present application is firstly performed a solder joint peeling test on a preset part using a stretching machine until the solder joint in the preset part fails, and a recognition model is established based on the image of the failed solder joint as a reference image, and the model is learned and iterated, thereby finally obtaining a recognition model that can identify solder joint failure. After the image to be identified containing the solder joint to be judged is input into the recognition model, whether the second solder joint has failed can be judged based on the output recognition result. Due to the use of solder joint peeling experiments and model learning, the efficiency and accuracy of solder joint failure identification can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0055] Figure 1 A flow chart of a solder joint failure identification method provided in an embodiment of the present application;

[0056] Figure 2 Another flow chart of the solder joint failure identification method provided in an embodiment of the present application;

[0057] Figures 3a-3c A schematic diagram of a failure experiment of the solder joint failure identification method provided in an embodiment of the present application;

[0058] Figure 4 This is a diagram of the architecture of the solder joint failure identification device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0060] The present application embodiment provides a method for identifying solder joint failure, such as Figure 1 As shown, the method includes steps S101, S102, S103, S104, S105, S106 and S107, wherein:

[0061] In step S101, a weld peeling test is performed on a preset part using a stretching machine until a first weld in the preset part fails and becomes a failed weld.

[0062] In step S102 , a first reference image including a failed solder joint is acquired.

[0063] In step S103 , image processing is performed on the first reference image to obtain a second reference image.

[0064] In step S104 , a recognition model is established according to the plurality of second reference images.

[0065] In step S105 , a first image to be recognized including the second welding point is acquired.

[0066] In step S106 , image processing is performed on the first image to be recognized to obtain a second image to be recognized.

[0067] In step S107 , the second image to be recognized is input into the recognition model to obtain an output recognition result, which indicates whether the second welding point is invalid.

[0068] In some optional embodiments, performing a failure test on a predetermined part using a stretching machine until a first weld in the predetermined part fails and becomes a failed weld includes:

[0069] A tensile test machine is used to perform one of a shear test, a cross tensile test, and a peel test on the preset part until a first weld point in the preset part fails and becomes a failed weld point.

[0070] In some optional embodiments, performing image processing on the first reference image to obtain the second reference image includes:

[0071] The first reference image is sequentially subjected to erosion processing, dilation processing, median filtering, and debinarization processing to obtain a first intermediate image.

[0072] The minimum bounding rectangle of the second welding point in the first intermediate image is determined, and the image within the minimum bounding rectangle is used as the second intermediate image.

[0073] The size of the second intermediate image is stretched to a preset size to obtain a second reference image.

[0074] In some optional embodiments, after using the image within the minimum bounding rectangle as the second intermediate image, the method further includes:

[0075] The size of the second intermediate image is stretched to a preset size to obtain a second original reference image.

[0076] The second original reference image is subjected to horizontal mirror transformation, vertical mirror transformation, and flip transformation to obtain a plurality of transformed expanded images.

[0077] The second original reference and the plurality of augmented images are used as second reference images.

[0078] In some optional embodiments, establishing a recognition model according to the second reference image includes:

[0079] A plurality of second reference images are acquired.

[0080] The plurality of second reference images are divided into training images and validation images.

[0081] Build a neural network model.

[0082] The training image and verification image are used as the input of the neural network model, and the neural network model is iteratively trained to obtain a set of model parameters with the highest recognition accuracy.

[0083] Substitute the model parameters into the neural network model to obtain the recognition model.

[0084] Using the solder joint failure identification method provided in the present application, a solder joint peeling test is first performed on a preset part using a stretching machine until the solder joint in the preset part fails and becomes a failed solder joint, and a recognition model is established based on the image of the failed solder joint as a reference image. After the image to be identified containing the solder joint to be judged is input into the recognition model, an output recognition result is obtained, and the recognition result indicates whether the second solder joint has failed. Due to the use of the solder joint peeling experiment and model learning, the efficiency and accuracy of solder joint failure identification can be improved.

[0085] The present application embodiment provides a method for identifying solder joint failure, such as Figure 2 As shown, the method includes steps S201, S202, S203, S204, S205, S206, S207, S208, S209, S210 and S211, wherein:

[0086] In step S201, a weld peeling test is performed on a preset part using a stretching machine until a first weld in the preset part fails and becomes a failed weld.

[0087] In some optional embodiments, performing a failure test on a predetermined part using a stretching machine until a first weld in the predetermined part fails and becomes a failed weld includes:

[0088] A tensile test machine is used to perform one of a shear test, a cross tensile test, and a peel test on the preset part until a first weld point in the preset part fails and becomes a failed weld point.

[0089] like Figure 3a As shown, the shear test refers to pulling the parallel overlapped welded parts 301a and 302a respectively, so that the first weld 303a fails and becomes a failed weld.

[0090] like Figure 3b As shown, the cross tensile test refers to pulling the vertically overlapped welded parts 301b and 302b respectively, so that the first weld 303b fails and becomes a failed weld.

[0091] like Figure 3c As shown, the peel test refers to pulling the bent and overlapped welded parts 301c and 302c respectively, so that the first welding point 303c fails and becomes a failed welding point.

[0092] An electronic universal testing machine can be used to perform a tensile test, with the tensile speed set at 5 mm / min, and the test is performed until the solder joint fails and separates.

[0093] In step S202 , a first reference image including a failed solder joint is acquired.

[0094] It is understood that there are three typical failure types, including interface tearing, complete nugget extraction, and partial nugget extraction. Interface tearing refers to the nugget of the weld point of two welded parts being torn into two halves, with the two halves remaining on the two parts. Complete nugget extraction refers to the nugget of the weld point of two welded parts being extracted, with only one of the parts remaining. Partial nugget extraction refers to the nugget of the weld point of two welded parts being partially extracted, with the extracted portion remaining on one of the parts, and the unextracted portion being very small compared to the extracted portion, remaining on the other part.

[0095] For failure types of complete or partial nugget extraction, image acquisition is performed on the side with more nugget attachment to obtain a first reference image containing the failed solder joint. For failure types of interface tearing, image acquisition is performed on either side to obtain a first reference image containing the failed solder joint.

[0096] In step S203, image processing is performed on the first reference image to obtain a second reference image.

[0097] In some optional embodiments, performing image processing on the first reference image to obtain the second reference image includes:

[0098] The first reference image is sequentially subjected to erosion processing, dilation processing, median filtering, and debinarization processing to obtain a first intermediate image, where:

[0099] Before the erosion process, a structuring element can also be defined. This generates a 3×3 matrix structuring element, where all elements are 1. This is used for subsequent image erosion and dilation. The structuring element is moved over the first reference image, with its center defined as the anchor point. The pixel values of the first reference image covered by the structuring element are calculated to replace the pixels at the anchor point.

[0100] The erosion process takes the minimum value within the range specified by the structuring element as the output grayscale value at that location. Because the minimum value within the range is taken for each location, the average overall brightness of the eroded output image will be lower than that of the original image. The brighter areas in the image will become smaller or even disappear, while the darker objects will be larger.

[0101] Dilation processing refers to taking the maximum value within the area specified by the structure element as the output grayscale value at that location. The dilation operation merges the background points that touch the current object (foreground) into the current object, thereby expanding the boundary points of the image outward.

[0102] Median filtering refers to selecting a 3×3 neighborhood window of a pixel in an image, arranging the grayscale values in the window from small to large, and taking the middle grayscale value as the grayscale value of the entire window. This process can eliminate isolated noise points and make the pixel value closer to the true value.

[0103] Debinarization means that for pixels with grayscale values greater than 80, their values are set to 0. For pixels with grayscale values less than or equal to 80, their values are set to 255.

[0104] After obtaining the first intermediate image, the minimum bounding rectangle of the second welding point in the first intermediate image is further determined, and the image within the minimum bounding rectangle is used as the second intermediate image.

[0105] It is understood that a contour detection algorithm can be used to find the minimum bounding rectangle of the second weld point in the first intermediate image. Based on the obtained minimum bounding rectangle, the coordinates of the upper left corner of the rectangle and the width and height of the rectangle are obtained. The second intermediate image within the rectangular area is cropped based on these values.

[0106] After obtaining the second intermediate image, the size of the second intermediate image is further stretched to a preset size to obtain a second reference image.

[0107] It's understandable that images can be uniformly stretched and resized to 100×100 pixels. Image resizing is a crucial step in model training, helping the model learn more efficiently, improving performance, and optimizing the training process. In practical applications, choosing the appropriate resizing strategy depends on the specific task and model requirements. This is particularly important in convolutional neural networks. Resizing ensures that all images are resized to the network's required size before being fed into the network.

[0108] To ensure higher accuracy and less overfitting in subsequent training of recognition models, a large amount of training data is typically required. However, in practice, obtaining large amounts of data is difficult, so dataset augmentation is necessary. This augmented dataset more closely reflects real-world diversity, enabling the model to better generalize to new, unseen data. Common image data augmentation methods include horizontal mirroring, vertical mirroring, and flipping. These operations augment solder joint images while ensuring that features such as defect morphology remain invariant to rotation and scale, ensuring sufficient solder joint data for model training.

[0109] Therefore, in some optional embodiments, after using the image in the minimum bounding rectangle as the second intermediate image, the method further includes:

[0110] The size of the second intermediate image is stretched to a preset size to obtain a second original reference image. It can be understood that the image size can be uniformly stretched and scaled to 100×100 pixels.

[0111] The second original reference image is subjected to horizontal mirror transformation, vertical mirror transformation, and flip transformation to obtain a plurality of transformed expanded images.

[0112] The second original reference image is horizontally mirrored to obtain a first extended image, vertically mirrored to obtain a second extended image, and flipped to obtain a third extended image, thereby obtaining a plurality of extended images after transformation.

[0113] The second original reference and the plurality of augmented images are used as second reference images.

[0114] In step S204 , a plurality of second reference images are acquired.

[0115] In step S205 , the plurality of second reference images are divided into training images and verification images.

[0116] In some optional embodiments, 80% of the second reference images in the plurality of second reference images may be used as training images, and 20% of the second reference images may be used as verification images.

[0117] In step S206, a neural network model is established.

[0118] It can be understood that building a neural network model involves using a CNN model to extract local image features through convolutional layers and reduce the spatial dimension of these features through pooling layers. Convolutional neural networks offer a great deal of flexibility, as the number of convolutional, pooling, and fully connected layers can be freely set, along with flexible parameters such as activation functions, kernel size, and stride.

[0119] During model building, we must prevent overfitting of the training data, which can lead to a decrease in the model's generalization ability on new data. When the data volume is small, models with overly deep network architectures should not be used. Practical analysis shows that using the AlexNet model structure can effectively avoid heavy computational loads and slow training.

[0120] The AlexNet model uses an eight-layer neural network architecture. The first five layers are convolutional layers C1, C2, C3, C4, and C5, and the last three layers are fully connected layers FC6, FC7, and FC8. The ReLU activation function is used in the hidden layers of AlexNet, allowing for deeper training and significantly improving training speed. Furthermore, the network incorporates a Dropout layer to effectively prevent overfitting. AlexNet employs a stacked pooling operation, where the pooling size is greater than the stride. This operation enables information exchange between adjacent pixels while preserving necessary connections.

[0121] In step S207, the training image and the verification image are used as inputs of the neural network model, and the neural network model is iteratively trained to obtain a set of model parameters with the highest recognition accuracy.

[0122] In step S208, the model parameters are substituted into the neural network model to obtain a recognition model.

[0123] In some optional embodiments, after obtaining the recognition model, the model may be evaluated:

[0124] Specifically, loss functions and evaluation functions can be used to evaluate the quality of the model and perform model optimization.

[0125] In some optional embodiments, a loss function can be used to calculate the degree of similarity between the model's predicted value and the expected value. The smaller the loss value, the better the model. There are many types of loss functions, and different loss functions are used for different problems. For regression problems, mean squared error or mean absolute error are often used. For binary classification problems, binary cross entropy can be used to evaluate the model. For multi-classification problems, multi-class cross entropy can be used to evaluate the degree of similarity between the expected output value and the predicted value.

[0126] In some optional embodiments, both the evaluation function and the loss value can be used to evaluate the quality of the model. In regression problems, the mean absolute error can be directly used to evaluate the quality of the model. However, in practice, the coefficient of determination (R2) is more commonly used to evaluate the predictive performance of the regression model. Its formula is shown in Equation (1). R2 represents the proportion of data variance that the model can explain. Simply put, R2 is a maximum of 1, and the larger the R2, the better the model performance.

[0127]

[0128] Where, represents the regression value of the dependent variable, that is, the predicted value, represents the mean of the dependent variable, and yi represents the i-th dependent variable.

[0129] While both loss functions and evaluation functions can evaluate model performance, the loss value provided by the loss function is less intuitive and is more suitable for evaluating whether a model has achieved optimal performance during training. In practical applications, however, it is often necessary to evaluate the performance of models with different types and parameters. Evaluation functions provide a more intuitive understanding of model performance, making them more convenient for selecting the optimal model.

[0130] Preferably, model optimization in the CNN model involves adjusting configurable parameters such as activation function, weight initialization, batch normalization, overfitting elimination, optimizer selection, learning rate, training batch size, and number of training runs. Model optimization is achieved by adjusting these parameters. After searching online, the TensorBoard tool can be used to visualize the model training accuracy under different parameter combinations. The model training results for different parameter combinations plotted by TensorBoard show that the optimal model achieves an accuracy of 98.18%.

[0131] In some optional embodiments, multi-classification cross entropy is used as the loss function, and classification accuracy is used as an indicator to evaluate the quality of the model. The calculation method is shown in formula (2):

[0132]

[0133] Where Correctnumbers is the number of correct predictions and Totalnumbers is the number of all data.

[0134] Model optimization involved a grid search for batch normalization, weight initialization, learning rate, training batch size, and number of training epochs. The model parameters that achieved the highest classification accuracy were selected, ultimately set to eliminate batch normalization, initialize weights using He_normal, set the learning rate to 0.001, train the batch size to 64, and train epochs to 200.

[0135] In step S209 , a first image to be recognized including the second welding point is acquired.

[0136] In step S210 , image processing is performed on the first image to be recognized to obtain a second image to be recognized.

[0137] In step S211, the second image to be recognized is input into the recognition model to obtain an output recognition result, which indicates whether the second welding point is invalid.

[0138] The solder joint failure identification method provided in the present application is firstly performed a solder joint peeling test on a preset part using a stretching machine until the solder joint in the preset part fails and becomes a failed solder joint, and a neural network model is established based on the image of the failed solder joint as a reference image, and the training image and the verification image are used as inputs of the neural network model. The neural network model is iteratively trained to obtain a recognition model, and then the image to be identified containing the solder joint to be judged is input into the recognition model to obtain an output recognition result, which indicates whether the second solder joint has failed. Due to the use of the solder joint peeling experiment and model learning iteration, the efficiency and accuracy of solder joint failure identification can be improved.

[0139] The present application also provides a device for identifying solder joint failures. Figure 4 As shown, the device includes:

[0140] The acquisition module 401 is configured to acquire a first reference image including a failed weld spot, where the first reference image is obtained after a failure test is performed on a preset part using a stretching machine until a first weld spot in the preset part fails and becomes a failed weld spot.

[0141] The processing module 402 is configured to perform image processing on the first reference image to obtain a second reference image.

[0142] The establishing module 403 is configured to establish a recognition model according to a plurality of second reference images.

[0143] The acquisition module 401 is further configured to acquire a first image to be identified that includes the second welding point.

[0144] The processing module 402 is further configured to perform image processing on the first image to be recognized to obtain a second image to be recognized.

[0145] The recognition module 404 is configured to input the second image to be recognized into the recognition model to obtain an output recognition result, where the recognition result indicates whether the second solder joint is failed.

[0146] In some optional embodiments, performing a failure test on a predetermined part using a stretching machine until a first weld in the predetermined part fails and becomes a failed weld includes:

[0147] A tensile test machine is used to perform one of a shear test, a cross tensile test, and a peel test on the preset part until a first weld point in the preset part fails and becomes a failed weld point.

[0148] In some optional embodiments, performing image processing on the first reference image to obtain the second reference image includes:

[0149] The first reference image is sequentially subjected to erosion processing, dilation processing, median filtering, and debinarization processing to obtain a first intermediate image.

[0150] The minimum bounding rectangle of the second welding point in the first intermediate image is determined, and the image within the minimum bounding rectangle is used as the second intermediate image.

[0151] The size of the second intermediate image is stretched to a preset size to obtain a second reference image.

[0152] In some optional embodiments, performing image processing on the first reference image to obtain the second reference image further includes:

[0153] After the image in the minimum circumscribed rectangle is used as the second intermediate image, the size of the second intermediate image is stretched to a preset size to obtain a second original reference image.

[0154] The second original reference image is subjected to horizontal mirror transformation, vertical mirror transformation, and flip transformation to obtain a plurality of transformed expanded images.

[0155] The second original reference and the plurality of augmented images are used as second reference images.

[0156] In some optional embodiments, the establishing module 403 is configured to:

[0157] A plurality of second reference images are acquired.

[0158] The plurality of second reference images are divided into training images and validation images.

[0159] Build a neural network model.

[0160] The training image and verification image are used as the input of the neural network model, and the neural network model is iteratively trained to obtain a set of model parameters with the highest recognition accuracy.

[0161] Substitute the model parameters into the neural network model to obtain the recognition model.

[0162] Using the solder joint failure identification device provided by the present application, a solder joint peeling test is first performed on a preset part using a stretching machine until the solder joint in the preset part fails and becomes a failed solder joint, and a recognition model is established based on the image of the failed solder joint as a reference image. After the image to be identified containing the solder joint to be judged is input into the recognition model, an output recognition result is obtained, and the recognition result indicates whether the second solder joint has failed. Due to the use of the solder joint peeling experiment and model learning, the efficiency and accuracy of solder joint failure identification can be improved.

[0163] In this application, it should be understood that the terms "first", "second", etc. are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features.

[0164] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only.

[0165] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

[0166] The above description is only for the purpose of facilitating those skilled in the art to understand the technical solution of this application and is not intended to limit this application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this application shall be included in the scope of protection of this application.

Claims

1. A method for identifying solder joint failure, characterized in that: The method comprises: Performing a weld peeling test on a predetermined part using a tensile machine until a first weld in the predetermined part fails and becomes a failed weld; Acquire a first reference image including the failed solder joint; performing image processing on the first reference image to obtain a second reference image; establishing a recognition model based on the plurality of second reference images; Acquire a first image to be identified including a second welding point; performing image processing on the first image to be identified to obtain a second image to be identified; The second image to be recognized is input into the recognition model to obtain an output recognition result, where the recognition result indicates whether the second welding point is invalid.

2. The solder joint failure identification method according to claim 1, characterized in that: The method of performing a failure test on a preset part by using a stretching machine until a first weld point in the preset part fails and becomes a failed weld point includes: A tensile test machine is used to perform one of a shear test, a cross tensile test, and a peel test on the preset part until the first weld point in the preset part fails and becomes a failed weld point.

3. The solder joint failure identification method according to claim 1, characterized in that: The performing image processing on the first reference image to obtain a second reference image includes: performing corrosion processing, dilation processing, median filtering, and debinarization processing on the first reference image in sequence to obtain a first intermediate image; determining a minimum bounding rectangle of the second welding point in the first intermediate image, and using the image within the minimum bounding rectangle as the second intermediate image; The size of the second intermediate image is stretched to a preset size to obtain the second reference image.

4. The solder joint failure identification method according to claim 3, characterized in that: After using the image in the minimum bounding rectangle as the second intermediate image, the method further includes: stretching the second intermediate image to the preset size to obtain a second original reference image; Performing a horizontal mirror transformation, a vertical mirror transformation, and a flip transformation on the second original reference image to obtain a plurality of transformed expanded images; The second original reference and the plurality of augmented images are used as the second reference images.

5. The solder joint failure identification method according to claim 1, characterized in that: The establishing the recognition model according to the second reference image comprises: acquiring a plurality of second reference images; dividing the plurality of second reference images into training images and verification images; Build a neural network model; Using the training image and the verification image as inputs of the neural network model, iteratively training the neural network model to obtain a set of model parameters with the highest recognition accuracy; Substituting the model parameters into the neural network model, the recognition model is obtained.

6. A solder joint failure identification device, characterized in that: The device comprises: an acquisition module configured to acquire a first reference image including a failed solder joint, wherein the first reference image is obtained after a failure test is performed on a predetermined part using a stretching machine until a first solder joint in the predetermined part fails and becomes the failed solder joint; a processing module, configured to perform image processing on the first reference image to obtain a second reference image; an establishing module configured to establish a recognition model according to a plurality of second reference images; The acquisition module is further configured to acquire a first image to be identified including the second welding point; The processing module is further configured to perform image processing on the first image to be identified to obtain a second image to be identified; The recognition module is configured to input the second image to be recognized into the recognition model to obtain an output recognition result, where the recognition result indicates whether the second welding point is invalid.

7. The solder joint failure identification device according to claim 6, characterized in that: The method of performing a failure test on a preset part by using a stretching machine until a first weld point in the preset part fails and becomes a failed weld point includes: A tensile test machine is used to perform one of a shear test, a cross tensile test, and a peel test on the preset part until the first weld point in the preset part fails and becomes the failed weld point.

8. The solder joint failure identification device according to claim 6, characterized in that: The performing image processing on the first reference image to obtain a second reference image includes: performing corrosion processing, dilation processing, median filtering, and debinarization processing on the first reference image in sequence to obtain a first intermediate image; determining a minimum bounding rectangle of the second welding point in the first intermediate image, and using the image within the minimum bounding rectangle as the second intermediate image; The size of the second intermediate image is stretched to a preset size to obtain the second reference image.

9. The solder joint failure identification device according to claim 8, characterized in that: The performing image processing on the first reference image to obtain a second reference image further includes: After taking the image in the minimum circumscribed rectangle as the second intermediate image, stretching the second intermediate image to the preset size to obtain a second original reference image; Performing a horizontal mirror transformation, a vertical mirror transformation, and a flip transformation on the second original reference image to obtain a plurality of transformed expanded images; The second original reference and the plurality of augmented images are used as the second reference images.

10. The solder joint failure identification device according to claim 6, characterized in that: The establishment module is configured to: acquiring a plurality of second reference images; dividing the plurality of second reference images into training images and verification images; Build a neural network model; Using the training image and the verification image as inputs of the neural network model, iteratively training the neural network model to obtain a set of model parameters with the highest recognition accuracy; Substituting the model parameters into the neural network model, the recognition model is obtained.

Images