Towel rack solder quantitative automatic supply method and system based on AI visual identification

Abstract

The invention discloses a towel rack welding flux quantitative automatic supply method and system based on AI visual identification, and relates to the technical field of intelligent welding, and the method comprises the steps: based on image data of a to-be-welded point of a towel rack, positioning the position of a welding point, extracting surface features, and generating a welding point feature vector set; according to the welding point feature vector set, the basic theoretical demand quantity of the welding flux and the welding flux dosage correction value are determined, and a quantitative supply instruction set is established by calculating the predicted welding flux dosage; based on the quantitative supply instruction set, the opening time and the rotating speed of a screw valve are controlled; and a welding result image is collected through the industrial camera, and the material characteristic correction coefficient is adjusted according to the welding result image so as to optimize the predicted amount of the welding flux in the next welding process. The welding flux consumption can be automatically adjusted according to the actual requirements of different welding positions, and the problem that according to traditional welding equipment, welding flux in some areas is too much, and other areas are insufficient is effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent welding technology, and in particular to a method and system for automatically supplying towel rack solder in a quantitative manner based on AI visual recognition. Background Art

[0002] With the rapid growth in demand for industrial automation, the welding process for metal products such as towel racks in the home furnishing manufacturing industry is gradually developing towards intelligent and precise processes. As an important component of bathroom products, the welding process in the production of towel racks directly affects the quality and lifespan of the product. Traditional welding processes mainly rely on manual operation, which not only has low production efficiency but also uneven product quality. With the development of artificial intelligence technology, the application of AI visual recognition systems in the manufacturing industry has become an inevitable trend to improve production efficiency and product quality. AI visual technology can accurately identify the location of welding points through methods such as image recognition and deep learning, realizing the automation and standardization of the welding process, providing a new technical path for the refined production of metal products such as towel racks.

[0003] Currently, towel rack welding processes primarily rely on manual spot soldering or semi-automated spot soldering equipment. In these processes, operators must determine solder dosage based on experience and manually adjust spot soldering device parameters, resulting in unstable welding quality. While some highly automated welding equipment, such as spot welders using basic vision systems, is available on the market, these devices generally lack intelligent recognition and dynamic adjustment capabilities. Existing equipment typically uses fixed parameters for solder supply, making it impossible to flexibly adjust to the varying specifications of towel rack structures and unable to monitor solder supply status in real time. This results in inaccurate solder dosage, significant waste, and the risk of poor welding. These issues severely restrict the quality improvement and production efficiency of towel rack products.

[0004] However, existing welding equipment is not intelligent enough, especially when it comes to supplying solder to towel racks. First, the lack of a precise positioning system causes solder point offsets, impacting welding quality. Second, the solder supply cannot be dynamically adjusted to meet the needs of different welding positions, resulting in excessive solder in some areas and insufficient solder in others. Third, existing welding equipment lacks real-time monitoring and feedback mechanisms, making it impossible to promptly adjust to abnormalities that arise during the welding process. Finally, the welding equipment's limited learning capabilities make it difficult to optimize welding parameters based on historical data, resulting in an inability to self-improve as production progresses. Frequent adjustments to equipment parameters are required, reducing production efficiency and increasing production costs.

[0005] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0006] In response to the problems in the related art, the present invention proposes a method and system for automatic quantitative supply of towel rack solder based on AI visual recognition, which has the advantages of accurately identifying solder joint characteristics and calculating the optimal solder dosage, thereby solving the problems in the existing technology that welding equipment is not intelligent enough, the solder supply is inaccurate, it cannot adapt to different solder joint conditions and lacks self-optimization capabilities.

[0007] To this end, the specific technical solutions adopted in the present invention are as follows:

[0008] According to one aspect of the present invention, a method for automatically supplying solder to a towel rack based on AI visual recognition is provided. The method for automatically supplying solder to a towel rack based on AI visual recognition comprises:

[0009] S1. Based on the image data of the towel rack's welding points, locate the welding points and extract surface features to generate a welding point feature vector set;

[0010] S2. Determine the basic theoretical demand for solder and the solder usage correction value based on the solder point feature vector set, and establish a quantitative supply instruction set by calculating the predicted solder usage;

[0011] S3. Based on the quantitative supply instruction set, the opening time and speed of the screw valve are controlled; the welding result image is collected through an industrial camera, and the material property correction coefficient is adjusted according to the welding result image to optimize the predicted amount of solder in the next welding process.

[0012] Furthermore, the welding point feature vector set includes geometric size features, surface roughness features and wettability features of the welding point;

[0013] Based on the image data of the towel rack's welded points, the weld points are located and surface features are extracted to generate a weld point feature vector set including:

[0014] S11. Use an industrial camera to collect multi-angle image data of the towel rack's welded points, locate the welded points using a target detection algorithm, and divide the welded point areas based on a threshold segmentation technique.

[0015] S12. According to the solder joint area, a deep residual network is used to extract the surface features of the solder joint and generate a solder joint feature vector set.

[0016] Furthermore, the basic theoretical demand for solder and the solder usage correction value are determined based on the solder point feature vector set, and the predicted solder usage is calculated to establish a quantitative supply instruction set including:

[0017] S21. Calculate the theoretical solder volume based on the solder joint feature vector set, correct it using the material property correction coefficient, and analyze the basic theoretical demand for solder in combination with the wettability characteristics;

[0018] S22, calculating the solder flow resistance coefficient based on the solder joint feature vector set, searching for similar solder joint cases, using physical information to guide the model, and determining the solder amount correction value;

[0019] S23. Based on the basic theoretical demand for solder and the solder usage correction value, the predicted solder usage is calculated through a weighted fusion algorithm, and a quantitative supply instruction set is generated.

[0020] Furthermore, the theoretical solder volume is calculated based on the solder joint feature vector set, and the material property correction coefficient is used for correction. Combined with the wettability characteristics, the basic theoretical demand for solder is analyzed, including:

[0021] S211. Calculating a theoretical solder volume required to fill a gap in a solder joint based on geometrical dimension features in a solder joint feature vector set;

[0022] S212. Correcting the theoretical solder volume using a material property correction coefficient to obtain a corrected solder volume; wherein the material property correction coefficient has an initial value of 1 and is dynamically adjusted based on historical supply deviation analysis results;

[0023] S213. Calculate the basic theoretical demand for solder based on the corrected solder volume and the wettability characteristics of the solder joint.

[0024] Furthermore, the solder flow resistance coefficient is calculated based on the solder joint feature vector set, and similar solder joint cases are retrieved. The physical information is used to guide the model to determine the solder amount correction value, including:

[0025] S221. Calculating a solder flow resistance coefficient based on surface roughness characteristics of a soldering point feature vector set;

[0026] S222. Based on the solder flow resistance coefficient, searching for solder joint cases whose similarity is not less than a preset threshold from a historical database to obtain a set of similar solder joint cases;

[0027] S223. Based on a case set of similar solder joints, a solder amount correction value for the current solder joint is generated by guiding the model through physical information.

[0028] Furthermore, according to the surface roughness characteristics of the solder joint feature vector set, the solder flow resistance coefficient is calculated including:

[0029] S2211. Extract surface roughness features based on the weld feature vector set and analyze the average roughness value, peak-to-valley distribution, and directionality index;

[0030] S2212. Establish a microscopic fluid dynamics model based on the average roughness value, peak-to-valley distribution, and directionality index;

[0031] S2213. Calculate the solder flow resistance coefficient of the current soldering point through a microscopic fluid dynamics model.

[0032] Furthermore, based on the solder flow resistance coefficient, solder joint cases with a similarity not less than a preset threshold are retrieved from the historical database, and a set of similar solder joint cases is obtained, including:

[0033] S2221. Combining the flow resistance coefficient of the current welding point with the surface roughness feature to form a query vector, and performing normalization processing on the query vector;

[0034] S2222. Calculate the similarity score between the query vector and the feature vectors of each welding point case in the historical database using a cosine similarity algorithm;

[0035] S2223. Filter historical welding point cases whose similarity scores are not lower than a preset threshold, sort them from high to low according to similarity, and take the top N cases as the similar welding point case set.

[0036] Furthermore, based on a set of similar solder joint cases, the physical information-guided model generates the solder amount correction value for the current solder joint, including:

[0037] S2231. Extract historical solder usage data and corresponding welding quality evaluation results from a collection of similar welding point cases;

[0038] S2232. Based on the deep neural network model, a physical constraint layer is added that uses the flow resistance coefficient as the main parameter input to obtain a physical information guidance model;

[0039] S2233. Use a set of similar welding point cases to train a physical information guided model, and use weighted averaging to perform feature fusion on similar cases;

[0040] S2234. Input the characteristic vector and flow resistance coefficient of the current soldering point into the trained physical information guidance model, and output the solder amount correction value of the current soldering point.

[0041] Furthermore, based on the quantitative supply instruction set, the opening time and speed of the screw valve are controlled; the welding result image is collected by an industrial camera, and the material property correction coefficient is adjusted based on the welding result image to optimize the predicted solder amount in the next welding process, including:

[0042] S31. According to the quantitative supply instruction set, the predicted amount of solder is converted into motor control parameters, and the screw valve motor is driven to rotate to achieve quantitative supply of solder;

[0043] S32. After the welding is completed and cooled, an industrial camera is used to capture an image of the welding result, measure the actual filling area and edge contour of the welding point, and compare the difference between the actual size of the welding point and the ideal size;

[0044] S33. Based on the comparison results, when a material shortage occurs at the welding point, it is recorded as a negative deviation; when a material overflow occurs at the welding point, it is recorded as a positive deviation;

[0045] S34. Determine an adjustment coefficient based on the deviation value. For a negative deviation, increase the correction coefficient by a preset ratio. For a positive deviation, decrease the correction coefficient by a preset ratio to optimize the predicted amount of solder in the next welding process.

[0046] According to another aspect of the present invention, there is also provided a towel rack solder quantitative automatic supply system based on AI visual recognition, the towel rack solder quantitative automatic supply system based on AI visual recognition comprising:

[0047] The welding point identification and positioning module is used to locate the welding point position and extract surface features based on the image data of the towel rack welding point to generate a welding point feature vector set;

[0048] The solder consumption prediction module is used to determine the basic theoretical demand for solder and the solder consumption correction value based on the solder point feature vector set, and to establish a quantitative supply instruction set by calculating the predicted solder consumption;

[0049] The feeding control optimization module is used to control the opening time and speed of the screw valve based on the quantitative feeding instruction set; the welding result image is collected through an industrial camera, and the material property correction coefficient is adjusted according to the welding result image to optimize the predicted solder amount in the next welding process.

[0050] The beneficial effects of the present invention are:

[0051] (1) The present invention realizes the precise positioning of welding points, feature extraction and precise control of solder quantity through the automatic quantitative supply method of towel rack solder based on AI visual recognition, effectively solving the problems of traditional welding equipment being not smart enough, inaccurate solder supply and unstable welding quality, and improving the automation level and product quality of towel rack production.

[0052] (2) The present invention is based on the extracted solder joint feature vector set, calculates the theoretical solder volume and corrects it in combination with the material property correction coefficient, and uses the flow resistance coefficient and physical information to guide the model to generate a solder quantity correction value, thereby achieving precise control of the solder supply amount and being able to automatically adjust the solder quantity according to the actual needs of different welding positions, effectively avoiding the problem of excessive solder in some areas and insufficient solder in other areas in traditional welding equipment.

[0053] (3) The present invention collects welding result images through industrial cameras, compares the difference between the actual size of the welding point and the ideal size, and dynamically adjusts the material property correction coefficient according to the deviation value, thereby constructing a complete closed-loop feedback mechanism. It can continuously learn and optimize from historical data, adapt to the welding requirements of towel racks of different materials and structures, realize continuous optimization and self-improvement of process parameters, and improve production efficiency and product consistency. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0055] Figure 1 1 is a flow chart of a method for automatically supplying solder to a towel rack based on AI visual recognition according to an embodiment of the present invention;

[0056] Figure 2 2. This is a schematic diagram of an application scenario of a method for automatically supplying solder to a towel rack based on AI visual recognition according to an embodiment of the present invention;

[0057] Figure 3 This is a partial schematic diagram of an application scenario of a method for automatically supplying quantitative solder to a towel rack based on AI visual recognition according to an embodiment of the present invention;

[0058] Figure 4 This is a principle block diagram of a towel rack solder quantitative automatic supply system based on AI visual recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0059] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. By referring to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0060] According to an embodiment of the present invention, a method and system for automatically supplying a quantitative amount of solder to a towel rack based on AI visual recognition are provided.

[0061] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1As shown, according to one embodiment of the present invention, a method for automatically supplying solder to a towel rack based on AI visual recognition is provided. The method for automatically supplying solder to a towel rack based on AI visual recognition includes:

[0062] S1. Based on the image data of the towel rack's welding points, locate the welding points and extract surface features to generate a welding point feature vector set;

[0063] S2. Determine the basic theoretical demand for solder and the solder usage correction value based on the solder point feature vector set, and establish a quantitative supply instruction set by calculating the predicted solder usage;

[0064] S3. Based on the quantitative supply instruction set, the opening time and speed of the screw valve are controlled; the welding result image is collected through an industrial camera, and the material property correction coefficient is adjusted according to the welding result image to optimize the predicted amount of solder in the next welding process.

[0065] In one embodiment, the welding point feature vector set includes geometric size features, surface roughness features, and wettability features of the welding point;

[0066] Based on the image data of the towel rack's welded points, the weld points are located and surface features are extracted to generate a weld point feature vector set including:

[0067] S11. Use an industrial camera to collect multi-angle image data of the towel rack's welded points, locate the welded points using a target detection algorithm, and divide the welded point areas based on a threshold segmentation technique.

[0068] S12. According to the solder joint area, a deep residual network is used to extract the surface features of the solder joint and generate a solder joint feature vector set.

[0069] Specifically, the target detection algorithm of the present invention adopts an improved YOLO algorithm, extracts features from the image through a pre-trained convolutional neural network, uses anchor boxes to generate candidate areas on the feature map, and screens out the welding point position coordinates with the highest confidence through non-maximum suppression; the threshold segmentation technology is based on the grayscale value distribution characteristics of the image, sets a threshold to separate the welding point area from the background, and forms a complete welding point area contour by removing noise and connecting adjacent pixels; the deep residual network extracts the texture, edge and shape information of the welding point through multi-layer convolution operations, and reduces the dimensionality of the extracted high-dimensional features through a fully connected layer, and finally outputs a feature vector set containing geometric dimensions, surface roughness and wettability.

[0070] In one embodiment, determining the basic theoretical demand for solder and the solder usage correction value based on the solder point feature vector set, and calculating the predicted solder usage to establish a quantitative supply instruction set includes:

[0071] S21. Calculate the theoretical solder volume based on the solder joint feature vector set, correct it using the material property correction coefficient, and analyze the basic theoretical demand for solder in combination with the wettability characteristics;

[0072] S22, calculating the solder flow resistance coefficient based on the solder joint feature vector set, searching for similar solder joint cases, using physical information to guide the model, and determining the solder amount correction value;

[0073] S23. Based on the basic theoretical demand for solder and the solder usage correction value, the predicted solder usage is calculated through a weighted fusion algorithm, and a quantitative supply instruction set is generated.

[0074] In one embodiment, the theoretical solder volume is calculated based on the solder joint feature vector set, and is corrected using the material property correction coefficient. Combined with the wettability characteristics, the basic theoretical solder requirement is analyzed, including:

[0075] S211. Calculating a theoretical solder volume required to fill a gap in a solder joint based on geometrical dimension features in a solder joint feature vector set;

[0076] S212. Correcting the theoretical solder volume using a material property correction coefficient to obtain a corrected solder volume; wherein the material property correction coefficient has an initial value of 1 and is dynamically adjusted based on historical supply deviation analysis results;

[0077] S213. Calculate the basic theoretical demand for solder based on the corrected solder volume and the wettability characteristics of the solder joint.

[0078] Specifically, when calculating the theoretical solder volume required to fill the gap between welding points, the present invention decomposes the gap between welding points into a combination of simple geometric shapes: butt welds are calculated according to a rectangular cross-section (length × width × depth), and fillet welds are calculated according to a triangular cross-section (0.5 × contact side length × weld height × length); a material property correction coefficient corrects the theoretical volume by multiplication, and the initial value of the correction coefficient is set to 1.0. After each welding, an adjustment of ±0.05 is made based on the ratio of the actual solder amount to the theoretical calculated amount, with an upper limit of 1.3 and a lower limit of 0.7; the wettability characteristic characterizes the spreading ability of the solder on the surface of the base material, and is determined by the ratio of the solder diffusion area to the theoretical area in the historical welding image, with a range of 0.5-1.5. When the wettability is good, if the coefficient is greater than 1, the amount can be reduced; when the wettability is poor, if the coefficient is less than 1, the amount needs to be increased.

[0079] In one embodiment, calculating the solder flow resistance coefficient based on the solder joint feature vector set, searching for similar solder joint cases, and using physical information to guide the model to determine the solder amount correction value includes:

[0080] S221. Calculating a solder flow resistance coefficient based on surface roughness characteristics of a soldering point feature vector set;

[0081] S222. Based on the solder flow resistance coefficient, searching for solder joint cases whose similarity is not less than a preset threshold from a historical database to obtain a set of similar solder joint cases;

[0082] S223. Based on a case set of similar solder joints, a solder amount correction value for the current solder joint is generated by guiding the model through physical information.

[0083] In one embodiment, calculating the solder flow resistance coefficient according to the surface roughness characteristics of the solder joint feature vector set includes:

[0084] S2211. Extract surface roughness features based on the weld feature vector set and analyze the average roughness value, peak-to-valley distribution, and directionality index;

[0085] S2212. Establish a microscopic fluid dynamics model based on the average roughness value, peak-to-valley distribution, and directionality index;

[0086] S2213. Calculate the solder flow resistance coefficient of the current soldering point through a microscopic fluid dynamics model.

[0087] Specifically, when the present invention extracts surface roughness features based on the weld feature vector set, the four texture statistics of image contrast, correlation, energy and homogeneity are first calculated through the gray level co-occurrence matrix (GLCM), and these statistics are normalized to the range of 0-100 to represent the average roughness value; the peak-to-valley distribution coefficient is obtained by performing wavelet transform analysis on the surface contour line, and the ratio of the peak-to-valley height variance to the density is calculated. The larger the value, the more significant the surface undulation; the directionality index analyzes the main direction angle and intensity of the surface texture through Fourier transform, and is quantified into a value between 0 and 1, where 0 indicates no obvious directionality and 1 indicates strong directionality.

[0088] Specifically, the present invention constructs a microscopic fluid dynamics model based on the modified Washburn equation, which is expressed as follows:

[0089] R=K1×R a λ (1+K2×P d )×(1+K3×D i );

[0090] Where R is the solder flow resistance coefficient; R a is the average roughness value; P d is the peak-valley distribution coefficient; D iis a directional index; K1, K2, K3 and λ are the roughness influence coefficient, peak-valley distribution influence coefficient, directionality influence coefficient and resistance adjustment constant, respectively, with values ​​of 0.05, 0.2, 0.15 and 1.5; the final flow resistance coefficient is standardized to the range of 0.8-2.0 for precise adjustment of subsequent solder dosage.

[0091] In one embodiment, based on the solder flow resistance coefficient, solder joint cases with similarity not less than a preset threshold are retrieved from a historical database, and a similar solder joint case set is obtained, including:

[0092] S2221. Combining the flow resistance coefficient of the current welding point with the surface roughness feature to form a query vector, and performing normalization processing on the query vector;

[0093] S2222. Calculate the similarity score between the query vector and the feature vectors of each welding point case in the historical database using a cosine similarity algorithm;

[0094] S2223. Filter historical welding point cases whose similarity scores are not lower than a preset threshold, sort them from high to low according to similarity, and take the top N cases as the similar welding point case set.

[0095] Specifically, when searching for similar welding point cases, the present invention first compares the flow resistance coefficient R of the current welding point with the surface roughness characteristics (including the average roughness value R a , peak-valley distribution P d and directional indicator D i ) into a multi-dimensional query vector [R,R a ,P d ,D i ,R×R a ,R×P d ,R×D i ,R a ×P d ×D i ], the query vector contains basic features and a small number of cross terms to reflect the correlation between features; in order to eliminate the influence of different dimensions, the query vector is normalized so that the data of each dimension are distributed in the same range, which is convenient for subsequent similarity calculation; the similarity calculation adopts the cosine similarity algorithm, which judges the degree of directional similarity by calculating the cosine value of the angle between vectors, and is more suitable for similarity evaluation in high-dimensional feature space than Euclidean distance; the preset similarity threshold can be adjusted according to the actual application scenario, and is usually set at a high level to ensure that the retrieved cases are sufficiently similar to the current weld; the case set size N value is determined based on statistical stability considerations. When the feature value is within the conventional range, it takes a smaller value, and when encountering unconventional working conditions, it is appropriately increased to balance the computational efficiency and the richness of reference information.

[0096] In one embodiment, generating a solder amount correction value for a current solder joint using a physical information-guided model based on a set of similar solder joint cases includes:

[0097] S2231. Extract historical solder usage data and corresponding welding quality evaluation results from a collection of similar welding point cases;

[0098] S2232. Based on the deep neural network model, a physical constraint layer is added that uses the flow resistance coefficient as the main parameter input to obtain a physical information guidance model;

[0099] S2233. Use a set of similar welding point cases to train a physical information guided model, and use weighted averaging to perform feature fusion on similar cases;

[0100] S2234. Input the characteristic vector and flow resistance coefficient of the current soldering point into the trained physical information guidance model, and output the solder amount correction value of the current soldering point.

[0101] Specifically, when generating a solder usage correction value, the present invention first extracts the actual solder usage, theoretical usage, flow resistance coefficient and welding quality score (excellent, qualified, unqualified) of historical solder joints from a set of similar solder joint cases, focusing on cases with an "excellent" quality evaluation; the physical constraint layer is designed with a three-layer structure, the input layer receives the flow resistance coefficient and other characteristic parameters, the middle layer adds the physical constraint equation P(R) = k×ln(R+1) (where R is the flow resistance coefficient and k is a learnable parameter), and the output layer generates the correction coefficient; during the training process, similar solder joint cases are weighted, and the higher the similarity, the greater the weight of the case, to avoid interference of low-correlation cases on the model; during feature fusion, the solder joint cases are first sorted by similarity, and then the feature vectors of the first N cases are weighted averaged, with the weight proportional to the similarity; finally, the current solder joint features and the flow resistance coefficient are input into the trained model, and a correction value between 0.85 and 1.15 is directly output to adjust the basic theoretical demand, thereby ensuring welding quality while avoiding material waste.

[0102] Specifically, the expression of the physical information guidance model in the present invention is:

[0103] M adj =f(X,R)=β×[NN(X)+α×P(R)];

[0104] Where M adjis the solder dosage correction value, which is used to adjust the basic theoretical demand; X is the solder joint feature vector set, which includes the aforementioned geometric size characteristics, surface roughness characteristics and wettability characteristics; R is the flow resistance coefficient, which is calculated by step S221; NN(X) is the processing output of the deep neural network for feature X; P(R) is the form of a physical constraint equation, which is a functional equation used to simulate the flow law of solder under different resistance conditions; α is the physical constraint weight coefficient, which is used to control the degree of influence of the physical law in the model; β is the global scaling factor, which is used to ensure that the correction value is within a reasonable range.

[0105] Specifically, the feature fusion process of the present invention uses a weighted average approach. First, the similarity between the current solder joint and each case in the historical case set is calculated as a weight. These similarity values ​​are calculated in step S2222. The feature vector of each historical case is then multiplied by its corresponding similarity weight, and all weighted feature vectors are summed up. Finally, this sum is divided by the sum of all similarity weights to obtain the fused feature vector. This approach ensures that historical cases that are more similar to the current solder joint have greater influence in the fusion process, effectively utilizing the most relevant information from historical experience.

[0106] In one embodiment, based on the basic theoretical demand for solder and the solder usage correction value, a weighted fusion algorithm is used to calculate the predicted solder usage, and a quantitative supply instruction set is generated, including:

[0107] S231. Calculate the predicted solder amount by combining the basic theoretical demand and the solder amount correction value through an adaptive weighted fusion algorithm;

[0108] S232, based on the predicted amount of solder, combined with process condition constraints and equipment parameters, converting it into instruction parameters executable by the feeding equipment;

[0109] S233. Generate a quantitative supply instruction set including a feeding rate, a feeding time, and a trigger condition according to the instruction parameters.

[0110] Specifically, the weighted fusion algorithm of the present invention adopts a dynamic weight allocation mechanism: first, the initial weight of the basic theoretical demand is set to 0.65, and the initial weight of the correction value is set to 0.35; then, according to the historical feeding accuracy evaluation results, the weight ratio is dynamically adjusted. When the historical data shows that the correction value prediction accuracy is higher than 95%, the correction value weight is increased by 0.05 each time, with an upper limit of 0.6; when the accuracy is lower than 85%, the correction value weight is reduced by 0.05 each time, with a lower limit of 0.25; the final predicted usage is obtained by weighted summation of the above weights.

[0111] Specifically, the present invention converts the predicted solder dosage into a set of quantitative supply instructions using PLC control technology combined with a mass flow ratio controller to calculate the required solder feeding duration and flow rate based on the predicted dosage. In a specific application, a Siemens S7 series PLC is interfaced with a dedicated screw valve control unit. The predicted dosage is converted into a 0-10V analog signal and a control pulse signal via an industrial communication bus. The analog signal controls the motor speed (corresponding to the flow rate), while the pulse signal controls the feeding time.

[0112] In one embodiment, based on a quantitative supply instruction set, the opening time and rotation speed of the screw valve are controlled; an industrial camera is used to capture a welding result image, and a material property correction coefficient is adjusted based on the welding result image to optimize the predicted amount of solder used in the next welding process, including:

[0113] S31. According to the quantitative supply instruction set, the predicted amount of solder is converted into motor control parameters, and the screw valve motor is driven to rotate to achieve quantitative supply of solder;

[0114] S32. After the welding is completed and cooled, an industrial camera is used to capture an image of the welding result, measure the actual filling area and edge contour of the welding point, and compare the difference between the actual size of the welding point and the ideal size;

[0115] S33. Based on the comparison results, when a material shortage occurs at the welding point, it is recorded as a negative deviation; when a material overflow occurs at the welding point, it is recorded as a positive deviation;

[0116] S34. Determine an adjustment coefficient based on the deviation value. For a negative deviation, increase the correction coefficient by a preset ratio. For a positive deviation, decrease the correction coefficient by a preset ratio to optimize the predicted amount of solder in the next welding process.

[0117] Specifically, the present invention adopts a piecewise linear mapping method when converting the predicted solder dosage into motor control parameters, and sets the corresponding screw valve speed and control signal strength according to different dosage ranges. A low speed is used in a small dosage range to improve accuracy, and the speed is increased accordingly in a large dosage range to ensure efficiency; the opening time is calculated by dividing the predicted solder dosage by the screw discharge rate, where the screw discharge rate is equal to the solder density multiplied by the screw displacement per revolution multiplied by the flow efficiency coefficient, which is used to compensate for material slippage and compression in the actual feeding process.

[0118] Specifically, the actual fill area of ​​the solder joint in the present invention is measured using an image processing algorithm based on OpenCV. First, Gaussian filtering is performed to eliminate noise, then the threshold segmentation method is used to extract the solder joint area, and finally the solder joint boundary is determined by the region growing method. The measured data is compared with the ideal size in the CAD model. A fill rate below 95% is recorded as a negative deviation, and a fill rate above 105% is recorded as a positive deviation. The calculation formula of the adjustment coefficient is: new correction coefficient = original correction coefficient × (1 ± |deviation rate| × 0.3), where the deviation rate is the percentage difference between the actual fill area and the ideal area. Positive deviations take a negative sign, and negative deviations take a positive sign. The single adjustment range is limited to within ±15% to prevent over-compensation of the system.

[0119] In order to facilitate understanding of the above technical solution of the present invention, the following is a detailed description using a towel rack welding production line as an example:

[0120] The present invention is applied to the visual point copper paste machine on the towel rack welding production line. Figure 2 As shown, the production line is equipped with two industrial cameras and a visual copper paste dot machine, which connects the manual loading station and the manual unloading station through an assembly line.

[0121] During the welding process, first, the loading worker places the towel rack parts to be welded on the assembly line and transports them to the first station of the visual point copper paste machine. Figure 3 As shown, the visual copper paste machine includes a display, a pressure regulating valve and pressure gauge, a copper paste barrel, and the core actuator—the screw valve. An industrial camera captures multi-angle images of the towel rack's solder joints. An improved YOLO algorithm accurately locates the solder joints. A deep residual network is then used to extract the solder joint's geometric dimensions, surface roughness, and wettability characteristics, forming a feature vector set.

[0122] Then, based on the acquired feature vector set, the theoretical solder volume required to fill the solder joint gap is calculated and corrected using the material property correction factor. Simultaneously, the solder joint surface roughness is analyzed to calculate the flow resistance coefficient. Similar solder joint cases are retrieved from a historical database, and the model is guided by physical information to generate a corrected solder volume. The basic theoretical requirement and the corrected volume are combined using an adaptive weighted fusion algorithm to produce an accurate predicted solder volume.

[0123] The controller of the visual copper paste dispenser then converts the predicted dosage into control parameters for the screw valve, adjusting the opening time and speed to achieve quantitative supply. The operating pressure can be adjusted in real time using a pressure regulating valve and pressure gauge, and the display shows the current parameters and recognition results in real time.

[0124] Finally, after the workpiece is fed, it moves to the welding station for welding. After the weld is complete and cooled, an industrial camera at the second station captures the weld image, measures the actual fill area and edge contour of the weld, and compares the actual dimensions with the ideal dimensions. Based on this comparison, it determines whether there is material shortage or overflow, and adjusts the material property correction factor accordingly to optimize the predicted amount of material for the next welding process.

[0125] This invention combines AI visual recognition technology with physical models to achieve precise quantitative solder supply. Practice has shown that using this method can control solder dosage deviations to within ±5%, reducing material waste by 30% and increasing production efficiency by 15%.

[0126] like Figure 4 As shown, according to another embodiment of the present invention, there is also provided a towel rack solder quantitative automatic supply system based on AI visual recognition, the towel rack solder quantitative automatic supply system based on AI visual recognition includes:

[0127] Welding point identification and positioning module 1 is used to locate the welding point position and extract surface features based on the image data of the towel rack welding point to generate a welding point feature vector set;

[0128] Solder usage prediction module 2 is used to determine the basic theoretical demand for solder and the solder usage correction value based on the solder point feature vector set, and to establish a quantitative supply instruction set by calculating the predicted solder usage;

[0129] The feeding control optimization module 3 is used to control the opening time and speed of the screw valve based on the quantitative feeding instruction set; the welding result image is collected by an industrial camera, and the material property correction coefficient is adjusted according to the welding result image to optimize the predicted amount of solder in the next welding process.

[0130] In summary, with the help of the above technical solutions of the present invention, the present invention realizes precise positioning of welding points, feature extraction and precise control of solder usage through the towel rack solder quantitative automatic supply method based on AI visual recognition, effectively solving the problems of inaccurate solder supply and unstable welding quality in traditional welding processes, and improving the automation level and product quality of towel rack production; based on the extracted solder point feature vector set, the present invention calculates the theoretical solder volume and corrects it in combination with the material property correction coefficient, and uses the flow resistance coefficient and physical information to guide the model to generate the solder usage correction value, thereby realizing precise control of the solder supply amount, and can automatically adjust the solder usage according to the actual needs of different welding positions, effectively avoiding the problem of excessive solder in some areas and insufficient solder in other areas in traditional welding equipment; the present invention uses an industrial camera to collect welding result images, compares the difference between the actual size of the solder point and the ideal size, and dynamically adjusts the material property correction coefficient according to the deviation value, constructing a complete closed-loop feedback mechanism, which can continuously learn and optimize from historical data, adapt to the welding needs of towel racks of different materials and structures, realize continuous optimization and self-improvement of process parameters, and improve production efficiency and product consistency.

[0131] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for automatically supplying solder to a towel rack based on AI visual recognition, characterized in that: The method for automatically supplying solder to a towel rack based on AI visual recognition includes: S1. Based on the image data of the towel rack's welding points, locate the welding points and extract surface features to generate a welding point feature vector set; S2. Determine the basic theoretical demand for solder and the solder usage correction value based on the solder point feature vector set, and establish a quantitative supply instruction set by calculating the predicted solder usage; S3. Based on the quantitative supply instruction set, the opening time and speed of the screw valve are controlled; the welding result image is collected through an industrial camera, and the material property correction coefficient is adjusted according to the welding result image to optimize the predicted amount of solder in the next welding process.

2. The method for automatically supplying solder to a towel rack based on AI visual recognition according to claim 1, characterized in that: The welding point feature vector set includes geometric size features, surface roughness features and wettability features of the welding point; The method of locating the welding point position and extracting the surface features based on the image data of the towel rack welding point to generate the welding point feature vector set includes: S11. Use an industrial camera to collect multi-angle image data of the towel rack's welded points, locate the welded points using a target detection algorithm, and divide the welded point areas based on a threshold segmentation technique. S12. According to the solder joint area, a deep residual network is used to extract the surface features of the solder joint and generate a solder joint feature vector set.

3. The method for automatically supplying solder to a towel rack based on AI visual recognition according to claim 1, characterized in that: The method of determining the basic theoretical demand for solder and the solder usage correction value based on the solder point feature vector set, and establishing a quantitative supply instruction set by calculating the predicted solder usage includes: S21. Calculate the theoretical solder volume based on the solder joint feature vector set, correct it using the material property correction coefficient, and analyze the basic theoretical demand for solder in combination with the wettability characteristics; S22, calculating the solder flow resistance coefficient based on the solder joint feature vector set, searching for similar solder joint cases, using physical information to guide the model, and determining the solder amount correction value; S23. Based on the basic theoretical demand for solder and the solder usage correction value, the predicted solder usage is calculated through a weighted fusion algorithm, and a quantitative supply instruction set is generated.

4. The method for automatically supplying solder to a towel rack based on AI visual recognition according to claim 3 is characterized in that: The theoretical solder volume is calculated based on the solder joint feature vector set, and is corrected using the material property correction coefficient. Combined with the wettability characteristics, the basic theoretical demand for solder is analyzed, including: S211. Calculating a theoretical solder volume required to fill a gap in a solder joint based on geometrical dimension features in a solder joint feature vector set; S212. Correcting the theoretical solder volume using a material property correction coefficient to obtain a corrected solder volume; wherein the material property correction coefficient has an initial value of 1 and is dynamically adjusted based on historical supply deviation analysis results; S213. Calculate the basic theoretical demand for solder based on the corrected solder volume and the wettability characteristics of the solder joint.

5. The method for automatically supplying solder to a towel rack based on AI visual recognition according to claim 3 is characterized in that: The calculation of the solder flow resistance coefficient based on the solder joint feature vector set, searching for similar solder joint cases, and using physical information to guide the model to determine the solder amount correction value includes: S221. Calculating a solder flow resistance coefficient based on surface roughness characteristics of a soldering point feature vector set; S222. Based on the solder flow resistance coefficient, searching for solder joint cases whose similarity is not less than a preset threshold from a historical database to obtain a set of similar solder joint cases; S223. Based on a case set of similar solder joints, a solder amount correction value for the current solder joint is generated by guiding the model through physical information.

6. The method for automatically supplying solder to a towel rack based on AI visual recognition according to claim 5, characterized in that: Calculating the solder flow resistance coefficient according to the surface roughness characteristics of the solder point feature vector set includes: S2211. Extract surface roughness features based on the weld feature vector set and analyze the average roughness value, peak-to-valley distribution, and directionality index; S2212. Establish a microscopic fluid dynamics model based on the average roughness value, peak-to-valley distribution, and directionality index; S2213. Calculate the solder flow resistance coefficient of the current soldering point through a microscopic fluid dynamics model.

7. The method for automatically supplying solder to a towel rack based on AI visual recognition according to claim 5, characterized in that: The method of retrieving solder joint cases whose similarity is not less than a preset threshold from a historical database based on the solder flow resistance coefficient to obtain a similar solder joint case set includes: S2221. Combining the flow resistance coefficient of the current welding point with the surface roughness feature to form a query vector, and performing normalization processing on the query vector; S2222. Calculate the similarity score between the query vector and the feature vectors of each welding point case in the historical database using a cosine similarity algorithm; S2223. Filter historical welding point cases whose similarity scores are not lower than a preset threshold, sort them from high to low according to similarity, and take the top N cases as the similar welding point case set.

8. The method for automatically supplying solder to a towel rack based on AI visual recognition according to claim 5, characterized in that: The method of generating a solder amount correction value for a current soldering point by guiding a model based on a set of similar soldering point cases includes: S2231. Extract historical solder usage data and corresponding welding quality evaluation results from a collection of similar welding point cases; S2232. Based on the deep neural network model, a physical constraint layer is added that uses the flow resistance coefficient as the main parameter input to obtain a physical information guidance model; S2233. Use a set of similar welding point cases to train a physical information guided model, and use weighted averaging to perform feature fusion on similar cases; S2234. Input the characteristic vector and flow resistance coefficient of the current soldering point into the trained physical information guidance model, and output the solder amount correction value of the current soldering point.

9. The method for automatically supplying solder to a towel rack based on AI visual recognition according to claim 1, characterized in that: The said quantitative supply instruction set is based on controlling the opening time and speed of the screw valve; The welding result image is captured by an industrial camera, and the material property correction coefficient is adjusted based on the welding result image to optimize the predicted solder amount in the next welding process, including: S31. According to the quantitative supply instruction set, the predicted amount of solder is converted into motor control parameters, and the screw valve motor is driven to rotate to achieve quantitative supply of solder; S32. After the welding is completed and cooled, an industrial camera is used to capture an image of the welding result, measure the actual filling area and edge contour of the welding point, and compare the difference between the actual size of the welding point and the ideal size; S33. Based on the comparison results, when a material shortage occurs at the welding point, it is recorded as a negative deviation; when a material overflow occurs at the welding point, it is recorded as a positive deviation; S34. Determine an adjustment coefficient based on the deviation value. For a negative deviation, increase the correction coefficient by a preset ratio. For a positive deviation, decrease the correction coefficient by a preset ratio to optimize the predicted amount of solder in the next welding process.

10. A towel rack solder quantitative automatic supply system based on AI visual recognition, used to implement the towel rack solder quantitative automatic supply method based on AI visual recognition according to any one of claims 1 to 9, characterized in that: The system includes: The welding point identification and positioning module is used to locate the welding point position and extract surface features based on the image data of the towel rack welding point to generate a welding point feature vector set; The solder consumption prediction module is used to determine the basic theoretical demand for solder and the solder consumption correction value based on the solder point feature vector set, and to establish a quantitative supply instruction set by calculating the predicted solder consumption; The feeding control optimization module is used to control the opening time and speed of the screw valve based on the quantitative feeding instruction set; the welding result image is collected through an industrial camera, and the material property correction coefficient is adjusted according to the welding result image to optimize the predicted solder amount in the next welding process.

Images