An execution system for an ultrasonic welding machine

Abstract

The present application relates to the technical field of welding machine, disclose a kind of execution system of ultrasonic welding machine, based on preset rule initialization welding point scheme to test base material;Welding point scheme includes: the position coordinates of M welding points and the welding sequence of corresponding welding point;Based on welding point scheme, welding is carried out to test base material, and test piece is obtained;For generating topological network based on welding point scheme;For obtaining N test pieces and N corresponding topological networks by data generation module, welding point welding module and topological construction module, and the corresponding welding score of N test pieces is obtained based on physical test;For constructing topological prediction model based on N topological networks and corresponding N welding scores;Based on topological prediction model, target welding scheme is obtained in combination with optimization algorithm.

Description

Technical Field

[0001] This invention relates to the field of welding machine technology, and more specifically, to an execution system for an ultrasonic welding machine. Background Technology

[0002] Existing ultrasonic welding systems typically employ a method of prioritizing weld point fixation on the base material before welding with wire bonding between these points. This approach aims to prevent the base material from shifting due to high-frequency friction from the welding head, thereby improving weld quality. However, current systems rely on manual setting of weld point positions and their sequence based on human experience, resulting in welding efficiency and quality that cannot meet the high demands of automation. Summary of the Invention

[0003] This invention provides an execution system for an ultrasonic welding machine, which solves the technical problems mentioned in the background art.

[0004] In a first aspect, the present invention provides an execution system for an ultrasonic welding machine, comprising:

[0005] The data generation module is used to initialize the weld point scheme for the test base material based on preset rules; the weld point scheme includes: the position coordinates of M weld points and the corresponding welding order of the weld points;

[0006] The solder joint welding module is used to weld the test base material based on the solder joint scheme to obtain the test piece;

[0007] The topology building module is used to generate topology networks based on solder joint schemes.

[0008] The training sample module is used to obtain N test pieces and N corresponding topology networks through the data generation module, weld point welding module and topology construction module, and to obtain the corresponding welding scores of the N test pieces based on physical testing;

[0009] The topology prediction module is used to construct a topology prediction model based on N topology networks and corresponding N welding scores.

[0010] The planning module is used to obtain the target welding scheme based on the topology prediction model and the optimization algorithm.

[0011] Furthermore, the solder joint initialization scheme based on preset rules includes:

[0012] Preset rules are used to constrain random solder joints in the solder joint scheme, specifically including:

[0013] The first and second weld points in the welding scheme are located at both ends of the weld on the test base material;

[0014] The spacing between solder joints that are not the first solder joint and the second solder joint in the solder joint scheme is less than a preset spacing threshold.

[0015] The unit weld density of the weld seam in the test base material is within the preset threshold range.

[0016] Furthermore, a topology network is generated based on the solder joint scheme, including:

[0017] Map each of the M solder joints to a topology node of the topology network;

[0018] Based on the welding order of the weld points, construct topological edges between topological nodes of adjacent welding orders;

[0019] The feature vector of the corresponding topological node is constructed based on the position coordinates of the solder joint.

[0020] Further physical testing includes weld strength testing, weld integrity testing, and appearance quality testing; the weld scores for N test pieces are obtained based on these physical tests, including:

[0021] Obtain the corresponding test data for welding strength, welding integrity, and appearance quality;

[0022] Calculate the ratio of the weld strength to the maximum weld strength to obtain the weld strength score;

[0023] Calculate the ratio of the non-welded intact area to the average non-welded intact area to obtain the weld integrity score;

[0024] The appearance quality score is determined based on the weld integrity test and the appearance quality test.

[0025] The weld strength score, weld integrity score, and appearance quality score are combined by weighted multiplication to obtain the first index value:

[0026] Calculate the weighted value of appearance quality based on the appearance quality score;

[0027] The normalized weld strength is added to the non-welded complete area to obtain the comprehensive index value of mechanical properties. Based on the comprehensive index value of mechanical properties, the weighted value of mechanical properties is calculated.

[0028] Calculate the ratio of the weighted value of appearance quality to the weighted value of mechanical properties, and then calculate its arctangent value based on the ratio:

[0029] The first index value is added to the arctangent value and then normalized to obtain the welding score of the test piece.

[0030] Furthermore, topology prediction models include:

[0031] A topology prediction model is constructed based on N topological networks and their corresponding N welding scores.

[0032] For each topology network, output an adjacency matrix. The element in the i-th row and j-th column of the adjacency matrix represents the connection relationship between the i-th node and the j-th node in the topology network. If there is an edge between the i-th node and the j-th node, the element value of the corresponding element is 1; otherwise, the element value of the corresponding element is 0.

[0033] For each topology network, output a diagonal sparse matrix. The element in the m-th row and m-th column of the diagonal sparse matrix represents the eigenvector of the m-th node. All other elements of the diagonal sparse matrix are 0.

[0034] The elements in the same row and column of the adjacency matrix and diagonal sparse matrix of each topology network are merged to obtain N feature matrices of size M×M.

[0035] The feature matrix is ​​used as the training sample, and the corresponding welding score is used as the sample label to train the topology prediction model, which includes a hidden layer and a classifier.

[0036] Hidden layers are used to perform nonlinear transformations on the input feature matrix to obtain hidden states;

[0037] A classifier is used as input to the hidden state and outputs a predicted welding score.

[0038] The mean error variance of the predicted welding score and the corresponding sample label is used as the loss function to update the weight matrix and bias matrix of the hidden layer of the topology prediction model through backpropagation.

[0039] Furthermore, the target welding scheme is obtained, including:

[0040] Step 1: Initialize the population. The population consists of R individuals that conform to preset rules, and each individual corresponds to a random welding scheme.

[0041] Step 2, the objective function is constructed based on the topology prediction model, specifically including:

[0042] Based on the welding scheme of an individual, a corresponding topology network is constructed, and the feature matrix of the topology network is obtained. The feature matrix is ​​input into the topology prediction model to obtain the predicted welding score of the topology prediction model. The predicted welding score is used as the objective function value of the corresponding individual.

[0043] Step 3: Obtain the average value of the objective function for each individual, retain individuals whose objective function value is higher than the average value, and use the remaining individuals as the parent generation for iteration;

[0044] Step 4: Repeat step 3 a preset number of times to obtain the welding scheme corresponding to the individual with the largest objective function value as the target welding scheme.

[0045] Furthermore, the remaining individuals are used as parents for iteration, including the first and second phases:

[0046] The first phase of updates is as follows:

[0047] Based on the individual encoding vector of the current iteration, multiply it by the first update weight vector to obtain the basic inheritance term; divide the baseline fitness constant by the individual's ranking position in the current iteration, and perform a nonlinear transformation on the fitness using a sine function to obtain the fitness adjustment term;

[0048] Extract the diagonal elements of the adjacency matrix corresponding to the current individual code, form a diagonal vector, and obtain the adjacency matrix perturbation term;

[0049] Calculate the individual encoding vector P(t+1) for the (t+1)th iteration based on the basic inheritance term, fitness adjustment term, and adjacency matrix perturbation term;

[0050] The second phase of updates is as follows:

[0051] Based on the individual encoding vector of the current iteration, multiply it by the random perturbation vector to obtain the random perturbation term;

[0052] Identify the individual with the largest objective function value in the current iteration, extract its encoding vector, calculate the absolute value of the encoding vector, and take the natural logarithm of the absolute value of the encoding vector to obtain the globally optimal guiding term:

[0053] The random perturbation term is added to the globally optimal guiding term and processed by the ReLU activation function to obtain the individual encoding vector P(t+1) for the (t+1)th iteration.

[0054] In a second aspect, a computing device includes:

[0055] One or more processors;

[0056] A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the system as described in any one of claims 1 to 8.

[0057] Thirdly, a computer-readable storage medium storing a program that, when executed by a processor, implements the system described herein.

[0058] The beneficial effects of this invention are as follows: by combining topological network modeling, deep learning prediction, and evolutionary optimization algorithms, it can efficiently optimize ultrasonic welding schemes and achieve intelligent layout and sequence optimization of welding points. Compared with traditional methods, it significantly improves welding quality, reduces manual intervention, and comprehensively enhances the automation and intelligence level of the welding process. Attached Figure Description

[0059] Figure 1 This is a block diagram of the execution system of an ultrasonic welding machine according to the present invention. Detailed Implementation

[0060] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0061] like Figure 1 As shown, an execution system for an ultrasonic welding machine includes:

[0062] The data generation module is used to initialize the weld point scheme for the test base material based on preset rules; the weld point scheme includes: the position coordinates of M weld points and the corresponding welding order of the weld points;

[0063] The solder joint welding module is used to weld the test base material based on the solder joint scheme to obtain the test piece;

[0064] The topology building module is used to generate topology networks based on solder joint schemes.

[0065] The training sample module is used to obtain N test pieces and N corresponding topology networks through the data generation module, weld point welding module and topology construction module, and to obtain the corresponding welding scores of the N test pieces based on physical testing;

[0066] The topology prediction module is used to construct a topology prediction model based on N topology networks and corresponding N welding scores.

[0067] The planning module is used to obtain the target welding scheme based on the topology prediction model and the optimization algorithm.

[0068] In one embodiment of the present invention, the solder joint initialization scheme based on preset rules includes:

[0069] Preset rules are used to constrain random solder joints in the solder joint scheme, specifically including:

[0070] The first and second weld points in the welding scheme are located at both ends of the weld on the test base material;

[0071] The spacing between solder joints that are not the first solder joint and the second solder joint in the solder joint scheme is less than a preset spacing threshold.

[0072] The unit weld density of the weld seam in the test base material is within the preset threshold range.

[0073] Specifically, the first and second weld points must be located at both ends of the weld. The main purpose of this rule is to ensure sufficient anchor points at both ends of the weld, preventing the base material from loosening or shifting during welding due to a lack of endpoint anchoring. This constraint provides stability for subsequent welding operations, significantly improving weld quality, especially when the base material is large or has a complex shape. In practice, the start and end points of the weld on the base material are automatically identified, and the positions of the first and second weld points are automatically assigned by the system based on the weld boundary. The spacing between non-endpoint weld points must be less than a preset spacing threshold. This rule limits the distance between weld points, ensuring a uniform distribution of overall weld strength. Excessive weld point spacing can lead to insufficient weld strength in high-stress areas, making it prone to fracture or performance problems. Setting a threshold can prevent such problems. In practice, the system dynamically adjusts the spacing threshold based on the size of the base material, the length of the weld, and the required weld strength. For example, the spacing threshold for smaller base materials can be set smaller to meet higher weld strength requirements. The unit weld point density must be within a preset threshold range. The unit weld point density refers to the number of weld points per unit length of weld. Excessive weld density can reduce welding efficiency and may even cause deformation of the base material; conversely, insufficient weld density can lead to inadequate overall strength and stability of the weld. Introducing a preset threshold range allows for a balance between welding quality and efficiency. The specific density range can be determined based on empirical data.

[0074] In one embodiment of the present invention, generating a topology network based on a solder joint scheme includes:

[0075] Map each of the M solder joints to a topology node of the topology network;

[0076] Based on the welding order of the weld points, construct topological edges between topological nodes of adjacent welding orders;

[0077] The feature vector of the corresponding topological node is constructed based on the position coordinates of the solder joint.

[0078] Specifically, the nodes in the topology network correspond to the weld points in the welding scheme, and the position of each weld point is directly mapped to a topology node in the network. This mapping transforms the physical distribution of weld points into network nodes, laying the foundation for subsequent construction of topology relationships and analysis of welding schemes. The number of M weld points corresponds one-to-one with the number of nodes in the network, and this mapping allows the distribution of weld points to be intuitively expressed through mathematical networks. Topology edges are the lines connecting topology nodes, representing the welding order of adjacent weld points during the welding process. If the welding order of the weld points is P1→P2→P3→......→PM, then in the network, nodes N1→N2→N3→......→NM are connected by edges in sequence. Based on the process constraints brought about by the welding order and the dynamic relationships during the welding process, the construction of topology edges can reflect the dependencies between weld points, thereby better analyzing the impact of different welding orders on the overall welding quality. Feature vectors are a set of values ​​used to represent the characteristics of weld points. They typically contain the position coordinates of the weld points (e.g., in a two-dimensional plane). or in three-dimensional space Each topological node represents the spatial location of the weld point through a feature vector. By transforming the weld point distribution into a topological network model, a novel approach to representing welding schemes is proposed, combining traditional welding processes with modern data-driven technologies (such as topology analysis and machine learning). Utilizing a topological network to express the order and relationships of weld points overcomes the limitations of traditional welding methods that rely solely on weld point location and order.

[0079] In one embodiment of the present invention, welding scores are obtained from N test pieces based on physical testing, including:

[0080] Physical tests include: weld strength test, weld integrity test, and appearance quality test;

[0081] Obtain the corresponding test data for weld strength, weld integrity, and appearance quality, and calculate the weld score of the test piece based on the test data. The calculation formula is as follows: ;

[0082] in, This indicates the welding score of the test piece. Indicates the weld strength of the test piece. Indicates the maximum weld strength of the test piece. Represents the natural base. This represents the area of ​​the unwelded, intact test piece. This represents the average area of ​​all unwelded intact sections of the test specimens. This indicates the appearance quality score of the test piece. , Represents the natural logarithm function. , and These represent the first, second, and third rating coefficients, respectively. , and Both are not 0, and , and The sum is 1. This indicates the adjusted weight for the appearance score, when ,but Equal to 1.5, when ,but It equals 0.8. Represents the arctangent function. express Activation function;

[0083] in, The value is obtained based on the average of several human ratings, with the human ratings ranging from 0 to 100.

[0084] Specifically, physical testing is used to evaluate weld quality, including the following three aspects: Weld strength testing: assesses the mechanical properties of the weld joint to ensure its strength and durability. Weld integrity testing: checks the continuity between weld joints and identifies unwelded or defective areas. Appearance quality testing: evaluates the weld based on its appearance characteristics (such as smoothness, uniformity, etc.). The results of these tests are combined to derive a weld score for the test piece, providing a quantitative basis for optimizing welding schemes. This comprehensive physical testing evaluation of weld quality, integrating strength, integrity, and appearance quality, forms a quantitative weld score. This score not only directly reflects the quality of the weld but also provides core data support for the subsequent training and optimization of the topology prediction model.

[0085] In one embodiment of the present invention, the topology prediction model includes:

[0086] A topology prediction model is constructed based on N topological networks and their corresponding N welding scores.

[0087] For each topology network, output an adjacency matrix. The element in the i-th row and j-th column of the adjacency matrix represents the connection relationship between the i-th node and the j-th node in the topology network. If there is an edge between the i-th node and the j-th node, the element value of the corresponding element is 1; otherwise, the element value of the corresponding element is 0.

[0088] For each topology network, output a diagonal sparse matrix. The element in the m-th row and m-th column of the diagonal sparse matrix represents the eigenvector of the m-th node. All other elements of the diagonal sparse matrix are 0.

[0089] The elements in the same row and column of the adjacency matrix and diagonal sparse matrix of each topology network are merged to obtain N feature matrices of size M×M.

[0090] The feature matrix is ​​used as the training sample, and the corresponding welding score is used as the sample label to train the topology prediction model, which includes a hidden layer and a classifier.

[0091] Hidden layers are used to perform nonlinear transformations on the input feature matrix to obtain hidden states;

[0092] A classifier is used as input to the hidden state and outputs a predicted welding score.

[0093] The mean error variance of the predicted welding score and the corresponding sample label is used as the loss function to update the weight matrix and bias matrix of the hidden layer of the topology prediction model through backpropagation.

[0094] Specifically, the adjacency matrix is ​​a mathematical tool used to represent the connectivity between weld points in a topological network. Each element in the matrix corresponds to whether there is a connection between two weld points (i.e., whether there is a welding order relationship). The adjacency matrix clearly represents the welding order and the relationships between weld points. The use of the adjacency matrix allows the dynamic relationships between weld points to be quantified and analyzed, providing a foundation for subsequent model training. The feature matrix is ​​another input matrix used to represent the feature information of the weld points. The features of each weld point can include its coordinates. The feature matrix assigns characteristics to each weld point, helping the model understand the role of the weld point in the network. It supplements the weld point location information that the adjacency matrix cannot reflect. The hidden layer processes the input adjacency matrix and feature matrix to extract the relationships and feature information between weld points. The hidden layer learns the latent characteristics of the welding network through multiple nonlinear transformations. The output of the hidden layer is input into the classifier to generate prediction results for welding scores. It learns the relationship between different topological structures (weld point order and connectivity) and welding scores, providing score predictions for new welding schemes. Through training and optimization, the model has strong generalization ability and can be applied to different welding scenarios. The optimized model can quickly predict the score of new solutions, significantly improving the efficiency of welding solution design. The innovation of the topology prediction model lies in combining welding solution quality assessment with topology network modeling. This method constructs a topology prediction model using adjacency and feature matrices, providing efficient and intelligent welding score prediction capabilities and laying the foundation for welding solution optimization.

[0095] In one embodiment of the present invention, the hidden layer calculation formula of the topology prediction model is as follows: ; ; ;

[0096] in, This represents the hidden state matrix output by the k-th hidden layer. This represents the hidden state matrix output by the (k-1)th hidden layer. Represents the normalized adjacency matrix. This represents the weight matrix of the k-th hidden layer. This represents the bias matrix of the k-th hidden layer. This represents the adjustment coefficient. This represents a diagonal matrix consisting of the diagonal elements of a normalized adjacency matrix. express Activation function Degree matrix, This represents the element in the i-th row and i-th column of the degree matrix, where i represents... index, This represents the element in the i-th row and j-th column of the adjacency matrix.

[0097] Specifically, the hidden layer extracts useful features from the input data through multiple processing and transformations. These features effectively describe the quality factors of the welding scheme, such as the spatial distribution of weld points and the rationality of the welding sequence. The hidden layer transforms the input data into intermediate representations, which are crucial for predicting the score. These intermediate representations contain the deep relationships between weld points and the global network structure.

[0098] In one embodiment of the present invention, the target welding scheme is obtained, including:

[0099] Step 1: Initialize the population. The population consists of R individuals that conform to preset rules, and each individual corresponds to a random welding scheme.

[0100] Step 2, the objective function is constructed based on the topology prediction model, specifically including:

[0101] Based on the welding scheme of an individual, a corresponding topology network is constructed, and the feature matrix of the topology network is obtained. The feature matrix is ​​input into the topology prediction model to obtain the predicted welding score of the topology prediction model. The predicted welding score is used as the objective function value of the corresponding individual.

[0102] Step 3: Obtain the average value of the objective function for each individual, retain individuals whose objective function value is higher than the average value, and use the remaining individuals as the parent generation for iteration;

[0103] Step 4: Repeat step 3 a preset number of times to obtain the welding scheme corresponding to the individual with the largest objective function value as the target welding scheme.

[0104] Specifically, the target welding scheme is generated through the following steps: Population initialization: Generate multiple random welding schemes, each scheme being an "individual". Objective function construction: Calculate the score for each welding scheme based on the topology prediction model; the score serves as the objective function value. High-quality individuals selection: Retain schemes with higher scores and use the remaining schemes for the next iteration. Iterative optimization: Repeat the above process until the welding scheme with the highest score is obtained, which is then used as the target welding scheme. Combining the topology prediction model with the optimization algorithm: The topology prediction model provides accurate scoring criteria for the welding schemes, while the optimization algorithm iteratively improves the schemes, achieving automated design through data-driven approaches and algorithm optimization. Intelligent welding scheme design:

[0105] It avoids the shortcomings of relying on human experience in traditional welding processes and provides a data-driven method for generating welding schemes, improving design efficiency and welding quality.

[0106] In one embodiment of the present invention, the remaining individuals are used as parents for iteration, including a first stage and a second stage:

[0107] The update formula for the first stage is as follows: ; ;

[0108] in, This represents the individual encoding vector in the (t+1)th iteration. This represents the individual encoding vector in the t-th iteration. This represents the first updated weight vector. This represents the first update adjustment coefficient. Indicates fitness. Represents the baseline fitness constant. This represents the descending rank of the objective function value of each individual in the t-th iteration. Let represent the diagonal vector consisting of the diagonal elements of the adjacency matrix corresponding to the individual encoding in the t-th iteration. This indicates the t-th iteration. This indicates the (t+1)th iteration;

[0109] The update formula for the second stage is as follows: ;

[0110] in, This represents the second update adjustment coefficient. This represents a random perturbation vector. The table shows the individual encoding vector of the individual with the largest objective function value among all individuals in the t-th iteration. It represents the absolute value.

[0111] Specifically, the individual iteration includes two stages of updates: a first stage and a second stage. By simulating different optimization strategies, the quality of welding schemes is gradually improved. The first stage update adjusts the parent individuals based on update rules using weights and fitness. It emphasizes the dominant role of high-quality welding schemes in the update, ensuring the schemes optimize towards a better direction. The second stage update introduces random perturbations to increase the diversity of schemes. It avoids optimization getting trapped in local optima and further improves the schemes through global search. The first stage update sorts each welding scheme according to its fitness (i.e., the score of the topology prediction model). A weight adjustment strategy is used to prioritize retaining individuals with higher fitness. High-scoring schemes are more likely to influence the next generation, ensuring the optimization process evolves towards the optimal direction. The combination of weights and fitness in the update rules allows for dynamic adjustment of the individual's weld point position and welding sequence. Fast convergence: The first stage focuses on utilizing information from existing high-quality schemes, allowing the optimization process to approach the target scheme more quickly. Building on the first stage, random perturbations are introduced into the individual schemes. Through global search, a wider solution space is explored, uncovering potential high-quality schemes. Randomness avoids the optimization process getting trapped in local optima. By combining the characteristics of the optimal individual with random adjustments, the solution is further improved. Random perturbation ensures the optimization algorithm's adaptability to different initial populations and welding scenarios. Global search can discover potential better welding solutions, improving the overall quality of the final solution. Through a two-stage update strategy, an efficient and intelligent welding solution optimization method is provided. The innovation of this two-stage update lies in balancing local optimization and global search, making the optimization results more efficient, higher quality, and more robust.

[0112] A computing device, comprising:

[0113] One or more processors;

[0114] A storage device for storing one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the system.

[0115] A computer-readable storage medium storing a program that, when executed by a processor, implements the system described above.

[0116] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. An execution system for an ultrasonic welding machine, characterized in that, include: The data generation module is used to initialize the weld point scheme of the test base material based on preset rules; The solder joint scheme includes: the position coordinates of M solder joints and the corresponding soldering order; The solder joint welding module is used to weld the test base material based on the solder joint scheme to obtain the test piece; The topology building module is used to generate topology networks based on solder joint schemes. The training sample module is used to obtain N test pieces and N corresponding topology networks through the data generation module, weld point welding module and topology construction module, and to obtain the corresponding welding scores of the N test pieces based on physical testing; The physical tests include weld strength testing, weld integrity testing, and appearance quality testing; the weld scores for N test pieces are obtained based on the physical tests, including: Obtain the corresponding test data for welding strength, welding integrity, and appearance quality; Calculate the ratio of the weld strength to the maximum weld strength to obtain the weld strength score; Calculate the ratio of the non-welded intact area to the average non-welded intact area to obtain the weld integrity score; The appearance quality score is determined based on the weld integrity test and the appearance quality test. The weld strength score, weld integrity score, and appearance quality score are combined by weighted multiplication to obtain the first index value: Calculate the weighted value of appearance quality based on the appearance quality score; The normalized weld strength is added to the non-welded complete area to obtain the comprehensive index value of mechanical properties. Based on the comprehensive index value of mechanical properties, the weighted value of mechanical properties is calculated. Calculate the ratio of the weighted value of appearance quality to the weighted value of mechanical properties, and then calculate its arctangent value based on the ratio: After adding the first index value to the arctangent value, normalization is performed to obtain the welding score of the test piece. The topology prediction module is used to construct a topology prediction model based on N topology networks and their corresponding N welding scores, including: A topology prediction model is constructed based on N topology networks and corresponding N welding scores; For each topology network, output an adjacency matrix. The element in the i-th row and j-th column of the adjacency matrix represents the connection relationship between the i-th node and the j-th node in the topology network. If there is an edge between the i-th node and the j-th node, the element value of the corresponding element is 1; otherwise, the element value of the corresponding element is 0. For each topology network, output a diagonal sparse matrix. The element in the m-th row and m-th column of the diagonal sparse matrix represents the eigenvector of the m-th node. All other elements of the diagonal sparse matrix are 0. The elements in the same row and column of the adjacency matrix and diagonal sparse matrix of each topology network are merged to obtain N feature matrices of size M×M. The feature matrix is ​​used as the training sample, and the corresponding welding score is used as the sample label to train the topology prediction model, which includes a hidden layer and a classifier. Hidden layers are used to perform nonlinear transformations on the input feature matrix to obtain hidden states; A classifier is used as input to the hidden state and outputs a predicted welding score. The mean error variance of the predicted welding score and the corresponding sample label is used as the loss function to update the weight matrix and bias matrix of the hidden layer of the topology prediction model through backpropagation. The planning module is used to obtain the target welding scheme based on the topology prediction model and the optimization algorithm.

2. The execution system of an ultrasonic welding machine according to claim 1, characterized in that, The solder joint scheme is initialized based on preset rules, including: Preset rules are used to constrain random solder joints in the solder joint scheme, specifically including: The first and second weld points in the welding scheme are located at both ends of the weld on the test base material; The spacing between solder joints that are not the first solder joint and the second solder joint in the solder joint scheme is less than a preset spacing threshold. The unit weld density of the weld seam in the test base material is within the preset threshold range.

3. The execution system of an ultrasonic welding machine according to claim 1, characterized in that, Generate a topology network based on the solder joint scheme, including: Map each of the M solder joints to a topology node of the topology network; Based on the welding order of the weld points, construct topological edges between topological nodes of adjacent welding orders; The feature vector of the corresponding topological node is constructed based on the position coordinates of the solder joint.

4. The execution system of an ultrasonic welding machine according to claim 3, characterized in that, The target welding scheme is obtained, including: Step 1: Initialize the population. The population consists of R individuals that conform to preset rules, and each individual corresponds to a random welding scheme. Step 2, the objective function is constructed based on the topology prediction model, specifically including: Based on the welding scheme of an individual, a corresponding topology network is constructed, and the feature matrix of the topology network is obtained. The feature matrix is ​​input into the topology prediction model to obtain the predicted welding score of the topology prediction model. The predicted welding score is used as the objective function value of the corresponding individual. Step 3: Obtain the average value of the objective function for each individual, retain individuals whose objective function value is higher than the average value, and use the remaining individuals as the parent generation for iteration; Step 4: Repeat step 3 a preset number of times to obtain the welding scheme corresponding to the individual with the largest objective function value as the target welding scheme.

5. The execution system of an ultrasonic welding machine according to claim 4, characterized in that, The remaining individuals are used as parents for iteration, including the first and second phases: The first phase of updates is as follows: Based on the individual encoding vector of the current iteration, multiply it by the first update weight vector to obtain the basic inheritance term; divide the baseline fitness constant by the individual's ranking position in the current iteration, and perform a nonlinear transformation on the fitness using a sine function to obtain the fitness adjustment term; Extract the diagonal elements of the adjacency matrix corresponding to the current individual code, form a diagonal vector, and obtain the adjacency matrix perturbation term; Calculate the individual encoding vector P(t+1) for the (t+1)th iteration based on the basic inheritance term, fitness adjustment term, and adjacency matrix perturbation term; The second phase of updates is as follows: Based on the individual encoding vector of the current iteration, multiply it by the random perturbation vector to obtain the random perturbation term; Identify the individual with the largest objective function value in the current iteration, extract its encoding vector, calculate the absolute value of the encoding vector, and take the natural logarithm of the absolute value of the encoding vector to obtain the globally optimal guiding term: The random perturbation term is added to the globally optimal guiding term and processed by the ReLU activation function to obtain the individual encoding vector P(t+1) for the (t+1)th iteration.

6. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the system as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a program that, when executed by a processor, implements the system as described in any one of claims 1 to 5.

Images