Intelligent stamping control method and system

Abstract

The present invention discloses an intelligent stamping control method and system, which relates to the field of stamping and artificial intelligence technology. The method includes: real-time monitoring of each link of the stamping process to obtain stamping data, the stamping data including pressure, speed, temperature, workpiece thickness, workpiece hardness, stamping stroke, sound, vibration and material quality; simulating the stamping data through a fusion simulation model to obtain the state of the current stamping workpiece, the fusion simulation model includes a boundary element analysis model and a discrete element analysis model, and the state of the current stamping workpiece includes the shape, size and surface quality of the workpiece; predicting the state of the current stamping workpiece through an online multimodal recognition model to obtain the abnormal probability of the stamping workpiece; automatically adjusting the stamping parameters according to the abnormal probability of the stamping workpiece, the stamping parameters including pressure and speed. The present invention performs simulation and recognition according to the stamping data, thereby improving the accuracy and efficiency of stamping detection.

Description

Technical Field

[0001] The present invention relates to the field of stamping and artificial intelligence technology, and in particular to an intelligent stamping control method and system. Background Art

[0002] Stamping is to apply external force to materials such as plates, strips, pipes and profiles through presses and dies to cause plastic deformation or separation, so as to obtain workpieces (stamping parts) of the required shape and size. The stamping detection system plays a very important role in the manufacturing process of stamping parts. First of all, the detection system can conduct a comprehensive inspection of the size, shape, surface quality, etc. of the stamping parts to ensure that the quality of the stamping parts meets the design requirements and production standards. Secondly, the detection system can monitor and feedback the process parameters in the stamping process in real time, helping operators to adjust the process parameters in time and improve production efficiency and product quality. In addition, the detection system can also quickly diagnose and locate faults and abnormalities in the production process, discover and solve problems in time, and avoid production line downtime and production losses.

[0003] The existing stamping inspection systems mainly include visual inspection, contact inspection, non-contact inspection and automated inspection methods. Among them, visual inspection is the most commonly used inspection method. The stamping parts are observed through human eyes or magnifying glasses to check whether their surface quality, dimensional accuracy, etc. meet the requirements. However, it is greatly affected by human factors and the accuracy is not high. Contact inspection refers to the inspection of stamping parts by touch, feel, etc. However, contact inspection requires manual operation, is greatly affected by human factors, and it is difficult to find some minor defects. Non-contact inspection refers to the inspection of stamping parts through various sensors, measuring instruments and other equipment without direct contact with the surface of the stamping parts. However, there is a lack of automatic processing and analysis of various sensor data, and the detection accuracy and efficiency need to be further improved.

[0004] Therefore, it is necessary to establish a more accurate and effective intelligent stamping control method and system. Summary of the invention

[0005] In view of the above problems, the present invention is proposed to provide an intelligent stamping control method and system that overcomes all or at least part of the above problems.

[0006] According to one aspect of the present invention, there is provided an intelligent stamping control method, comprising:

[0007] Real-time monitoring of each step of the stamping process to obtain stamping data, wherein the stamping data includes at least one of the following: pressure, speed, temperature, workpiece thickness, workpiece hardness, stamping stroke, sound, vibration and material quality;

[0008] The stamping data is simulated by a fusion simulation model to obtain the current state of the stamping workpiece, wherein the fusion simulation model includes a boundary element analysis model and a discrete element analysis model, and the current state of the stamping workpiece includes the shape, size and surface quality of the workpiece;

[0009] Predicting the state of the current stamping workpiece by an online multimodal recognition model to obtain an abnormal probability of the stamping workpiece;

[0010] The stamping parameters are automatically adjusted according to the abnormal probability of the stamping workpiece, and the stamping parameters include pressure and speed.

[0011] Furthermore, the fusion simulation model adopts a bidirectional coupling method to achieve real-time interaction and data sharing between the boundary element analysis model and the discrete element analysis model;

[0012] The bidirectional coupling method is specifically as follows:

[0013] The boundary element analysis model and the discrete element analysis model are respectively initialized, and the state of the current stamping workpiece is input into the boundary element analysis model and the discrete element analysis model. During the calculation process, real-time data exchange and information sharing are performed between the models to achieve two-way coupling.

[0014] Furthermore, the boundary element analysis model introduces high-order shape functions and boundary integral equations, wherein the high-order shape functions are used to describe the geometric shape and displacement field of the workpiece boundary, and the boundary integral equations are used to capture the influence of the workpiece shape and surface quality on the simulation results.

[0015] Furthermore, the calculation formula of the boundary integral equation is:

[0016]

[0017] Where Γ is the boundary of the workpiece, U(x) is the displacement field function at position x, and T ij (x) is the stress tensor at position x, U j (x′) is the displacement field function at position x′, U i (x) is the gradient of the displacement field function at position x, T jk (x) is the gradient of the stress tensor at position x, U k (x′) is the gradient of the displacement field function at position x′, n j (x′) is the external normal vector at position x′, F i (x) is the external force vector at position x, and dΓ(x′) is the boundary element area element at position x′.

[0018] Furthermore, the calculation formula of the high-order shape function is:

[0019]

[0020] Among them, N I (x) is the value of the I-th higher-order shape function at position x, φ i (x) is the value of the ith basis function at position x, n is the number of basis functions, I is the index of the higher-order shape function, which is used to distinguish different shape functions, and x is the position vector, indicating the location of the shape function. is the preset coefficient of the adaptive algorithm.

[0021] Furthermore, the online multimodal recognition model optimizes recognition performance by introducing a reward matrix, wherein the reward matrix calculates reward values ​​according to feature weights of different modalities and target categories, and uses the reward values ​​as supervisory information for training the online multimodal recognition model.

[0022] Furthermore, the calculation formula of the reward matrix is:

[0023] R = w softmax(Y)

[0024] Among them, R is the reward matrix, w is the feature weight vector, Y is the probability distribution vector of the target category, and softmax(.) is the activation function.

[0025] Furthermore, the reward matrix further calculates the reward value according to the feature weights of different modalities and the target category, including:

[0026] The reward matrix determines the reward value by comparing the similarities between feature vectors of different modalities;

[0027] The calculation formula for the similarity between the feature vectors is:

[0028]

[0029] Where sim(x,y) is the similarity between feature vectors x and y, x·y is the dot product of x and y, ||x|| and ||y|| are the norms of x and y respectively, α is the learning rate parameter, and N is the number of feature vectors.

[0030] Furthermore, the online multimodal recognition model optimizes recognition performance by introducing a reward matrix, further comprising:

[0031] The reward matrix is ​​iteratively updated to optimize the performance of the online multimodal recognition model. The iterative update formula of the reward matrix is:

[0032] R t+1 =R t +α·(R t -Rt-1 )

[0033] Among them, R t+1 is the reward matrix for the t+1th iteration, R t and R t-1 are the reward matrices for the t-th and t-1-th iterations respectively, and α is the learning rate parameter.

[0034] According to another aspect of the present invention, there is provided an intelligent stamping control system, comprising:

[0035] A data acquisition module, used for real-time monitoring of each link of the stamping process to obtain stamping data, wherein the stamping data includes at least one of the following: pressure, speed, temperature, workpiece thickness, workpiece hardness, stamping stroke, sound, vibration and material quality;

[0036] A simulation module is used to simulate the stamping data through a fusion simulation model to obtain the current state of the stamping workpiece, wherein the fusion simulation model includes a boundary element analysis model and a discrete element analysis model, and the current state of the stamping workpiece includes the shape, size and surface quality of the workpiece;

[0037] An abnormality recognition module, used to predict the state of the current stamping workpiece through an online multimodal recognition model to obtain an abnormality probability of the stamping workpiece;

[0038] The automatic adjustment module is used to automatically adjust the stamping parameters according to the abnormal probability of the stamping workpiece, and the stamping parameters include pressure and speed.

[0039] According to the scheme provided by the present invention, each link of the stamping process is monitored in real time to obtain stamping data, and the stamping data includes at least one of the following: pressure, speed, temperature, workpiece thickness, workpiece hardness, stamping stroke, sound, vibration and material quality; the stamping data is simulated by a fusion simulation model to obtain the state of the current stamping workpiece, the fusion simulation model includes a boundary element analysis model and a discrete element analysis model, and the state of the current stamping workpiece includes the shape, size and surface quality of the workpiece; the state of the current stamping workpiece is predicted by an online multimodal recognition model to obtain the abnormal probability of the stamping workpiece; the stamping parameters are automatically adjusted according to the abnormal probability of the stamping workpiece, and the stamping parameters include pressure and speed. The present invention performs simulation and recognition according to the stamping data, thereby improving the accuracy and efficiency of stamping detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0041] Figure 1 A schematic diagram showing a flow chart of an intelligent stamping control method according to an embodiment of the present invention;

[0042] Figure 2 A schematic diagram showing the structure of an intelligent stamping control system according to an embodiment of the present invention DETAILED DESCRIPTION

[0043] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0044] Figure 1 The flowchart of the intelligent stamping control method according to the embodiment of the present invention is shown. Specifically, the method comprises the following steps:

[0045] Step S101, real-time monitoring of each link of the stamping process to obtain stamping data, wherein the stamping data includes at least one of the following: pressure, speed, temperature, workpiece thickness, workpiece hardness, stamping stroke, sound, vibration and material quality.

[0046] In this embodiment, various aspects of the stamping process are monitored in real time, and a variety of stamping data can be obtained. Among them, pressure is one of the key factors affecting the deformation and forming of the workpiece. By monitoring the pressure data in real time, it is possible to analyze whether the pressure is within a suitable range and whether the pressure is evenly distributed, thereby ensuring the forming quality and production efficiency of the workpiece. By monitoring the speed data in real time, the stamping speed can be controlled within a suitable range to avoid problems such as workpiece cracking and wrinkling caused by too fast or too slow speed. Too high or too low temperature may affect the forming quality of the workpiece. By monitoring the temperature data in real time, the temperature can be controlled within a suitable range to avoid problems such as deformation and cracking of the workpiece caused by abnormal temperature. By monitoring the workpiece thickness data in real time, the thickness change of the workpiece during the stamping process can be analyzed, and problems such as uneven thickness can be discovered and solved in time. By monitoring the hardness data in real time, the hardness can be controlled within a suitable range to ensure the forming quality and durability of the workpiece. The stamping stroke refers to the distance the punch moves during the stamping process. By monitoring the stroke data in real time, it is possible to analyze whether the stroke is within a suitable range and whether the stroke is stable, thereby ensuring the forming quality and production efficiency of the workpiece. The sound during the stamping process can reflect the state of the workpiece and the forming effect. By real-time monitoring of the sound changes of the workpiece during the stamping process, the problems represented by abnormal sounds can be discovered and solved in time. By real-time monitoring of vibration data, the operating state of the equipment and the forming effect of the workpiece can be analyzed, and the problems caused by vibration can be discovered and solved in time. For example, various sensors such as pressure sensors, speed sensors, temperature sensors, etc. are installed on the stamping equipment to monitor various data in the stamping process in real time, and then the data collected by the sensors (such as pressure distribution, speed changes, temperature fluctuations, etc.) are transmitted to the computer for processing and analysis.

[0047] Step S102, simulating the stamping data through a fusion simulation model to obtain the current state of the stamped workpiece, the fusion simulation model includes a boundary element analysis model and a discrete element analysis model, and the current state of the stamped workpiece includes the workpiece shape, size and surface quality.

[0048] In this embodiment, by simulating the stamping data by fusing the simulation model, the state of the current stamping workpiece can be obtained, including information such as the shape, size and surface quality of the workpiece. Among them, the boundary element analysis model (Boundary Element Analysis, BEM) analyzes the deformation and stress distribution of the workpiece under pressure in the stamping simulation. By inputting stamping data (such as pressure, speed, temperature, etc.) to the boundary element analysis model, the deformation process of the workpiece can be simulated, and the final shape and size of the workpiece can be predicted. The discrete element analysis model (Discrete Element Method, DEM) is used to analyze the numerical value of the behavior of non-continuous media. In the stamping simulation, the discrete element analysis model simulates the movement and deformation of particles or grains inside the workpiece. By inputting stamping data to the discrete element analysis model, the movement and interaction of particles or grains inside the workpiece can be simulated, and the internal structure and deformation behavior of the workpiece can be predicted. By fusing the boundary element analysis and discrete element analysis models, more accurate stamping simulation results can be obtained. The fusion of these two models can achieve complementary advantages, making the simulation results closer to the actual state of the stamping workpiece.

[0049] Specifically, the stamping data is input into the fusion simulation model, which includes two parts: boundary element analysis and discrete element analysis. Among them, the boundary element analysis part simulates the deformation process of the workpiece according to the input stamping data and predicts the final shape and size of the workpiece. The discrete element analysis part simulates the movement and interaction of particles or grains inside the workpiece and predicts the internal structure and deformation behavior of the workpiece. The results of boundary element analysis and discrete element analysis are integrated to obtain more accurate stamping simulation results, realize real-time monitoring and simulation of the stamping process, timely discover and solve potential problems, and improve the forming quality and production efficiency of the workpiece.

[0050] Furthermore, the fusion simulation model adopts a bidirectional coupling method to achieve real-time interaction and data sharing between the boundary element analysis model and the discrete element analysis model;

[0051] The bidirectional coupling method is specifically as follows:

[0052] The boundary element analysis model and the discrete element analysis model are respectively initialized, and the state of the current stamping workpiece is input into the boundary element analysis model and the discrete element analysis model. During the calculation process, real-time data exchange and information sharing are performed between the models to achieve two-way coupling.

[0053] In this embodiment, a bidirectional coupling method is used to realize real-time interaction and data sharing between the boundary element analysis model and the discrete element analysis model, so that more accurate and efficient stamping simulation can be achieved.

[0054] Specifically, a boundary element analysis model and a discrete element analysis model are established to describe the deformation behavior of the workpiece during the stamping process. The boundary element analysis model is used to analyze the macroscopic deformation and stress distribution of the workpiece, while the discrete element analysis model is used to simulate the movement and interaction of particles or grains inside the workpiece. After the model is established, the relevant parameters and data, such as the material properties, initial state, stamping conditions, etc. of the workpiece, are initialized for subsequent simulation calculations. A two-way coupling method is used to allow the boundary element analysis model and the discrete element analysis model to interact and share data in real time. That is, the two models will perform calculations at the same time and transfer data and information to each other. In each time step, the boundary element analysis model transfers the calculated stress distribution to the discrete element analysis model, and the discrete element analysis model updates the motion state of the particles or grains according to these stress distributions. At the same time, the discrete element analysis model also feeds back its own motion state and interaction results to the boundary element analysis model for correcting the macroscopic deformation analysis. Through repeated iterative calculations, the two models will gradually reach a dynamic equilibrium state, and the simulation results obtained at this time will be closer to the state of the actual stamping workpiece. In the iterative calculation process, it is necessary to set appropriate convergence conditions. When the difference between the two models is less than a certain threshold, the simulation results can be considered to have converged. Through real-time interaction and data sharing between the boundary element analysis model and the discrete element analysis model, the accuracy and efficiency of stamping simulation are improved, providing strong support for actual production.

[0055] Furthermore, the boundary element analysis model introduces high-order shape functions and boundary integral equations, wherein the high-order shape functions are used to describe the geometric shape and displacement field of the workpiece boundary, and the boundary integral equations are used to capture the influence of the workpiece shape and surface quality on the simulation results.

[0056] In this embodiment, the introduction of high-order shape functions and boundary integral equations into the boundary element analysis model can improve the accuracy of the description of the workpiece boundary geometry and displacement field, and better capture the influence of the workpiece shape and surface quality on the simulation results. Among them, the high-order shape function, also known as the high-order basis function or the high-order interpolation function, is a mathematical tool used to describe the properties and behaviors of complex functions, and can provide a more accurate function approximation. In the stamping simulation, it can provide a more accurate description of the displacement field. In the boundary element analysis model, the high-order shape function is used to represent the node displacement on the workpiece boundary. Among them, the boundary integral equation is used to describe the relationship between the physical quantity on the boundary and the physical quantity in the region. In engineering and scientific computing, the boundary integral equation is widely used to solve the numerical solution of partial differential equations. In the stamping simulation, it describes the relationship between the stress distribution on the workpiece boundary and the external load. In the boundary element analysis model, by discretizing the workpiece boundary, the stress distribution is expressed as a function of the node displacement on the boundary. After the boundary integral equation is established, it is solved by numerical methods. Commonly used solution methods include finite element method, finite difference method, etc. By solving the boundary integral equation, the node displacement and stress distribution on the workpiece boundary can be obtained.

[0057] Furthermore, the calculation formula of the boundary integral equation is:

[0058]

[0059] Where Γ is the boundary of the workpiece, U(x) is the displacement field function at position x, and T ij (x) is the stress tensor at position x, U j (x′) is the displacement field function at position x′, U i (x) is the gradient of the displacement field function at position x, T jk (x) is the gradient of the stress tensor at position x, U k (x′) is the gradient of the displacement field function at position x′, n j (x′) is the external normal vector at position x′, F i (x) is the external force vector at position x, and dΓ(x′) is the boundary element area element at position x′.

[0060] In this embodiment, the boundary of the workpiece refers to the outer surface of the workpiece or the surface in contact with the outside world. In the stamping simulation, it is directly related to the deformation and stress distribution of the workpiece. The displacement field function describes the displacement of a point inside or on the surface of an object, that is, how much the object moves at a certain position. The displacement field function describes the deformation of the workpiece under the action of an external force. The stress tensor describes the distribution of the force at a point inside or on the surface of an object. The stress tensor is a mathematical tool used to describe the complexity and directionality of this distribution. The stress tensor can be a second-order symmetric tensor, which represents the relationship between the three principal stresses and the three principal shear forces. It can be expressed as the direction, magnitude and change trend of the stress. By calculating the stress tensor, the stress distribution inside the workpiece can be obtained, and then the deformation and strength of the workpiece can be analyzed. The gradient of the displacement field function describes the rate of change of the function value with the spatial position, which is used to describe the direction and change trend of the displacement. The gradient of the stress tensor is similar to the gradient of the displacement field function, which describes the rate of change of the stress distribution with the spatial position. The external normal vector is a vector tangent to the boundary of the workpiece (pointing to the outside), which is used to describe the direction and shape of the boundary. The external force vector refers to the external force acting on the workpiece, which is used to describe the pressure and impact force of the stamping equipment. Boundary element area element is used to solve partial differential equations. Boundary element area element refers to a small area or unit divided on the boundary of the workpiece. The boundary integral equation can be used to describe the deformation and stress distribution of the workpiece, and how these distributions are affected by external forces and boundary conditions. By solving the boundary integral equation, more accurate and reliable stamping simulation results can be obtained.

[0061] Furthermore, the calculation formula of the high-order shape function is:

[0062]

[0063] Among them, N I (x) is the value of the I-th higher-order shape function at position x, φ i (x) is the value of the ith basis function at position x, n is the number of basis functions, I is the index of the higher-order shape function, which is used to distinguish different shape functions, and x is the position vector, indicating the location of the shape function. is the preset coefficient of the adaptive algorithm.

[0064] In this embodiment, high-order shape functions are used to describe the boundary and displacement field of the workpiece. The value of the I-th high-order shape function at position x represents the value of the I-th high-order shape function at this position, which is used to determine the displacement and stress state of the workpiece at this position. The value of the i-th basis function at position x is used to describe the displacement field and stress distribution of the workpiece. The value of the i-th basis function at position x represents the value of the i-th basis function at this position. n is the number of basis functions, which determines the accuracy and complexity of describing the state of the workpiece. I is the index of the high-order shape function, which is used to distinguish different high-order shape functions. By using different high-order shape functions, the complex boundary and displacement field of the workpiece can be better fitted. The position vector is used to describe the position of a point in space. In stamping simulation, the position vector is used to determine the position of a point inside or on the surface of the workpiece. The preset coefficients of the adaptive algorithm are used to optimize the accuracy and convergence speed of the simulation.

[0065] Step S103, predicting the state of the current stamping workpiece by using an online multimodal recognition model to obtain the abnormality probability of the stamping workpiece.

[0066] By predicting the current state of the stamping workpiece through the online multimodal recognition model, the abnormal probability of the stamping workpiece can be obtained, so that potential problems can be discovered and handled in time.

[0067] Furthermore, the online multimodal recognition model optimizes recognition performance by introducing a reward matrix, wherein the reward matrix calculates reward values ​​according to feature weights of different modalities and target categories, and uses the reward values ​​as supervisory information for training the online multimodal recognition model.

[0068] In this embodiment, the reward matrix is ​​used to guide how the model learns and optimizes, and the reward values ​​are calculated according to the feature weights and target categories of different modes. These reward values ​​are used as supervisory information for training online multimodal recognition models. Specifically, the weight calculation of the features of each mode can be achieved by various feature selection methods, such as statistical-based methods, model-based methods, integrated learning, etc. The target category of each mode is clarified. For example, for the state prediction of stamping workpieces, the target category may be normal, abnormal, damaged, etc. The reward value is calculated according to the feature weight and the target category. The reward value reflects the prediction accuracy of the model for the target category under a certain mode. When calculating the reward value, factors such as the contribution of the mode to the overall prediction and the size of the classification error can be considered. The reward values ​​of each mode are combined into a reward matrix, which not only contains the reward values ​​of each mode, but also reflects the relationship between different modes. The reward matrix is ​​used as supervisory information to train the online multimodal recognition model. During the training process, the model adjusts its parameters and structure according to the reward values ​​in the reward matrix to improve the prediction accuracy and stability under various modes. By introducing the reward matrix, the online multimodal recognition model is able to trade-off and optimize between different modalities and target categories, thereby improving its prediction accuracy and reliability of the stamping workpiece state.

[0069] Furthermore, the calculation formula of the reward matrix is:

[0070] R = w softmax(Y)

[0071] Among them, R is the reward matrix, w is the feature weight vector, Y is the probability distribution vector of the target category, and softmax(.) is the activation function.

[0072] In this embodiment, the online multimodal recognition model can better understand and process multimodal data and extract valuable information from it by combining the reward matrix, feature weight vector and probability distribution vector of the target category. Among them, the activation function converts the output of the model into a probability distribution form so that the sum of the probabilities of each target category is 1. By calculating the probability of each target category, the confidence of the model for different categories can be measured, thereby providing a basis for the calculation of the reward matrix.

[0073] Furthermore, the reward matrix further calculates the reward value according to the feature weights of different modalities and the target category, including:

[0074] The reward matrix determines the reward value by comparing the similarities between feature vectors of different modalities;

[0075] The calculation formula for the similarity between the feature vectors is:

[0076]

[0077] Where sim(x,y) is the similarity between feature vectors x and y, x·y is the dot product of x and y, ||x|| and ||y|| are the norms of x and y respectively, α is the learning rate parameter, and N is the number of feature vectors.

[0078] In this embodiment, the reward value in the reward matrix is ​​determined by comparing the similarity between the feature vectors of different modes. The similarity is calculated using the cosine similarity method, or the Euclidean distance method. In the similarity calculation process, the learning rate parameter is introduced. Used to control the step size of model parameter updates to avoid overfitting or underfitting.

[0079] Furthermore, the iterative update formula of the reward matrix is:

[0080] R t+1 =R t +α·(R t -R t-1 )

[0081] Among them, R t+1 is the reward matrix for the t+1th iteration, R t and R t-1 are the reward matrices for the t-th and t-1-th iterations respectively, and α is the learning rate parameter.

[0082] In this embodiment, by comparing the reward matrices of the t-th and t-1 iterations, the changes in the reward matrix can be detected to better balance the stability and adaptability of the model. Among them, the reward matrix of the t+1 iteration is the reward matrix currently being calculated, which is used to guide the learning and optimization of the model in the t+1 iteration. The reward matrix of the t iteration is the reward matrix of the previous iteration. By comparing it with the reward matrix of the t-1 iteration, the dynamic changes of the reward matrix can be obtained. The learning rate parameter can control the update speed of the reward matrix. A larger learning rate may cause the reward matrix to change rapidly, while a smaller learning rate may cause the reward matrix to adjust slowly.

[0083] Step S104: automatically adjusting stamping parameters according to the abnormal probability of the stamping workpiece, wherein the stamping parameters include pressure and speed.

[0084] In this embodiment, the stamping parameters are automatically adjusted according to the abnormal probability of the stamped workpiece in order to optimize the stamping process and ensure the quality of the workpiece and the stability of production. For example, when the abnormal probability of the stamped workpiece increases, the system automatically reduces the stamping pressure to prevent the workpiece from being damaged or defective due to excessive pressure. When the abnormal probability of the workpiece increases, the system automatically reduces the stamping speed to give the workpiece more forming time and reduce problems caused by rapid stamping. When the abnormal probability is low, the system may increase the stamping speed to improve production efficiency. In addition to adjusting the pressure and speed separately, a comprehensive adjustment strategy can also be formulated based on the abnormal probability. For example, when the abnormal probability increases, the system can reduce the pressure and speed at the same time to handle the workpiece more carefully. In practical applications, the appropriate adjustment strategy can be selected to optimize the stamping process according to specific needs and equipment capabilities.

[0085] According to the scheme provided by the present invention, each link of the stamping process is monitored in real time to obtain stamping data, and the stamping data includes at least one of the following: pressure, speed, temperature, workpiece thickness, workpiece hardness, stamping stroke, sound, vibration and material quality; the stamping data is simulated by a fusion simulation model to obtain the state of the current stamping workpiece, the fusion simulation model includes a boundary element analysis model and a discrete element analysis model, and the state of the current stamping workpiece includes the shape, size and surface quality of the workpiece; the state of the current stamping workpiece is predicted by an online multimodal recognition model to obtain the abnormal probability of the stamping workpiece; the stamping parameters are automatically adjusted according to the abnormal probability of the stamping workpiece, and the stamping parameters include pressure and speed. The present invention performs simulation and recognition according to the stamping data, thereby improving the accuracy and efficiency of stamping detection.

Claims

1. An intelligent stamping control method, It is characterized in that include: Real-time monitoring of each step of the stamping process to obtain stamping data, wherein the stamping data includes at least one of the following: pressure, speed, temperature, workpiece thickness, workpiece hardness, stamping stroke, sound, vibration and material quality; The stamping data is simulated by a fusion simulation model to obtain the current state of the stamping workpiece, wherein the fusion simulation model includes a boundary element analysis model and a discrete element analysis model, and the current state of the stamping workpiece includes the shape, size and surface quality of the workpiece; Predicting the state of the current stamping workpiece by an online multimodal recognition model to obtain an abnormal probability of the stamping workpiece; Automatically adjust stamping parameters according to the abnormal probability of the stamping workpiece, the stamping parameters including pressure and speed; The fusion simulation model adopts a bidirectional coupling method to realize real-time interaction and data sharing between the boundary element analysis model and the discrete element analysis model; The bidirectional coupling method is specifically as follows: Initializing the boundary element analysis model and the discrete element analysis model respectively, inputting the state of the current stamping workpiece into the boundary element analysis model and the discrete element analysis model, and performing real-time data exchange and information sharing between the models during the calculation process to achieve bidirectional coupling; The boundary element analysis model introduces high-order shape functions and boundary integral equations, wherein the high-order shape functions are used to describe the geometric shape and displacement field of the workpiece boundary, and the boundary integral equations are used to capture the influence of the workpiece shape and surface quality on the simulation results; The calculation formula of the boundary integral equation is: Where Γ is the boundary of the workpiece, U(x) is the displacement field function at position x, and T ij (x) is the stress tensor at position x, U j (x′) is the displacement field function at position x′, U i (x) is the gradient of the displacement field function at position x, T jk (x) is the gradient of the stress tensor at position x, U k (x′) is the gradient of the displacement field function at position x′, n j (x′) is the external normal vector at position x′, F i (x) is the external force vector at position x, dΓ(x′) is the boundary element area element at position x′; The online multimodal recognition model optimizes recognition performance by introducing a reward matrix, wherein the reward matrix calculates reward values ​​according to feature weights of different modalities and target categories, and uses the reward values ​​as supervision information for training the online multimodal recognition model; The calculation formula of the reward matrix is: R = w softmax(Y) Among them, R is the reward matrix, w is the feature weight vector, Y is the probability distribution vector of the target category, and softmax(.) is the activation function; The online multimodal recognition model further includes: The reward matrix is ​​iteratively updated to optimize the performance of the online multimodal recognition model. The iterative update formula of the reward matrix is: R t+1 =R t +α·(R t -R t-1 ) Among them, R t+1 is the reward matrix for the t+1th iteration, R t and R t-1 are the reward matrices for the t-th and t-1-th iterations respectively, and α is the learning rate parameter.

2. The intelligent stamping control method according to claim 1, It is characterized in that The calculation formula of the high-order shape function is: Among them, N I (x) is the value of the I-th higher-order shape function at position x, φ i (x) is the value of the ith basis function at position x, n is the number of basis functions, I is the index of the higher-order shape function, which is used to distinguish different shape functions, and x is the position vector, indicating the location of the shape function. is the preset coefficient of the adaptive algorithm.

3. The intelligent stamping control method according to claim 2, It is characterized in that The reward matrix further calculates the reward value according to the feature weights of different modalities and the target category: The reward matrix determines the reward value by comparing the similarities between feature vectors of different modalities; The calculation formula for the similarity between the feature vectors is: Where sim(x,y) is the similarity between feature vectors x and y, x·y is the dot product of x and y, ||x|| and ||y|| are the norms of x and y respectively, α is the learning rate parameter, and N is the number of feature vectors.

4. An intelligent stamping control system, the system is based on the intelligent stamping control method according to any one of claims 1 to 3, It is characterized in that include: A data acquisition module, used for real-time monitoring of each link of the stamping process to obtain stamping data, wherein the stamping data includes at least one of the following: pressure, speed, temperature, workpiece thickness, workpiece hardness, stamping stroke, sound, vibration and material quality; A simulation module is used to simulate the stamping data through a fusion simulation model to obtain the current state of the stamping workpiece, wherein the fusion simulation model includes a boundary element analysis model and a discrete element analysis model, and the current state of the stamping workpiece includes the shape, size and surface quality of the workpiece; An abnormality recognition module, used to predict the state of the current stamping workpiece through an online multimodal recognition model to obtain an abnormality probability of the stamping workpiece; The automatic adjustment module is used to automatically adjust stamping parameters according to the abnormal probability of the stamping workpiece, and the stamping parameters include pressure and speed.

Images