Intelligent shaping machining control method and system based on machine learning

Abstract

The application discloses an intelligent shaping machining control method based on machine learning, comprising the following steps: S1, obtaining a machine learning prediction model, wherein the machine learning prediction model is an MLP model; S2, placing a product to be machined into the machine learning prediction model trained offline, then predicting a machining value, and feeding back the relevant value to a mechanical arm pressure head to perform shaping machining on the product; S3, re-measuring product data after machining and performing judgment to obtain a machining result; and S4, performing online training on the machine learning prediction model according to a time interval set by a user. The application further discloses an intelligent shaping machining control system based on machine learning. Compared with the prior art, the application adopts a data processing and machine learning method to predict a machining value, and separately designs a model for each side edge of a product to be shaped to perform training and prediction.

Description

Technical Field

[0001] This invention relates to the field of intelligent workpiece shaping technology, and in particular to an intelligent shaping control method and system based on machine learning. Background Technology

[0002] Material shaping is an important part of material processing. Shaping is a process that utilizes the plasticity of metal to obtain the shape of a product in a time-saving, labor-saving, and material-saving manner. In current shaping systems, for original square materials, a robotic arm pressure head is used to press and shape the front, left, and right sides according to a specific stroke, so that the height values ​​of the three sides are within the standard range.

[0003] Currently, the traditional method for obtaining processing volume mainly relies on predicting processing values ​​using traditional empirical models.

[0004] An experience-based model refers to a predictive model for processing accuracy established by summarizing and analyzing the accumulated experience and knowledge from long-term engineering practice in the front line of processing and production. This model starts with existing data, forms corresponding tables based on experience, and establishes an expert experience database. The database mainly contains the initial measured height values ​​of the front, left, and right sides of materials processed in the past, as well as the corresponding processing amounts for each side. The processing values ​​for the first and second shaping are obtained by looking up the tables. These values ​​are all determined manually through trial and error. Each time new material is added to the processing table, the computer lookup program searches the manual experience database, finds information matching the material data, and then retrieves the corresponding processing value for processing. This approach requires manual trial shaping for each type of workpiece in advance to determine the appropriate processing amounts for different shapes, sizes, and materials. Furthermore, it requires dedicated engineers to be on-site around the production line equipment to make timely adjustments, undoubtedly wasting a significant amount of human resources and reducing material shaping efficiency.

[0005] In actual production practice, based on traditional shaping and processing prediction models (as shown in the attached figure) Figure 1 Cosmetic surgery has the following disadvantages:

[0006] 1) The processing is easily affected by noise interference, material elastic deformation and unknown hidden variables. It is difficult to take these factors into account based on the expert experience base, which makes the decision-making and processing process usually deviate significantly from the expectations.

[0007] 2) The current method of obtaining processing volume by looking up tables in the expert experience database is dependent on the sample size, making it difficult to measure all situations and making sampling difficult;

[0008] 3) The expert database is subject to recent adjustments and has limited space, which makes the expert experience database suitable for short-term cosmetic procedures and unsuitable for long-term use;

[0009] 4) Lack of adaptive capability. Obtaining the processing quantity of each side through the expert experience database requires looking up tables for each of the three sides separately. However, the processing of one side of the product will also affect the other sides. Therefore, the processing quantity obtained by this method will have errors and needs to be adjusted manually. It lacks adaptive capability. Summary of the Invention

[0010] The purpose of this invention is to provide an intelligent plastic surgery processing control method and system based on machine learning, which uses data processing and machine learning methods to predict processing values, and designs and trains and predicts a separate model for each side of the product to be plasticized.

[0011] To achieve the above objectives, this invention discloses an intelligent plastic surgery control method based on machine learning, comprising the following steps:

[0012] S1. Using all the previously collected processing data, a training set is established in combination with existing samples, and offline training is performed to obtain a machine learning prediction model, which is an MLP model.

[0013] S2. The product to be processed is put into the offline trained machine learning prediction model to predict the processing value, and the relevant value is fed back to the robotic arm press head to perform product shaping processing.

[0014] S3. After processing, remeasure the product data and make a judgment to obtain the processing result;

[0015] S4. Train the machine learning prediction model online according to the time interval set by the user.

[0016] As a further description of the above technical solution:

[0017] In step S1, when establishing the MLP model, two models are established for the front, left and right sides of the product to be processed. Each side of the product to be processed is set with a primary shaping MLP model and a secondary shaping MLP model. The primary shaping MLP model and the secondary shaping MLP model are respectively for the primary shaping and secondary shaping of one side of the product to be processed.

[0018] As a further description of the above technical solution:

[0019] In step S1, the process of establishing the MLP model includes the following steps:

[0020] S11. Use Tensorflow.keras to introduce the MLP model, stack the multi-layer structure, and complete the model construction.

[0021] S12. For each prediction process of each edge, a separate model is trained. The data used for training the MLP model is the raw processing data obtained from the actual production process. The raw processing data includes the product output information, the point values ​​and straightness before and after each processing, and the target specification range information of the product. The raw processing data is classified and extracted according to the edge and the number of processing to obtain six sets of data, which are the two processing data of the front, left, and right edges.

[0022] S13. After obtaining the processing data for the front, left, and right sides of the product, define the input parameters of the MLP model. Set the value of the first hidden layer to 32, the value of the second hidden layer to 64, and the value of the third hidden layer to 32. Finally, output a value as the predicted processing quantity. The input parameters are defined as follows: S14. The original dataset consisting of the two processing data of the front, left and right sides of the product is divided into a 2:8 ratio to obtain the test set and the training set. The MLP model is trained using the training set.

[0023] S15. Use the MAE metric to evaluate the MLP model. The MAE calculation formula is as follows: ,in, The total number of datasets in the test set; The target value corresponding to the test set; These are the predicted performance values ​​for the corresponding test set.

[0024] As a further description of the above technical solution:

[0025] In step S2, the product shaping process includes the following steps:

[0026] S21. After the product to be processed is placed into the shaping machine, it is first fixed on the shaping platform. The laser scanner scans the points on the front, left and right sides of the product to be processed to obtain the height value of all points and then transmits it to the industrial computer.

[0027] S22. After receiving the height value of the material point, the industrial computer obtains the specification range input by the user on the front-end interface, puts all these data into the offline trained machine learning prediction model, and waits for the shaping and processing results.

[0028] S23. After the data is fed into the machine learning prediction model, the model calculates and predicts the corresponding processing value, which is then submitted to the industrial computer.

[0029] S24. After receiving the predicted processing value, the industrial computer feeds it back to the controller, which then controls the robotic arm pressure head to perform a pressing operation to complete the shaping process.

[0030] On the other hand, the present invention also discloses an intelligent shaping and processing control system based on machine learning, comprising:

[0031] An industrial computer for running shaping control programs and machine learning algorithms, wherein the shaping control programs include instructions for performing any of the methods described above;

[0032] A scanner is used to measure the height of points on each side of a product to be processed, and then transmits the height values ​​of the points to an industrial computer to realize machine learning algorithms and shaping control.

[0033] A shaping platform, used to hold products to be processed;

[0034] The robotic arm pressure head is used to compress and deform the product being processed and hold it for a period of time.

[0035] The controller receives signals from the industrial computer, converts the signals, and transmits them to the robotic arm to control the robotic arm's pressure head.

[0036] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0037] 1. This invention replaces the traditional expert experience base model with a machine learning prediction model. Data analysis and processing of workpiece parameters are used to train a prediction model, which automatically calculates the stroke amount for each shaping process. The machine learning prediction model provides a predicted value for the processing amount, enabling prediction of workpiece processing amount and judgment of product qualification during the workpiece material shaping process. This system can learn and update online dynamically. It has strong adaptability, effectively reducing the impact of uncertainties such as elastic deformation, and has small errors. In actual production processing, it can improve processing efficiency and reduce labor costs. This technology can be used in flexible processing applications such as product processing and shaping.

[0038] 2. In this invention, the number of measurement points involved in the calculation changes from three points to multiple points. Traditional shaping schemes determine straightness specifications based on three key points of each edge, without performing any related operations or calculations on the remaining points. In this application, to obtain the processing status of each edge, the system sets 60 points. The machine learning-based scheme considers all points and aims to process all points to the "optimal value," that is, the midpoint of the specification range. This means that the height values ​​of all points and the midpoint of the specification range are included in the model, allowing the predicted processing amount to better consider the relationship between all points and the midpoint of the range.

[0039] 3. In this invention, data processing and machine learning methods are used to predict processing values, which effectively reduces the impact of abnormal data, resulting in better performance and stronger adaptability. This makes the processing process more intelligent, improves processing efficiency, and reduces labor costs. By feeding a portion of previous production data for a specific material into the model for training, a corresponding model is obtained. Compared to manual experience, this method requires less data.

[0040] 4. In this invention, considering the mutual influence between the three sides of the product, a separate model is designed for each side of the product to be shaped for training and prediction, thereby improving the accuracy. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0042] Legend

[0043] Figure 1 A flowchart of the shaping process based on traditional experience models in existing technologies.

[0044] Figure 2 This is a flowchart of step S2 in a machine learning-based intelligent shaping and processing control method.

[0045] Figure 3 This is a schematic diagram of one side of the product to be processed before and after shaping in an intelligent shaping and processing control method based on machine learning.

[0046] Figure 4 This is a schematic diagram of straightness calculation in an intelligent shaping and processing control method based on machine learning.

[0047] Figure 5 This is a schematic diagram of the MLP model in a machine learning-based intelligent shaping process control method. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0049] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0050] System composition:

[0051] The entire system mainly consists of three parts: the device layer, the data layer, and the algorithm layer.

[0052] 1. Equipment Layer: This layer includes

[0053] ① Scanner: Performs statistical analysis of material-related parameters, then transmits these parameters to an industrial computer to implement relevant algorithms and control. It features a built-in high-precision laser measurement subsystem that travels along the workpiece edge and measures each sampling point using a discrete sampling method.

[0054] ② Shaping platform: Used to place materials and perform shaping processing on them.

[0055] ③ Controller: After receiving the signal from the industrial computer, the controller converts the signal and transmits it to the robotic arm's pressure head to control the robotic arm.

[0056] ④ The robotic arm pressure head is a rubber block with a specific curved surface shape installed at the top of the robotic arm. During the shaping process, the robotic arm drives the shaping pressure head to act on the workpiece being shaped, compressing and deforming the workpiece for a period of time, thereby achieving the shaping effect.

[0057] ⑤ Industrial computer: This is a computer that runs shaping control programs and machine learning algorithms. It is connected to the scanner and robotic arm via a data cable to control the entire shaping system.

[0058] 2. Data Layer:

[0059] ① The system first needs to acquire relevant product data using a scanner and camera. To obtain the processing status of each side, the system sets up 60 points: 30 points corresponding to the front, 20 points to the left, and the remaining 20 points to the right. Then, a Cartesian coordinate system is constructed on the cross-section of a certain side of the material, initially with the plane where the material is placed as the x-axis and the vertical height as the y-axis. The processing status of that side is then displayed. For example... Figure 3 As shown.

[0060] The upper curve represents the material after shaping, the lower curve represents the material before shaping, C11-C95 represents the left side of the material, and the vertical axis represents the height relative to the initial horizontal plane. By combining the data from the scanner, the condition of the left side before and after shaping can be displayed.

[0061] The acquired data cannot be used directly for training. Unnecessary data such as "temperature" and "time" must be removed and standardized before it can be used in the model. The main method is to use z-score standardization to convert data of different magnitudes into a uniform metric for easy use.

[0062] ② Introduce the concept of straightness. After scanning, take the value of the point with the largest absolute value among the three middle points of each side as the straightness value. For example... Figure 4 As shown.

[0063] Straightness is an important parameter for subsequent testing and also a standard for judging whether a material meets the requirements.

[0064] ③ In addition, to facilitate model training and prediction, it is also necessary to set relevant properties of the material, including the type of material.

[0065] 3. Algorithm layer:

[0066] ① The introduction of machine learning prediction models is achieved through Tensorflow.keras. The main model is the Sequential model, which stacks multiple fully connected layer structures to complete the model building.

[0067] ② To obtain the data needed for model training, the processed data needs to be parameterized. A corresponding model needs to be built for each processing step of each edge. The trained model is then used to predict the processing volume.

[0068] Hardware and software platform:

[0069] Industrial computer 61045H, CPU: i7-9700, Memory: 16G, Hard drive: 256G SSD + 1T HDD; Laser rangefinder scanner: Keyence (Japan) CL-P015; Camera: Keyence (Japan) CA-H200MX / CA-H200CX; Developed using Python and combined with the TensorFlow deep learning framework, it can run on Windows and Linux.

[0070] Please see Figure 1-5 On the one hand, this invention discloses an intelligent plastic surgery control method based on machine learning, comprising the following steps:

[0071] S1. Using all the previously collected processing data, a training set is established in combination with existing samples, and offline training is performed to obtain a machine learning prediction model, which is an MLP model.

[0072] S2. The product to be processed is put into the offline trained machine learning prediction model to predict the processing value, and the relevant value is fed back to the robotic arm press head to perform product shaping processing.

[0073] S3. After processing, remeasure the product data and make a judgment to obtain the processing result;

[0074] S4. The system trains the machine learning prediction model online according to the time interval set by the user. Every once in a while, the system will add the newly processed data to the sample library to retrain the model. The time interval can be 24 hours or the time to process 100 workpieces, depending on the standard specified by the user.

[0075] In step S1, when establishing the MLP model, two models are created for the front, left, and right sides of the product to be processed. For each side of the product to be processed, a primary shaping MLP model and a secondary shaping MLP model are set up. The primary shaping MLP model and the secondary shaping MLP model are respectively for the primary shaping and secondary shaping of one side of the product to be processed. In this way, when predicting the processing amount, the deformation error of other sides caused by the deformation of one side is reduced.

[0076] In step S1, the process of establishing the MLP model includes the following steps:

[0077] S11. Use Tensorflow.keras to introduce the MLP model, stack the multi-layer structure, and complete the model construction.

[0078] S12. For each prediction process of each edge, a separate model is trained. The data used for training the MLP model is the raw processing data obtained from the actual production process. The raw processing data includes the product output information, the point values ​​and straightness before and after each processing, and the target specification range information of the product. The raw processing data is classified and extracted according to the edge and the number of processing to obtain six sets of data, which are the two processing data of the front, left, and right edges.

[0079] S13. After obtaining the processing data for the front, left, and right sides of the product, define the input parameters of the MLP model. Set the value of the first hidden layer to 32, approximately three times the input dimension, to amplify its impact on the results. Set the value of the second hidden layer to 64 and the value of the third hidden layer to 32. Finally, output a value as the predicted processing quantity. Due to the different types of materials and the fact that some materials are already processed, the parameters to be input into the model are (taking the processing of the left side as an example; the right side is similar to the front): The input parameters are defined as follows: The parameter dimensions are described as follows: ① Raw: 1 dimension;

[0080] Note: There can be multiple types of materials, such as three categories: A, B, and C. Therefore, they need to be digitized. If the material is category A, the Raw value is 0; if it is category B, the Raw value is 1; and if it is category C, the Raw value is 2.

[0081] ② FR: 1 dimension;

[0082] Note: There are two properties: new and old. The new property is set to 1, and the old property is set to 0.

[0083] ③ aimSPC: 1 dimension, the definition of optimal straightness is as follows: ④ Press: 1 dimension;

[0084] ⑤ SPC_left: 1 dimension;

[0085] ⑥ SPC_left_res: 1 dimension;

[0086] ⑦ Ci_before: 20 dimensions;

[0087] Note: For the preceding terms, Ci_before, Ci, and Ci_res each have 31 dimensions.

[0088] ⑧ Ci: It has 20 dimensions.

[0089] ⑨ Ci_res: There are 20 dimensions. The specific definition of Ci_res is as follows: S14. The original dataset consisting of the two processing data of the front, left and right sides of the product is divided into a 2:8 ratio to obtain the test set and the training set. The MLP model is trained using the training set.

[0090] S15. Use the MAE metric to evaluate the MLP model. The MAE calculation formula is as follows: .in, The total number of datasets in the test set; The target value corresponding to the test set; The predicted values ​​are the corresponding performance values ​​in the test set. MAE uses the norm between the predicted and the true values ​​to define the error, which has a high tolerance for outliers and better robustness to errors.

[0091] To prevent overfitting, each layer is activated using the ReLU function with a 15% dropout rate, and the MAE function is used for backpropagation correction. The trained model is evaluated using the test set, and the dataset is adjusted. Training ends when both the training and test sets are well-fitted, and the MAE loss function of both is less than 0.2.

[0092] The MLP model in this system has a four-layer structure, as shown in the attached figure. Figure 5 As shown.

[0093] It's important to note that because the front of the product being shaped is longer than the left and right sides, it has more measurement points. Therefore, the dimensions of the two front models are different from those of the left and right models. The left and right sides each have 20 measurement points, while the front has 31. The input dimensions are 66 dimensions for the left and right sides and 99 dimensions for the front.

[0094] In step S2, the product shaping process includes the following steps:

[0095] S21. After the product to be processed is placed into the shaping machine, it is first fixed on the shaping platform. The laser scanner scans the points on the front, left and right sides of the product to be processed to obtain the height value of all points and then transmits it to the industrial computer.

[0096] S22. After receiving the height value of the material point, the industrial computer obtains the specification range input by the user on the front-end interface, puts all these data into the offline trained machine learning prediction model, and waits for the shaping and processing results.

[0097] S23. After the data is put into the machine learning prediction model, the dimensions of the input data are 66 (left and right) and 99 (front). After the model calculates, the corresponding processing value is predicted and submitted to the industrial computer.

[0098] S24. After receiving the predicted processing value, the industrial computer feeds it back to the controller, which then controls the robotic arm pressure head to perform a pressing operation to complete the shaping process.

[0099] On the other hand, the present invention also discloses an intelligent shaping and processing control system based on machine learning, comprising:

[0100] An industrial computer for running shaping control programs and machine learning algorithms, wherein the shaping control programs include instructions for performing any of the methods described above;

[0101] A scanner is used to measure the height of points on each side of a product to be processed, and then transmits the height values ​​of the points to an industrial computer to realize machine learning algorithms and shaping control.

[0102] A shaping platform, used to hold products to be processed;

[0103] The robotic arm pressure head is used to compress and deform the product being processed and hold it for a period of time.

[0104] The controller receives signals from the industrial computer, converts the signals, and transmits them to the robotic arm to control the robotic arm's pressure head.

[0105] An industrial computer connects the scanner and robotic arm via a data cable to control the entire shaping system. The robotic arm's pressure head is a rubber block with a specific curved surface, mounted at the top of the robotic arm. During shaping, the robotic arm drives the pressure head to act on the workpiece, compressing and deforming it for a period of time to achieve the shaping effect. The scanner is a laser scanner that performs statistical analysis of material-related parameters, which are then transmitted to the industrial computer for algorithmic control. The scanner's built-in high-precision laser measurement subsystem travels along the workpiece edge, measuring each sampling point using a discrete sampling method.

[0106] Working principle:

[0107] This system replaces traditional expert experience base models with machine learning prediction models. It analyzes and processes workpiece parameters to train a prediction model that automatically calculates the stroke amount for each shaping process. The machine learning model then provides a predicted value for the processing amount, enabling prediction of workpiece processing quantity and judgment of product quality during the workpiece material shaping process. This system can learn and update dynamically online. It exhibits strong adaptability, effectively reducing the impact of uncertainties such as elastic deformation, and has low error. In actual production, it can improve processing efficiency and reduce labor costs. This technology can be applied to flexible processing applications such as product processing and shaping.

[0108] In this invention, the number of measurement points involved in the calculation changes from three points to multiple points (all measurement points). Traditional shaping schemes determine straightness specifications based on three key points of each edge, without performing any related operations or calculations on the remaining points. In this application, to obtain the processing status of each edge, the system sets 60 points. The machine learning-based scheme considers all points and aims to process all points to the "optimal value," that is, the midpoint of the specification interval. This means that the height values ​​of all points and the midpoint of the specification interval are included in the model, allowing the predicted processing amount to better take into account the relationship between all points and the midpoint of the interval.

[0109] Employing data processing and machine learning methods for predicting processing values ​​effectively reduces the impact of outliers, resulting in better performance and stronger adaptability. This makes the processing more intelligent, improves efficiency, and reduces labor costs. Training a model with past production data for a specific material reduces the amount of data required compared to manual experience. Considering the interrelationships between the three sides of the product requiring shaping, a separate model is designed and trained for each side, leading to higher shaping accuracy.

[0110] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A machine learning-based intelligent shaping and processing control method, characterized in that, Includes the following steps: S1. Using all the previously collected processing data, a training set is established in combination with existing samples, and offline training is performed to obtain a machine learning prediction model, which is an MLP model. S2. The product to be processed is put into the offline trained machine learning prediction model to predict the processing value, and the relevant value is fed back to the robotic arm press head to perform product shaping processing. S3. After processing, remeasure the product data and make a judgment to obtain the processing result; S4. Train the machine learning prediction model online according to the time intervals set by the user; In step S1, when establishing the MLP model, two models are established for the front, left, and right sides of the product to be processed. Each side of the product to be processed is provided with a primary shaping MLP model and a secondary shaping MLP model. The primary shaping MLP model and the secondary shaping MLP model are respectively for the primary shaping and secondary shaping of one side of the product to be processed. The process of building an MLP model includes the following steps: S11. Use Tensorflow.keras to introduce the MLP model, stack the multi-layer structure, and complete the model construction. S12. For each prediction process of each edge, a separate model is trained. The data used for training the MLP model is the raw processing data obtained from the actual production process. The raw processing data includes the product output information, the point values ​​and straightness before and after each processing, and the target specification range information of the product. The raw processing data is classified and extracted according to the edge and the number of processing to obtain six sets of data, which are the two processing data of the front, left, and right edges. S13. After obtaining the processing data for the front, left, and right sides of the product, define the input parameters of the MLP model. Set the value of the first hidden layer to 32, the value of the second hidden layer to 64, and the value of the third hidden layer to 32. Finally, output a value as the predicted processing quantity. The input parameters are defined as follows: S14. The original dataset consisting of the two processing data of the front, left and right sides of the product is divided into a 2:8 ratio to obtain a test set and a training set. The MLP model is trained using the training set. S15. Use the MAE metric to evaluate the MLP model. The MAE calculation formula is as follows: ,in, The total number of datasets in the test set; The target value corresponding to the test set; These are the predicted performance values ​​for the corresponding test set.

2. The intelligent shaping and processing control method based on machine learning according to claim 1, characterized in that, In step S2, the product shaping process includes the following steps: S21. After the product to be processed is placed into the shaping machine, it is first fixed on the shaping platform. The laser scanner scans the points on the front, left and right sides of the product to be processed to obtain the height value of all points and then transmits it to the industrial computer. S22. After receiving the height value of the material point, the industrial computer obtains the specification range input by the user on the front-end interface, puts all these data into the offline trained machine learning prediction model, and waits for the shaping and processing results. S23. After the data is fed into the machine learning prediction model, the model calculates and predicts the corresponding processing value, which is then submitted to the industrial computer. S24. After receiving the predicted processing value, the industrial computer feeds it back to the controller, which then controls the robotic arm pressure head to perform a pressing operation to complete the shaping process.

3. A machine learning-based intelligent shaping and processing control system, characterized in that, include: An industrial computer for running a shaping control program and a machine learning algorithm, the shaping control program including instructions for performing any of the methods according to claim 1 or 2; A scanner is used to measure the height of points on each side of a product to be processed, and then transmits the height values ​​of the points to an industrial computer to realize machine learning algorithms and shaping control. A shaping platform, used to hold products to be processed; The robotic arm pressure head is used to compress and deform the product being processed and hold it for a period of time. The controller receives signals from the industrial computer, converts the signals, and transmits them to the robotic arm to control the robotic arm's pressure head.

Images