Pet trolley production fault image diagnosis method and system

Abstract

The invention provides a pet trolley production fault image diagnosis method and system, and belongs to the field of fault image diagnosis, and the method comprises the steps: S1, collecting a pet trolley production field image, adjusting the image contrast of the pet trolley production field image, and generating a first to-be-diagnosed production fault image; s2, inputting the first to-be-diagnosed production fault image into a multi-task feature extraction network for production fault feature processing to obtain a fault sensitive feature map; s3, performing feature enhancement on the fault sensitive feature map to obtain a fault feature map, and performing fault classification and positioning based on the fault feature map to obtain a fault diagnosis result; s4, combining the fault diagnosis result with the real-time process parameters of the production line, and generating an optimization suggestion for optimizing the production process; and S5, the optimization suggestion is fed back to the production line, and the process parameters of the production line are adjusted. According to the invention, the fault detection precision and efficiency are improved, and the omission ratio and the false detection ratio are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of fault image diagnosis, and particularly to a method and system for diagnosing production faults of pet trolleys. Background Art

[0002] During the production process of pet trolleys, a large number of sensors are distributed on the production line. These sensors continuously collect multi-dimensional data such as the operating status of equipment, environmental parameters, and process parameters. However, these data often have problems such as inconsistent sampling frequencies, uneven data quality, and severe signal interference, which increase the difficulty of data preprocessing and affect the subsequent fault diagnosis effect. Especially in the case of sudden faults during the production process, there is often a lack of complete historical data records, making it more difficult to extract fault features and perform pattern recognition.

[0003] To address these problems, currently, rule-based expert systems and statistical analysis methods are commonly used in industrial sites for fault diagnosis. These methods mainly rely on manually preset diagnostic rules and thresholds. Although they are simple to implement, they lack adaptability and intelligence. When faced with changes in production conditions or new production fault modes, these fixed diagnostic rules are often difficult to update and optimize in a timely manner, resulting in a decrease in diagnostic accuracy. At the same time, traditional methods have low utilization efficiency of historical data and cannot fully explore the hidden fault features and patterns in the data.

[0004] In order to improve the diagnostic effect, in recent years, some factories have begun to try to introduce intelligent diagnostic technologies based on machine learning. However, these technologies still face many challenges in practical applications. Existing intelligent diagnostic models are mostly designed for specific types of faults, lacking generality and transferability, and are difficult to meet the diagnostic requirements of different production lines and different working conditions. Secondly, existing fault diagnosis systems often operate independently and have a low degree of integration with other production management systems, resulting in the inability to track and manage fault information throughout the process, and the diagnostic results are difficult to effectively support decision-making optimization and process improvement.

[0005] Therefore, finding a method that can not only timely capture fault features to identify complex faults but also efficiently manage production efficiency and quality is a technical problem that needs to be urgently solved by those skilled in the art. Summary of the Invention

[0006] The present invention provides a method and system for diagnosing production faults of pet trolleys, which are used to solve the defect in the prior art that it is difficult to identify faults, resulting in the inability to efficiently manage production efficiency and quality, improve the accuracy and efficiency of fault detection, and reduce the missed detection rate and false detection rate.

[0007] The present invention provides a method for diagnosing production faults of pet trolleys, including the following steps: S1. Collect the images of the pet stroller production site and adjust the image contrast of the pet stroller production site images to generate the first production fault image to be diagnosed; S2. Input the first production fault image to be diagnosed into a multi-task feature extraction network for production fault feature processing to obtain a fault-sensitive feature map; S3. Perform feature enhancement on the fault-sensitive feature map to obtain a fault feature map, and perform fault classification and localization based on the fault feature map to obtain a fault diagnosis result; S4. Combine the fault diagnosis result with the real-time process parameters of the production line to generate an optimization suggestion for optimizing the production process; S5. Feed back the optimization suggestion to the production line and adjust the process parameters of the production line.

[0008] According to a production fault image diagnosis method of a pet stroller provided by the present invention, the production fault feature processing specifically includes: Extract features from the first production fault image to be diagnosed to obtain an initial fault feature map of the pet stroller; Divide the initial fault feature map of the pet stroller according to the historical fault risk frequency of the pet stroller to obtain the key area and non-key area of the pet stroller, enhance the response of the key area of the pet stroller, and suppress the interference of the non-key area of the pet stroller to obtain the global fault feature; Divide the initial fault feature map of the pet stroller according to the structure of the pet stroller to obtain multiple local structure feature areas; wherein the structure of the pet stroller includes a welded joint area, a support frame area, a connecting piece area, and a surface material area; Extract the local fault features of multiple local structure feature areas and calculate the local structure attention to obtain local structure attention features, and fuse the global fault features and local structure attention features to obtain a fault-sensitive feature map of the pet stroller.

[0009] According to a production fault image diagnosis method of a pet stroller provided by the present invention, the extraction of the local fault features of multiple local structure feature areas and the calculation of the local structure attention specifically include: Adopt differential convolution operations on multiple local structure feature areas of the pet stroller to extract local fault features respectively, perform local pooling operations on the local fault features to obtain local description vectors of the local structure feature areas; Use a fully connected layer to perform linear mapping and normalization processing on the local description vectors to obtain the normalized attention weights of the pet stroller; Multiply the normalized attention weights of the pet stroller element by element with the local fault features to obtain local structure attention features.

[0010] A method for diagnosing production fault images of a pet stroller provided by the present invention, in which during the training process of the multi-task feature extraction network, sparse regularization constraints are used to optimize the attention weight distribution of the multi-task feature extraction network, specifically including: Calculate the attention weight distribution of the multi-task feature extraction network to obtain the initial attention weight distribution; Compare and analyze the initial attention weight with a preset prior constant to obtain a weight deviation value; Adjust the initial attention weight distribution according to the weight deviation value to obtain the attention distribution of the key area; Sparsify the attention distribution of the key area to obtain the constraint of the final attention weight of the fault feature area.

[0011] A method for diagnosing production fault images of a pet stroller provided by the present invention, the feature enhancement of the fault-sensitive feature map specifically includes: Perform feature aggregation on the fault-sensitive feature map to obtain the global feature information of the pet stroller; Based on the global feature information of the pet stroller, perform local feature extraction, and use local convolution and dual attention mechanisms to obtain the local fault weight and local fault features of the pet stroller; Obtain the fault prior information of the pet stroller, and fuse the global feature information and local fault features of the pet stroller based on the fault prior information of the pet stroller to obtain a fault feature map.

[0012] A method for diagnosing production fault images of a pet stroller provided by the present invention, the fault classification and location based on the fault feature map specifically includes: Calculate the fault classification probability based on the fault feature map, generate candidate regions through a dynamic sliding window based on the fault classification probability, and use boundary regression to calculate the position information of the candidate regions to determine the initial fault position; Use non-maximum suppression to optimize the initial fault position to obtain a fault diagnosis result.

[0013] A method for diagnosing production fault images of a pet stroller provided by the present invention, step S4 specifically includes: Obtain the production historical data of the pet stroller and construct a fault association model between fault features and production process parameters; Compare the diagnosis result with the production historical data of the pet stroller to identify the abnormal process parameters of the current production fault; According to preset rules, calculate the adjustment range of process parameters for the current production fault, input the process adjustment range of the current production fault into the fault correlation model, determine the production link of the current production fault, and generate optimization suggestions.

[0014] The present invention also provides a production fault image diagnosis system for a pet stroller, which implements the production fault image diagnosis method as described above, including: An image acquisition and processing module, configured to acquire an image of the pet stroller production site and adjust the image contrast of the image of the pet stroller production site to generate a first production fault image to be diagnosed; A production fault feature processing module, configured to input the first production fault image to be diagnosed into a multi-task feature extraction network for production fault feature processing to obtain a fault-sensitive feature map; A classification and positioning module, configured to enhance the features of the fault-sensitive feature map to obtain a fault feature map, and perform fault classification and positioning based on the fault feature map to obtain a fault diagnosis result; An optimization suggestion module, configured to combine the fault diagnosis result with the real-time process parameters of the production line to generate optimization suggestions for optimizing the production process; A feedback module, configured to feedback the optimization suggestions to the production line to adjust the process parameters of the production line.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the production fault image diagnosis method as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the production fault image diagnosis method as described above.

[0017] A production fault image diagnosis method and system for a pet stroller provided by the present invention, by obtaining production line data, combining a multi-task feature extraction network for fault identification, and combining the diagnosis result with real-time process parameters to generate optimization suggestions, realizes a closed-loop control from fault detection to process adjustment, overcomes the defects of poor adaptability to new production faults and low data utilization rate of traditional methods, improves the fault detection accuracy and efficiency, reduces the missed detection rate and false detection rate, and effectively solves the quality monitoring problem in the production process of pet strollers. Description of the Drawings

[0018] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 is a flowchart of the production fault image diagnosis method provided by the present invention; Figure 2 is a schematic structural diagram of the electronic device provided by the present invention. Detailed implementation manners

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0021] As Figure 1 shown, the present invention provides a production fault image diagnosis method for a pet stroller, including the following steps: S1. Collect images of the pet stroller production site and adjust the image contrast of the pet stroller production site images to generate the first production fault image to be diagnosed.

[0022] It can be understood that in the present invention, an image acquisition device is installed on the production line of the pet stroller, and the image data of the pet stroller production line is obtained through the image acquisition device. The image acquisition device can be an industrial camera, which can capture fine defects (such as cracks and surface scratches), or a laser scanning camera, which can perform three-dimensional imaging of the product and has high precision in detecting the three-dimensional structure.

[0023] Due to the complex production line environment of the pet stroller, the images collected may be affected by various factors. For example, unstable lighting conditions may cause strong light, shadows, or reflections in the images, thereby obscuring detailed information; the industrial camera may capture noise information in the production background, interfering with the recognition of target features; at the same time, due to the different workstation settings and production batches on the pet stroller production line, the imaging range may be inconsistent. These problems will all cause the quality of the image data to decline and affect the accuracy of subsequent diagnosis. Therefore, it is necessary to preprocess the production image dataset to unify the image features and optimize the image quality.

[0024] In an embodiment of the present invention, adjusting the image contrast of the pet stroller production site image specifically includes: enhancing the brightness and contrast of the production image dataset by using histogram equalization. The preprocessing method combining histogram equalization with noise reduction and sharpening can effectively optimize the brightness, noise, and edge features simultaneously, and can adapt to the complex environmental change conditions in the actual production process, effectively detecting faults such as coating surface defects and welding cracks during the production of pet strollers.

[0025] S2. Input the first production fault image to be diagnosed into the multi-task feature extraction network for production fault feature processing to obtain a fault-sensitive feature map.

[0026] Specifically, the production fault feature processing specifically includes: Performing feature extraction on the first production fault image to be diagnosed to obtain an initial fault feature map of the pet stroller; Dividing the initial fault feature map of the pet stroller according to the historical fault risk frequency of the pet stroller to obtain the key area and non-key area of the pet stroller, enhancing the response to the key area of the pet stroller, and suppressing interference to the non-key area of the pet stroller to obtain a global fault feature; Dividing the initial fault feature map of the pet stroller according to the structure of the pet stroller to obtain multiple local structure feature areas; wherein the structure of the pet stroller includes a welding joint area, a support frame area, a connecting piece area, and a surface material area; Extracting the local fault features of multiple local structure feature areas and calculating the local structure attention to obtain local structure attention features, and fusing the global fault features and local structure attention features to obtain a fault-sensitive feature map of the pet stroller.

[0027] It can be understood that through the statistics of the historical fault data of the pet stroller, the fault-prone area is used as the key area, and the area with a lower fault incidence rate is classified as the non-key area. For example, if a certain local area of the pet stroller has a problem during production, it will directly threaten the overall safety (such as weld or frame fracture), then it is classified as the key area; if the fault of a certain part of the pet stroller only affects the secondary function or the maintenance cost is low (such as a slight scratch on the coating), then it is classified as the non-key area. In the fault-sensitive feature map, different colors, brightness, and textures represent the fault possibilities and severity of different structures of the pet stroller. That is, the "fault-sensitive feature map" obtained after fusion is like a "perspective diagnosis map" of the pet stroller production. In the fault-sensitive feature map, the welding joint area may present an abnormal red area, indicating a high risk; while the support frame area may present a uniform green, indicating a complete structure. This visual diagnosis method converts complex technical details into intuitive fault signals. By dividing the key regions of the initial fault feature map of the pet stroller, enhancing the response to the key regions, then dividing the local structural feature regions according to the structure of the pet stroller, and calculating the local structural attention features, the fault feature map of the pet stroller obtained by fusing the global fault features and the local structural attention features not only contains local structural information but also highlights the important risk priorities.

[0028] In an embodiment of the present invention, the multi-task feature extraction network includes an initial feature extraction layer and a global-local dual attention mechanism. The initial feature extraction layer is used to extract features from the first production fault image to be diagnosed, and the global-local dual attention mechanism is used to divide regions of the initial fault feature map of the pet stroller.

[0029] It can be understood that the global attention mechanism is used to enhance the key features in the initial fault feature map, and the local attention mechanism is used to capture the detailed features of the local structure of the initial fault feature map. Among them, response enhancement means automatically identifying and amplifying the most critical and vulnerable structural regions in the production of pet strollers. For example, welding joints, connections of support frames, key stress-bearing parts, etc. These regions are often decisive factors for product quality and safety. Interference suppression means automatically inputting interference information in non-critical regions through interference suppression techniques, such as shooting environment, lighting conditions, non-critical surface textures, etc.

[0030] The global-local dual attention mechanism adopted in the present invention, through region division and feature processing of the first production fault image to be diagnosed, enhances the response to key regions and suppresses interference in non-critical regions at the same time. It can effectively cope with problems such as complex environment, unstable lighting, and inconsistent imaging range in the pet stroller production line, improving the model's ability to extract fault features under different working conditions, especially significantly improving the recognition accuracy of subtle faults such as welding cracks and surface defects.

[0031] In an embodiment of the present invention, the initial feature extraction layer includes a convolution module, a pooling module, a non-linear activation function, and a feature fusion module. The convolution module includes 3 consecutive convolution layers, and the convolution kernel size of each convolution layer is 3×3. The pooling module performs downsampling on the output of part of the convolution module.

[0032] Specifically, the operation of the initial feature extraction layer is as follows: Input the first production fault image to be diagnosed into the convolution module for convolution to obtain the first feature; Input a part of the first feature into the pooling module for processing to obtain the second feature; Input another part of the first feature into the non-linear activation function to obtain the third feature; The second feature and the third feature are fused using a feature fusion module to obtain an initial fault feature map of the pet stroller; The ratio of the first feature input to the pooling module and the non-linear activation function can be set according to actual usage requirements, and the present invention does not make specific limitations thereto.

[0033] In an embodiment of the present invention, a normalization layer is connected after each convolutional layer in the convolutional module, which can effectively suppress overfitting.

[0034] In an embodiment of the present invention, the fusion module can be channel fusion splicing, that is, the second feature and the third feature are spliced on the channel, or 1×1 convolution fusion can be used, that is, the second feature is first subjected to channel change using 1×1 convolution, and then added to the third feature element by element.

[0035] Specifically, the initial fault feature map of the pet stroller is regionally divided according to the historical fault risk frequency of the pet stroller, specifically including: A global pooling operation is performed on the initial fault feature map of the pet stroller to obtain a global feature representation, and a weight distribution calculation is performed on the global feature representation to obtain the weight distribution of the initial fault feature map; Based on the weight distribution of the initial fault feature map, the key area and the non-key area of the pet stroller are determined.

[0036] In an embodiment of the present invention, the weight distribution calculation specifically includes: A global pooling operation is performed on the global feature representation to obtain a global feature vector by taking the mean in the spatial dimension (i.e., all pixel points); A fully connected layer is used to map the global feature vector, and then the softmax function is applied for normalization to obtain the weight distribution of the initial fault feature map.

[0037] It can be understood that by performing a weight distribution calculation on the global feature representation, the importance of different channels can be quantified, and the softmax function ensures that all weights are between 0 and 1 and the sum is 1.

[0038] In an embodiment of the present invention, the weight distribution of the initial fault feature map is set according to a weight threshold, where the weight threshold can be set according to actual usage requirements. The positions with weights higher than the weight threshold are regarded as key areas, and the positions with weights lower than the weight threshold are regarded as non-key areas. This distinction can selectively enhance the response of key areas (such as areas with significant fault features) while reducing the interference of noise.

[0039] In an embodiment of the present invention, the extraction of local fault features of multiple local structural feature regions and the calculation of local structural attention specifically include: Differentiated convolution operations are performed on multiple local structural feature regions of the pet stroller to extract local fault features respectively. Local pooling operations are performed on the local fault features to obtain local description vectors of the local structural feature regions; A fully connected layer is used to perform linear mapping and normalization processing on the local description vectors to obtain the normalized attention weights of the pet stroller; The normalized attention weights of the pet stroller are multiplied element by element with the local fault features to obtain local structural attention features.

[0040] In an embodiment of the present invention, the differentiated convolution operations include: using a fine-grained convolution kernel for the welded joint area to extract the continuity and penetration depth features of the weld; using a large convolution kernel for the support frame area to capture the overall structural deformation and stress distribution features; selecting an edge-sensitive convolution kernel for the connecting piece area to extract the bolt loosening and buckle integrity features; using a texture-enhanced convolution kernel for the surface material area to extract the scratch, peeling and wear features.

[0041] In an embodiment of the present invention, the normalization processing is normalization processing.

[0042] It can be understood that by calculating the local attention in the initial fault feature map of the pet stroller, the detection ability for subtle local anomalies (such as small cracks or local scratches) can be effectively improved.

[0043] In an embodiment of the present invention, when using a fully connected layer to perform linear mapping on the local description vectors, a bias term can also be added so that the features of each local area can be mapped into a predefined dimensional space.

[0044] In an embodiment of the present invention, the multi-task feature extraction network uses sparse regularization constraints to optimize the attention weight distribution of the multi-task feature extraction network during the training process, specifically including: Calculating the attention weight distribution of the multi-task feature extraction network to obtain the initial attention weight distribution; Comparing and analyzing the initial attention weights with a preset prior constant to obtain the weight deviation value; Adjusting the initial attention weight distribution according to the weight deviation value to obtain the attention distribution of the key area; Performing sparsification processing on the attention distribution of the key area to obtain the constraint of the final attention weight of the fault feature area.

[0045] Among them, the preset prior constant can be determined based on empirical values or the production fault feature distribution pre-statistically obtained from the training data. The sparse regularization can be L1 regularization.

[0046] It is understandable that in the multi-task feature extraction network, the role of the attention mechanism is to enable the model to focus on specific key regions while ignoring irrelevant or interfering regions, so as to improve the extraction and classification effects of fault features. However, in actual applications, the attention distribution may be too "dispersed", that is, unnecessary high weights are assigned to all regions instead of concentrating on important key regions. Therefore, it is necessary to constrain the attention weights through sparse regularization to make the attention weights concentrate on a few key regions, while reducing the influence of useless regions and achieving more efficient feature extraction.

[0047] S3. Perform feature enhancement on the fault-sensitive feature map to obtain a fault feature map, and perform fault classification and location based on the fault feature map to obtain a fault diagnosis result, where the fault diagnosis result includes the fault category, fault location, and fault severity.

[0048] In an embodiment of the present invention, performing feature enhancement on the fault-sensitive feature map specifically includes: Perform feature aggregation on the fault-sensitive feature map to obtain the global feature information of the pet stroller; Perform local feature extraction based on the global feature information of the pet stroller, and use local convolution and a dual attention mechanism to obtain the local fault weight and local fault features of the pet stroller; Obtain the fault prior information of the pet stroller, and fuse the global feature information and local fault features of the pet stroller based on the fault prior information of the pet stroller to obtain a fault feature map. The calculation formula is: Among them, represents the fault feature map at position (i, j), represents the non-linear activation function, represents the local fault weight, represents the feature vector of the fault feature map at position (i, j), represents the adjustment parameter, represents the fault prior information of the pet stroller, represents the global feature information, represents the row coordinate in the fault feature map, represents the column coordinate in the fault feature map.

[0049] It is understandable that due to the diverse fault types and unfixed positions in the production of pet strollers, by fusing and utilizing local fault features (such as welding cracks) and global feature information (such as overall structural offset), the comprehensiveness of fault detection is improved. The fusion formula of this application enhances the feature extraction ability for key regions through local fault weights and prior information, while suppressing the interference of non-key regions. Integrating the prior information of production line failures into the feature extraction process, β × prior(i,j) × G realizes the combination of global feature information and prior knowledge, modulates the influence degree of global feature information through prior probability, and while ensuring global information, pays attention to high-risk areas according to historical experience. It can adaptively enhance features at different positions and highlight the feature expressions in key failure areas.

[0050] Among them, the prior information of pet stroller failures includes historical failure data, process parameter impacts, etc.

[0051] In an embodiment of the present invention, the operation of feature aggregation specifically includes: Performing global average pooling operation on the failure-sensitive feature map to obtain a fourth feature; Taking the mean value of the fourth feature in the spatial dimension (i.e., all pixel points), compressing the two-dimensional feature map into a one-dimensional feature vector to obtain global feature information.

[0052] It can be understood that feature aggregation is a process of integrating global information in the spatial dimension of the failure-sensitive feature map, and its core operation is to use global average pooling (Global Average Pooling, GAP) to capture the global statistical information of the entire feature map. The prior information of failures is obtained by statistically analyzing historical production failure data, which reflects the prior probability of failures occurring in key parts of the pet stroller during the manufacturing process (such as frame joints, welding areas, fastening parts), and is used to guide the key strengthening of key areas when integrating global and local features.

[0053] The feature enhancement of the present invention realizes the full utilization of historical failure data, can focus on detecting key positions such as the frame joints, welding areas, and fastening parts of the pet stroller, improves the sensitivity to failure features, and reduces the misidentification of irrelevant features.

[0054] In an embodiment of the present invention, the failure classification and localization based on the failure feature map specifically include: Calculating the failure classification probability based on the failure feature map, generating candidate regions through a dynamic sliding window based on the failure classification probability, and using boundary regression to calculate the position information of the candidate regions to determine the initial failure position; the calculation formula for the position information of the candidate regions is: ; ; Among them, represents the confidence score of the candidate region, represents the sigmoid activation function, Denote the convolutional weight matrix for candidate region generation, Denote the convolutional operation, Denote the bias term of the convolutional operation, Denote the boundary coordinates of the candidate region, Denote the weight matrix for boundary regression, Denote the bias term of the boundary regression module; Optimize the initial fault location using non-maximum suppression to obtain the fault diagnosis result.

[0055] It can be understood that the present invention uses a dynamic sliding window to achieve precise positioning of the fault region. It utilizes the fault classification probability as the search prior, calculates the confidence of the candidate region using the sigmoid function, and introduces boundary regression to calculate the precise location. By performing differential searches on different structures of the pet stroller, such as using windows of different sizes to capture features in the welding joint area and the support frame area, it can capture fault features of different scales.

[0056] Furthermore, calculate the fault classification probability based on the fault feature map, specifically including: Perform global average pooling on the fault feature map to generate an image-level fault description vector, and input it into the fully connected layer. Use the softmax function to output the fault classification probability distribution. The calculation formula of the fault classification probability distribution is: ; Wherein, Denote the fault classification probability distribution, Denote the softmax activation function, Denote the weight matrix of the fully connected layer, Denote the global average pooling operation, Denote the fault feature map, Denote the bias vector of the fully connected layer.

[0057] It can be understood that boundary regression is used to precisely locate the candidate region. According to the fault feature map, regression prediction is performed on the boundary of the candidate region to output accurate region coordinates. Since it is necessary to convert the extracted fault features into clear fault category identifiers (such as loose screws, welding cracks, etc.), the present invention captures the statistical characteristics of the entire fault feature map through global average pooling to judge the overall fault type of the pet stroller. The fully connected layer provides the mapping ability from features to categories, can learn the distinguishing features of different faults, and uses the softmax function to ensure that the output result is a valid probability distribution. Among them, Compress the two-dimensional fault feature map into a one-dimensional feature vector, perform a linear transformation on the feature vector through the fully connected layer, and use the softmax function to convert the output into a probability distribution, making the sum of the probabilities of each category equal to 1.

[0058] After determining the type of pet stroller failure, it is also necessary to locate the failure position. Therefore, the local features of the failure feature map are extracted through convolution operations to effectively capture position-related features. The sigmoid function is used to limit the confidence interval between 0 and 1, facilitating the detection threshold. By predicting the boundary coordinates of the candidate regions, indicating the precise position and scope of the failure, and obtaining the precise failure type and position.

[0059] The present invention outputs the failure classification probability distribution through global average pooling and the softmax function, and calculates the position information of the candidate regions using the dynamic sliding window and the boundary regression module. After combining the two, non-maximum suppression is performed to achieve precise identification of the failure category, position, and severity, providing a more accurate basis for subsequent process adjustment.

[0060] In an embodiment of the present invention, the boundary regression includes convolution operations and fully connected operations. The boundary regression utilizes the features within the candidate regions to predict the four coordinates of the candidate regions (or forms such as the center coordinate plus width and height) through convolution and linear regression.

[0061] S4. Combine the failure diagnosis result with the real-time process parameters of the production line to generate optimization suggestions for optimizing the production process; wherein the optimization suggestions include the production line temperature, pressure, and speed.

[0062] Specifically, step S4 specifically includes: Obtain the production historical data of the pet stroller and construct a failure correlation model between the failure features and the production process parameters; Compare the diagnosis result with the production historical data of the pet stroller to identify the abnormal process parameters of the current production failure; According to the preset rules, calculate the process parameter adjustment range of the current production failure, input the process adjustment range of the current production failure into the failure correlation model, determine the production link of the current production failure, and generate optimization suggestions.

[0063] Among them, the preset rules can be set according to the actual usage requirements of the pet stroller production line, and the present invention does not make specific limitations on this.

[0064] It can be understood that there is an inherent correlation between the failure types and the process parameters in the production of pet strollers. The present invention fully considers the particularity of pet stroller production, such as the relationship between welding temperature and welding quality, the correlation between assembly pressure and fastening effect, etc. By combining the failure diagnosis result with the real-time process parameters, targeted optimization suggestions are automatically generated to complete the closed-loop control of "detection - analysis - adjustment", reducing the need for manual intervention.

[0065] Specifically, a specific embodiment is used for illustration: Obtain the production historical data of the pet stroller production line in the past year, including complete production parameter records, quality inspection reports, repair records, and equipment operation logs, and construct a multi-dimensional fault correlation model; Compare the diagnosis result with the production historical data of the pet stroller. When there is a micro-crack fault in the welding area of the support frame, the abnormal process parameters include: the welding current fluctuation range exceeds the standard value by ±5%, the contact angle between the welding head and the workpiece has a deviation of 0.3°, and the welding cooling rate is uneven; Calculate the adjustment range of the process parameters according to the preset fault diagnosis rules: adjust the welding current from the original 120±5A to 115±3A, calibrate the welding head angle to the standard contact angle, with the error controlled within ±0.1°, and optimize the cooling air flow distribution; Input the adjustment range of the process parameters into the fault correlation model, determine the specific production links, and generate optimization suggestions.

[0066] S5. Feed back the optimization suggestions to the production line and adjust the process parameters of the production line.

[0067] In an embodiment of the present invention, the process optimization suggestions are encapsulated through standard interfaces and data formats (such as JSON, XML), and adjustment instructions are sent to the production line control system in real time using industrial control protocols. After receiving the instructions, the production line control system automatically adjusts the corresponding process parameters according to preset thresholds, and feedbacks the adjustment effect through the monitoring system to form a closed-loop control, realizing intelligent monitoring and automatic adjustment of the production process, significantly improving the automation level and product quality stability of the production line, and reducing the frequency of manual intervention and production costs.

[0068] The present invention obtains production line data through image acquisition devices, combines a multi-task feature extraction network for fault identification, and combines the diagnosis result with real-time process parameters to generate optimization suggestions, realizing a closed-loop control from fault detection to process adjustment, overcoming the defects of poor adaptability to new production faults and low data utilization rate in traditional methods, improving the fault detection accuracy and efficiency, reducing the missed detection rate and false detection rate, and effectively solving the quality monitoring problem in the production process of pet strollers.

[0069] The present invention also provides a production fault image diagnosis system for pet strollers, which realizes the production fault image diagnosis method as described above, including: An image acquisition and processing module, which is used to acquire images of the pet stroller production site and adjust the image contrast of the pet stroller production site images to generate the first production fault image to be diagnosed; A production fault feature processing module, which is used to input the first production fault image to be diagnosed into a multi-task feature extraction network for production fault feature processing to obtain a fault-sensitive feature map; A classification and localization module, which is used to enhance the features of the fault-sensitive feature map to obtain a fault feature map, and perform fault classification and localization based on the fault feature map to obtain a fault diagnosis result; An optimization suggestion module, which is used to combine the fault diagnosis result with the real-time process parameters of the production line to generate an optimization suggestion for optimizing the production process; A feedback module, which is used to feedback the optimization suggestion to the production line to adjust the process parameters of the production line.

[0070] The production fault image diagnosis device provided by the present invention will be described below. The production fault image diagnosis device described below can be correspondingly referred to the production fault image diagnosis method described above.

[0071] Figure 2 An example of the physical structure diagram of an electronic device is shown as Figure 2 As shown, the electronic device may include: a processor 210, a communication interface 220, a memory 230, and a communication bus 240. Among them, the processor 210, the communication interface 220, and the memory 230 complete mutual communication through the communication bus 240. The processor 210 can call the logical instructions in the memory 230 to execute the production fault image diagnosis method, and the method includes: collecting an image of the pet stroller production site and adjusting the image contrast of the pet stroller production site image to generate a first production fault image to be diagnosed; inputting the first production fault image to be diagnosed into a multi-task feature extraction network for production fault feature processing to obtain a fault-sensitive feature map; enhancing the features of the fault-sensitive feature map to obtain a fault feature map, and performing fault classification and localization based on the fault feature map to obtain a fault diagnosis result; combining the fault diagnosis result with the real-time process parameters of the production line to generate an optimization suggestion for optimizing the production process; feedbacking the optimization suggestion to the production line to adjust the process parameters of the production line.

[0072] In addition, when the logical instructions in the above-mentioned memory 230 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0073] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the production fault image diagnosis method provided by the above-mentioned various methods. The method includes: collecting an image of the pet stroller production site and adjusting the image contrast of the image of the pet stroller production site to generate a first production fault image to be diagnosed; inputting the first production fault image to be diagnosed into a multi-task feature extraction network for production fault feature processing to obtain a fault-sensitive feature map; performing feature enhancement on the fault-sensitive feature map to obtain a fault feature map, and performing fault classification and localization based on the fault feature map to obtain a fault diagnosis result; combining the fault diagnosis result with the real-time process parameters of the production line to generate an optimization suggestion for optimizing the production process; and feeding back the optimization suggestion to the production line to adjust the process parameters of the production line.

[0074] In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the production fault image diagnosis method provided by the above-mentioned various methods. The method includes: collecting an image of the pet stroller production site and adjusting the image contrast of the image of the pet stroller production site to generate a first production fault image to be diagnosed; inputting the first production fault image to be diagnosed into a multi-task feature extraction network for production fault feature processing to obtain a fault-sensitive feature map; performing feature enhancement on the fault-sensitive feature map to obtain a fault feature map, and performing fault classification and localization based on the fault feature map to obtain a fault diagnosis result; combining the fault diagnosis result with the real-time process parameters of the production line to generate an optimization suggestion for optimizing the production process; and feeding back the optimization suggestion to the production line to adjust the process parameters of the production line.

[0075] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0076] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. 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 invention.

Claims

1. A method for diagnosing production fault images of a pet stroller, characterized in that, It includes the following steps: S1. Collect images of the pet stroller production site and adjust the image contrast of the pet stroller production site images to generate the first production fault image to be diagnosed; S2. Input the first production fault image to be diagnosed into a multi-task feature extraction network for production fault feature processing to obtain a fault-sensitive feature map; S3. Enhance the features of the fault-sensitive feature map to obtain a fault feature map, and perform fault classification and localization based on the fault feature map to obtain a fault diagnosis result; S4. Combine the fault diagnosis result with the real-time process parameters of the production line to generate an optimization suggestion for optimizing the production process; S5. Feed back the optimization suggestion to the production line to adjust the process parameters of the production line.

2. The production fault image diagnosis method of a pet trolley according to claim 1, characterized in that, The production fault feature processing specifically includes: Extract features from the first production fault image to be diagnosed to obtain an initial fault feature map of the pet stroller; Divide the initial fault feature map of the pet stroller into key areas and non-key areas according to the historical fault risk frequency of the pet stroller, enhance the response of the key areas of the pet stroller, and suppress the interference of the non-key areas of the pet stroller to obtain global fault features; Divide the initial fault feature map of the pet stroller according to the structure of the pet stroller to obtain multiple local structure feature areas; wherein the structure of the pet stroller includes a welded joint area, a support frame area, a connecting piece area, and a surface material area; Extract local fault features of multiple local structure feature areas and calculate local structure attention to obtain local structure attention features, and fuse the global fault features and local structure attention features to obtain a fault-sensitive feature map of the pet stroller.

3. A method for diagnosing production fault images of a pet stroller according to claim 2, characterized in that, The extraction of local fault features of multiple local structure feature areas and the calculation of local structure attention specifically include: Adopt differential convolution operations on multiple local structure feature areas of the pet stroller to extract local fault features respectively, and perform local pooling operations on the local fault features to obtain local description vectors of the local structure feature areas; Use a fully connected layer to perform linear mapping and normalization processing on the local description vectors to obtain the normalized attention weights of the pet stroller; Multiply the normalized attention weights of the pet stroller element by element with the local fault features to obtain local structure attention features.

4. A method for diagnosing production fault images of a pet trolley according to claim 3, characterized in that, During the training process of the multi-task feature extraction network, the attention weight distribution of the multi-task feature extraction network is optimized using sparse regularization constraints, specifically including: Calculate the attention weight distribution of the multi-task feature extraction network to obtain an initial attention weight distribution; Compare and analyze the initial attention weights with a preset prior constant to obtain a weight deviation value; Adjust the initial attention weight distribution according to the weight deviation value to obtain the attention distribution of the key areas; Perform sparsification processing on the attention distribution of the key areas to obtain the constraint of the final attention weights of the fault feature areas.

5. A method for diagnosing production fault images of a pet stroller according to claim 1, characterized in that, The feature enhancement of the fault-sensitive feature map specifically includes: Feature aggregation is performed on the fault-sensitive feature map to obtain the global feature information of the pet stroller; Based on the global feature information of the pet stroller, local feature extraction is performed, and local convolution and a dual attention mechanism are used to obtain the local fault weight and local fault features of the pet stroller; The fault prior information of the pet stroller is obtained, and based on the fault prior information of the pet stroller, the global feature information and local fault features of the pet stroller are fused to obtain a fault feature map.

6. The production fault image diagnosis method of a pet stroller according to claim 5, characterized in that The fault classification and localization based on the fault feature map specifically include: Calculating the fault classification probability based on the fault feature map, generating candidate regions through a dynamic sliding window based on the fault classification probability, and using boundary regression to calculate the position information of the candidate regions to determine the initial fault position; Using non-maximum suppression to optimize the initial fault position to obtain the fault diagnosis result.

7. A method for diagnosing production fault images of a pet stroller according to claim 1, characterized in that, Step S4 specifically includes: Obtaining the production historical data of the pet stroller and constructing a fault association model between fault features and production process parameters; Comparing the diagnosis result with the production historical data of the pet stroller to identify the abnormal process parameters of the current production fault; According to the preset rules, calculating the process parameter adjustment range of the current production fault, and inputting the process adjustment range of the current production fault into the fault association model to determine the production link of the current production fault and generate an optimization suggestion.

8. A production fault image diagnosis system for a pet stroller, characterized in that, Implementing the production fault image diagnosis method according to any one of claims 1-7, including: An image acquisition and processing module for acquiring an image of the pet stroller production site and adjusting the image contrast of the pet stroller production site image to generate a first production fault image to be diagnosed; A production fault feature processing module for inputting the first production fault image to be diagnosed into a multi-task feature extraction network for production fault feature processing to obtain a fault-sensitive feature map; A classification and localization module for performing feature enhancement on the fault-sensitive feature map to obtain a fault feature map, and performing fault classification and localization based on the fault feature map to obtain a fault diagnosis result; An optimization suggestion module for combining the fault diagnosis result with the real-time process parameters of the production line to generate an optimization suggestion for optimizing the production process; A feedback module for feeding back the optimization suggestion to the production line to adjust the process parameters of the production line.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the production fault image diagnosis method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the production fault image diagnosis method according to any one of claims 1 to 7.

Images