A complex curved surface stamping process control method and system based on process energy

Abstract

The present application belongs to the technical field of numerical control machining, and particularly relates to a complex curved surface stamping machining control method and system based on process energy, and corresponding stamping equipment.S1: a blank holder force prediction network is constructed.S2: a stamping total stroke of a workpiece to be machined and a reference process energy curve of ideal machining conditions are preset.S3: displacement and pressure signals are collected, real-time process energy is calculated, and the reference process energy is inquired.S4: process energy deviation is calculated, and the following decisions are made: S41: whether the process energy deviation is out of limit is judged;S42: whether the real-time stamping depth is out of limit is judged;S43: if the former is out of limit and the latter is not out of limit, the blank holder force prediction network is used to predict the blank holder force control amount;S44: the blank holder force of the next stamping action is adjusted.S5: after each stamping action is completed, the steps of S3-S4 are repeated until the stamping workpiece machining is completed.The present application solves the problem that the traditional method cannot achieve good control effect in the stamping workpiece machining process of a complex curved surface.

Description

Technical Field

[0001] This invention belongs to the field of CNC machining technology, specifically relating to a method and system for controlling the stamping of complex curved surfaces based on process energy, as well as stamping equipment using the corresponding control system. Background Technology

[0002] Stamping is one of the main methods for plastic deformation of metals. Its forming process includes basic steps such as blanking, deep drawing, bending, punching, and flanging. The stamping process is a complex mechanical process. Besides the varying shapes of the workpieces, many factors, such as blank holder force, lubrication method, blank size, and processing environment, affect the final forming quality of the stamped parts. Without proper process measures, common quality problems such as wrinkling, cracking, and partial springback may occur during the forming process, making it difficult to obtain qualified products. Improving the quality of formed parts and reducing the probability of scrap is one of the current hot topics and challenges in stamping technology.

[0003] Traditional quality control methods primarily rely on tracking the blank holder force curve. While this allows for precise control of the stamping action, it fails to dynamically monitor the quality of the stamped parts. Consequently, it cannot adjust the stamping action based on the actual stamping effect. Furthermore, stamping is a complex and diverse process with varied workpiece material characteristics, including bending, dents, and protrusions. The methods for monitoring workpiece deformation also differ depending on the type of workpiece being processed. Therefore, traditional quality control methods have significant limitations when applied to different types of workpieces and cannot meet the quality control requirements of complex curves. Summary of the Invention

[0004] To address the problem that traditional stamping quality control methods cannot achieve good control results in the processing of stamped parts with complex curved surfaces, this invention provides a method and system for controlling the stamping process of complex curved surfaces based on process energy, as well as stamping equipment using the corresponding control system.

[0005] This invention is achieved using the following technical solution:

[0006] A method for controlling the stamping process of complex curved surfaces based on process energy is disclosed. This method dynamically adjusts the impact blank holder force according to the process energy deviation in each stamping cycle of the stamping equipment, thereby improving the processing quality of the stamped parts. The method for controlling the stamping process of complex curved surfaces includes the following steps:

[0007] S1: Construct and train a stacking framework-based network for predicting blanking force. The input to the blanking force prediction network is the stamping depth h. p Initial blank holder force F p Process energy deviation ΔEp The output is the blank holder force adjustment amount ΔF.

[0008] S2: Preset total stamping stroke h of the workpiece to be processed max A reference process energy curve under ideal processing conditions. The reference process energy curve includes a real-time stamping depth h. p,i Compared with the reference process energy E i The mapping relationship between them.

[0009] S3: Collect displacement and pressure signals at preset measurement points during the stamping process of the stamping equipment, and calculate the corresponding real-time process energy E. p,i And obtain the corresponding reference process energy E by querying the reference process energy curve. i .

[0010] S4: Calculate the process energy deviation ΔE corresponding to each stamping action performed by the stamping equipment. p,i ΔE p,i =E p,i -E i Based on the process energy deviation, the following decisions are made:

[0011] S41: Determine the process energy deviation ΔE p,i If the deviation exceeds the preset upper limit, proceed to the next step; otherwise, execute according to the preset stamping plan.

[0012] S42: Determine the real-time stamping depth h at the current moment. p,i Does it exceed h? max If the upper limit is reached, the stamping process ends; otherwise, proceed to the next step.

[0013] S43: Real-time stamping depth h p,i Real-time blank holder force F p,i and real-time process energy deviation ΔE p,i The input is fed into the blank holder force prediction network, and the corresponding blank holder force control amount ΔF is output.

[0014] S44: The blank holder force F to be executed in the next stamping action is determined based on the blank holder force adjustment amount ΔF. p,i+1 Adjustments to be made: F p,i+1 =F p,i +ΔF.

[0015] S5: After each stamping action, repeat steps S3-S4 until the stamped part is completed.

[0016] As a further improvement to this invention, the blank holder force prediction network includes several baseline models and one meta-model. The input to the baseline model is the stamping depth h. p Initial blank holder force F p and process energy deviation ΔEp Output the predicted blank holder force offset ΔF p The meta-model is based on the blank holder force offset ΔF predicted by each baseline model. p Output a corresponding blank holder force adjustment amount ΔF.

[0017] Furthermore, the construction and training process of the edge pressure force prediction network includes the following steps:

[0018] S01: Using SVM, Xgboost, LSTM and KNN as baseline models, construct a stacking framework based on ensemble learning to predict edge pressure force.

[0019] S02: Collect the raw data corresponding to the stamping task in the simulation software. The raw data collection process is as follows:

[0020] S021: The total stamping depth h of any stamping task max Divided into m equal parts, the i-th stamping depth can be expressed as h. p,i , where i = (1,2,…,m).

[0021] S022: The initial blank holder force F for the stamping task p Divided into k equal parts, the u-th initial blank holder force can be expressed as F. p,u , where u = (1,2,…,k).

[0022] S023: Perform stamping simulation using simulation software to calculate the stamping depth h at any given depth. p,i and initial blank holder force F p,u

[0023] The process energy value under the given conditions is denoted as E. p,i,u .

[0024] S024: Generate and record arbitrary stamping depth h p,i Under certain conditions, the adjustment amount of blank holder force ΔF compared with a specified initial blank holder force under different initial blank holder forces is denoted as ΔF. w w represents the amount of data for adjusting the blank holder force, w = (1, 2, ..., u-1).

[0025] S025: Generate and record different stamping depths h p,i Conditions, process energy deviation ΔE caused by different initial blank holder forces p .

[0026] S03: Based on the raw data collected in the previous step, according to the process energy deviation ΔE p Initial blank holder force F p The mapping relationship between the blank holder force control amount ΔF and the corresponding original dataset (ΔE, F) is generated. p,ΔF).

[0027] S04: Divide the original dataset to obtain the training and test sets for the training baseline model and the meta-model, and complete the training of the edge pressure force prediction network.

[0028] As a further improvement of the present invention, the method for selecting the baseline model in step S01 is as follows:

[0029] (1) Select several different types of machine learning models, and set the input of the machine learning model as the stamping depth h. p Initial blank holder force F p and process energy deviation ΔE p The output is set to the blank holder force control amount (ΔF).

[0030] (2) Pre-train each machine learning model using a pre-prepared dataset.

[0031] (4) Perform Pearson correlation analysis on the output values ​​of each selected machine learning model. The Pearson correlation coefficient r between any two machine learning models is calculated. xy It is expressed as follows:

[0032]

[0033] In the above formula, ΔF p,i,x and ΔF p,i,y These are the i-th predicted values ​​output by the x-th and y-th models, respectively; ΔF p,x and ΔF p,y , respectively, are the average values ​​of the x-th and y-th model predictions; I represents the number of samples.

[0034] (5) Based on the analysis results of the Pearson correlation coefficient, select a specified number of machine learning models with relatively low correlation coefficients as the required baseline models.

[0035] As a further improvement of the present invention, in step S04, the content of the edge clamping force prediction network during the training and testing phases is as follows:

[0036] (1) Based on the number of baseline models n, the original dataset (ΔE, F) is divided into two parts. p The dataset is divided into n subsets on average, ΔF and ΔE. i ,F p,i ,ΔF i ), i = 1, 2, ..., n}.

[0037] (2) First, train each baseline model n times, and then use the training result of the i-th time, i.e., the i-th subset (ΔE) i ,F p,i ,ΔF iThe test set is used as the test set, and the remaining n-1 subsets are kept as the training set.

[0038] (3) Combine the prediction results (ΔF) of the n baseline models p ) are constructed into a new n×J vector for predicting the blank holder force, and then these prediction results (ΔF) are used to construct the vector. p A new dataset (ΔF) is constructed by combining the measured blank holder force control amount ΔF with the actual measured blank holder force control amount ΔF. p ,ΔF).

[0039] (4) Finally, use the new dataset (ΔF) p The training set of the meta-model (ΔF) is used to complete the training of the edge pressure force prediction network.

[0040] (5) After training, the blank holder force prediction network is trained using the test set. The parameters of the network model with the best evaluation index are retained as the blank holder force prediction network required to predict the blank holder force control amount.

[0041] As a further improvement of the present invention, in the original dataset (ΔE,F) p In ΔF), ΔE is a 2×J matrix used to store and characterize the process energy deviation, expressed as:

[0042] ΔE=(h p ,ΔE p )

[0043] In the above formula, h p Let J be a 1×J stamping depth vector, denoted as (h p =[h p,1 ,h p,2 ,...,h p,j ] -1 );ΔE p Let ΔE be a 1×J process energy deviation vector, denoted as (ΔE) p =[ΔE p,1 ,ΔE p,2 ,...,ΔE p,J ] -1 ).

[0044] Define F p A measured blank holder force vector of 1×J is expressed as:

[0045] F p =[F p,1 ,F p,2 ,...,F p,J ] -1

[0046] ΔF is the vector of 1×J blank holder force control, which can be expressed as:

[0047] ΔF = [ΔF1, ΔF2, ..., ΔF J ] -1

[0048] Then, ΔE i F p,i ΔF i The process energy deviation matrix ΔE and blank holder force vector F correspond to 2×J, 1×J, and 1×J respectively. p The elements of the i-th subprocess in the blank holder force control vector ΔF

[0049] ΔE i Represented as:

[0050] ΔE i =(h p,i ,ΔE p,i )

[0051] Where h p,i Let ΔE be the vector representing the depth of the i-th stamping process of 1×J. p,i Let J be the process energy deviation vector of the i-th subprocess.

[0052] As a further improvement of the present invention, during the testing process of the blank holder force prediction network, the average absolute percentage error e is used. MA and mean square error e MS Two metrics are used to evaluate the predictive performance of the meta-model, and e is selected. MA and e MS The minimum network model serves as the final meta-model; percentage error e MA and mean square error e MS The expression is as follows:

[0053]

[0054]

[0055] In the above formula, ΔF i ΔF represents the edge clamping force adjustment amount in the test dataset. p,i I represents the predicted edge pressure adjustment amount by the model; I represents the number of samples.

[0056] As a further improvement of the present invention, in step S3, the real-time stamping depth h of the stamping process is first calculated based on the displacement signal data. p,i Then, by querying the reference process energy curve, the corresponding reference process energy E is obtained. i Finally, by combining the displacement and pressure signal data, the required real-time process energy E is obtained by integrating the measured impact force with respect to the real-time displacement. p,i Its expression is as follows:

[0057]

[0058] In the above formula, h p,i F(h) represents the real-time stamping depth during the stamping process, and F(h) is a function characterizing the stamping force F corresponding to each impact depth h during the current stamping process of the stamping equipment.

[0059] The present invention also includes a process energy-based complex surface stamping control system, which is installed in a stamping device and used to evaluate the effect of each stamping action in the stamping process using the aforementioned process energy-based complex surface stamping control method, and to provide feedback control to the stamping device based on the evaluation results. The process energy-based complex surface stamping control system includes: a displacement sensor, a pressure sensor group, a data acquisition module, a data processing module, and a feedback control module.

[0060] A displacement sensor is installed between the punch and die in the stamping equipment to measure the stamping depth h in real time. Pressure sensor arrays are installed below the blank holder and on the punch in the stamping equipment; these pressure sensor arrays are used to measure the real-time blank holder force and real-time stamping force during the stamping process. A data acquisition module is used to acquire the measurement signals from the displacement sensor and pressure sensor array in real time and obtain the corresponding measurement data.

[0061] The data processing module includes a storage unit, a calculation unit, and a judgment unit. The storage unit stores the total stamping stroke h of the workpiece currently being processed. max The system includes a reference process energy curve for ideal processing conditions and a trained blank holder force prediction network. The computational unit calculates the corresponding real-time process energy E based on the measurement data acquired by the data acquisition module during each stamping operation. p,i And obtain the corresponding reference process energy E by querying the reference process energy curve. i Finally, the corresponding process energy deviation ΔE was calculated. p The judgment unit is used to first determine the current stamping stroke h. p Has the preset total stamping stroke h been reached? max When h p ≥h max Then the stamping task ends when h p <h max Then continue to determine the process energy deviation ΔE of the computing unit. p Does it exceed the preset safety range ΔE? pmax When ΔE p ≥ΔE pmax If ΔE is applied, the blank holder force for the next stamping cycle will be adjusted; when ΔE p <ΔE pmax Then wait to execute the next round of stamping.

[0062] The feedback control module is connected to the main control system of the stamping equipment. The feedback control module obtains the result of the judgment unit in the data processing module and makes the following decisions based on the judgment result: (1) When the judgment unit determines that the stamping task should be ended, a stop command is issued to the main control system of the stamping equipment. (2) When the judgment unit determines that the blank holder force of the next stamping action should be adjusted, the blank holder force prediction network in the storage unit is run and called to obtain a corresponding blank holder force adjustment amount ΔF. Then, an initial blank holder force adjustment command is issued to the main control system of the stamping equipment. The adjusted initial blank holder force is F. p,i+1 =F p,i +ΔF.

[0063] The present invention also includes a stamping device that employs the aforementioned complex curved surface stamping processing control system based on process energy, thereby enabling online evaluation of processing quality during the stamping process and dynamic adjustment of the initial blank holder force in the stamping action based on the deviation of process energy, so as to improve the stamping quality of the workpiece.

[0064] The technical solution provided by this invention has the following beneficial effects:

[0065] This invention addresses the shortcomings of low control accuracy in stamping quality control systems during the machining of complex curved surfaces. It improves the feedback control method by using process energy as the control variable and process energy deviation as the feedback variable, thereby more accurately reflecting and controlling the forming quality during the stamping process. Based on this, a new stamping quality feedback control system is designed. This new process energy-based control method for complex curved surface stamping achieves better control and improvement of the final quality of stamped parts.

[0066] In addition, in order to correct the deviation of process energy in the processing, this invention also designs a new blank holder force prediction network. This scheme is established by an ensemble learning method, which can integrate the characteristics and advantages of multiple machine learning models. It is applicable to stamping scenarios with various complex workpiece materials and diverse characteristics, and the prediction results have stronger accuracy and generalization ability. Attached Figure Description

[0067] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0068] Figure 1 This is a flowchart of the complex surface stamping control method based on process energy provided in Embodiment 1 of the present invention.

[0069] Figure 2 This is the architecture of the edge pressure force prediction network designed in Embodiment 1 of the present invention.

[0070] Figure 3 A flowchart illustrating the steps involved in constructing a method for predicting edge pressure force.

[0071] Figure 4 This is a block diagram of the complex curved surface stamping control system based on process energy designed in Embodiment 2 of the present invention.

[0072] Figure 5 This is a simplified structural diagram of a stamping die.

[0073] Figure 6 This is a flowchart illustrating the operation of the stamping equipment using a complex curved surface stamping processing control system based on process energy, as described in Embodiment 3 of the present invention. Detailed Implementation

[0074] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0075] Example 1

[0076] This embodiment provides a control method for complex curved surface stamping based on process energy. It dynamically adjusts the impact blank holder force according to the process energy deviation in each stamping cycle of the stamping equipment, thereby improving the processing quality of the stamped parts. The idea behind this complex curved surface stamping control method is to create a feedback control system outside the existing stamping equipment, monitor the process energy of each stamping cycle, determine the process energy deviation caused by each cycle, and finally adjust the initial blank holder force for the next cycle based on the process energy deviation.

[0077] The core of this scheme comprises two aspects. First, it involves constructing a feedback control system that enables online monitoring and evaluation of stamping quality, and then uses the evaluation results to adjust the stamping action of the stamping equipment in real time. Second, it utilizes machine learning algorithms to learn from simulation data of the stamping process, thereby obtaining a neural network that can predict the initial edge pressure adjustment range based on detected process energy deviations.

[0078] Regarding the first aspect of the work, the workflow of the feedback control system designed in this embodiment is as follows: Figure 1 As shown. Specifically, this method for controlling the stamping process of complex curved surfaces includes the following steps:

[0079] S1: Construct and train a stacking framework-based network for predicting blanking force. The input to the blanking force prediction network is the stamping depth h. p Initial blank holder force F p Process energy deviation ΔE pThe output is the blank holder force adjustment amount ΔF.

[0080] S2: Preset total stamping stroke h of the workpiece to be processed max A reference process energy curve under ideal processing conditions. The reference process energy curve includes a real-time stamping depth h. p,i Compared with the reference process energy E i The mapping relationship between them.

[0081] S3: Collect displacement and pressure signals at preset measurement points during the stamping process of the stamping equipment, and calculate the corresponding real-time process energy E. p,i And obtain the corresponding reference process energy E by querying the reference process energy curve. i The specific process is as follows:

[0082] First, calculate the real-time stamping depth h during the stamping process based on the displacement signal data. p,i Then, by querying the reference process energy curve, the corresponding reference process energy E is obtained. i Finally, by combining the displacement and pressure signal data, the required real-time process energy E is obtained by integrating the measured impact force with respect to the real-time displacement. p,i Its expression is as follows:

[0083]

[0084] In the above formula, h p,i F(h) represents the real-time stamping depth during the stamping process, and F(h) is a function characterizing the stamping force F corresponding to each impact depth h during the current stamping process of the stamping equipment.

[0085] S4: Calculate the process energy deviation ΔE corresponding to each stamping action performed by the stamping equipment. p,i ΔE p,i =E p,i -E i Based on the process energy deviation, the following decisions are made:

[0086] S41: Determine the process energy deviation ΔE p,i If the deviation exceeds the preset upper limit, proceed to the next step; otherwise, execute according to the preset stamping plan.

[0087] S42: Determine the real-time stamping depth h at the current moment. p,i Does it exceed h? max If the upper limit is reached, the stamping process ends; otherwise, proceed to the next step.

[0088] S43: Real-time stamping depth h p,i Real-time blank holder force F p,i and real-time process energy deviation ΔE p,iThe input is fed into the blank holder force prediction network, and the corresponding blank holder force control amount ΔF is output.

[0089] S44: The blank holder force F to be executed in the next stamping action is determined based on the blank holder force adjustment amount ΔF. p,i+1 Adjustments to be made: F p,i+1 =F p,i +ΔF.

[0090] S5: After each stamping action, repeat steps S3-S4 until the stamped part is completed.

[0091] Regarding the second aspect of the work, this embodiment adopts the following approach: Figure 2 The stacking framework of ensemble learning shown is used to design the required blank holder force prediction network. This network model effectively overcomes the difficulty of obtaining accurate prediction results in the stamping parameter analysis scenario of this embodiment due to the inability of a single relation prediction model. Specifically, the blank holder force prediction network designed in this embodiment includes four baseline models and one meta-model. The input of the baseline model is the stamping depth h. p Initial blank holder force F p and process energy deviation ΔE p Output the predicted blank holder force offset ΔF p The meta-model is based on the blank holder force offset ΔF predicted by each baseline model. p Output a corresponding blank holder force adjustment amount ΔF.

[0092] Specifically, such as Figure 3 As shown, the construction and training process of the edge pressure force prediction network includes the following steps:

[0093] S01: Using SVM, Xgboost, LSTM and KNN as baseline models, construct a stacking framework based on ensemble learning to predict edge pressure force.

[0094] In this embodiment, the baseline model is selected as follows:

[0095] (1) Select several different types of machine learning models, and set the input of the machine learning model as the stamping depth h. p Initial blank holder force F p and process energy deviation ΔE p The output is set to the blank holder force control amount (ΔF).

[0096] (2) Pre-train each machine learning model using a pre-prepared dataset.

[0097] (4) Perform Pearson correlation analysis on the output values ​​of each selected machine learning model. The Pearson correlation coefficient r between any two machine learning models is calculated. xy It is expressed as follows:

[0098]

[0099] In the above formula, ΔF p,i,x and ΔF p,i,y These are the i-th predicted values ​​output by the x-th and y-th models, respectively; ΔF p,x and ΔF p,y , respectively, are the average values ​​of the x-th and y-th model predictions; I represents the number of samples.

[0100] (5) Based on the analysis results of the Pearson correlation coefficient, select a specified number of machine learning models with relatively low correlation coefficients as the required baseline models.

[0101] S02: Collect the raw data corresponding to the stamping task in the simulation software. The raw data collection process is as follows:

[0102] S021: The total stamping depth h of any stamping task max Divided into m equal parts, the i-th stamping depth can be expressed as h. p,i , where i = (1,2,…,m).

[0103] S022: The initial blank holder force F for the stamping task p Divided into k equal parts, the u-th initial blank holder force can be expressed as F. p,u , where u = (1,2,…,k).

[0104] S023: Perform stamping simulation using simulation software to calculate the stamping depth h at any given depth. p,i and initial blank holder force F p,u The process energy value under the given conditions is denoted as E. p,i,u ; Among them, i=(1,2,…,m), u=(1,2,…,k).

[0105] S024: Generate and record arbitrary stamping depth h p,i Under certain conditions, the adjustment amount of blank holder force ΔF compared with a specified initial blank holder force under different initial blank holder forces is denoted as ΔF. w This includes: ΔF1 = F p,1 -F p,u ΔF2=F p,2 -F p,u , ..., ΔF w =F p,w -F p,u , ..., ΔF u =F p,u+1 -F p,u , ..., ΔF k-1 =F p,k -F p,uWhere w represents the amount of data for adjusting the blank holder force, w = (1,2,…,u-1).

[0106] S025: Generate and record different stamping depths h p,i Conditions, process energy deviation ΔE caused by different initial blank holder forces p Including: ΔE p,1 =E p,i,1 -E p,i,u ΔE p,2 =E p,i,2 -E p,i,u ..., ΔE p,w =E p,i,w -E p,i,u ..., ΔE p,u =E p,i,u+1 -E p,i,u ..., ΔE p,k-1 =E p,i,k -E p,i,u .

[0107] S03: Based on the raw data collected in the previous step, according to the process energy deviation ΔE p Initial blank holder force F p The mapping relationship between the blank holder force control amount ΔF and the corresponding original dataset (ΔE, F) is generated. p ,ΔF).

[0108] S04: Split the original dataset to obtain the training and test sets for the baseline model and the meta-model, and complete the training of the edge clamping force prediction network. The content of the edge clamping force prediction network in the training and testing phases is as follows:

[0109] (1) Based on the number of baseline models n, the original dataset (ΔE, F) is divided into two parts. p The dataset is divided into n subsets on average, ΔF and ΔE. i ,F p,i ,ΔF i ), i = 1, 2, ..., n}.

[0110] In the original dataset (ΔE, F) p In ΔF), ΔE is a 2×J matrix used to store and characterize the process energy deviation, expressed as:

[0111] ΔE=(h p ,ΔE p )

[0112] In the above formula, h p Let J be a 1×J stamping depth vector, denoted as (h p =[h p,1 ,h p,2 ,...,hp,j ] -1 );ΔE p Let ΔE be a 1×J process energy deviation vector, denoted as (ΔE) p =[ΔE p,1 ,ΔE p,2 ,...,ΔE p,J ] -1 ).

[0113] Define F p A measured blank holder force vector of 1×J is expressed as:

[0114] F p =[F p,1 ,F p,2 ,...,F p,J ] -1

[0115] ΔF is the vector of 1×J blank holder force control, which can be expressed as:

[0116] ΔF = [ΔF1, ΔF2, ..., ΔF J ] -1

[0117] Then, ΔE i F p,i ΔF i The process energy deviation matrix ΔE and blank holder force vector F correspond to 2×J, 1×J, and 1×J respectively. p The elements of the i-th subprocess in the blank holder force control vector ΔF

[0118] ΔE i Represented as:

[0119] ΔE i =(h p,i ,ΔE p,i )

[0120] Where h p,i Let ΔE be the vector representing the depth of the i-th stamping process of 1×J. p,i Let J be the process energy deviation vector of the i-th subprocess.

[0121] (2) First, train each baseline model n times, and then use the training result of the i-th time, i.e., the i-th subset (ΔE) i ,F p,i ,ΔF i The test set is used as the test set, and the remaining n-1 subsets are kept as the training set.

[0122] (3) Combine the prediction results (ΔF) of the n baseline models p) are constructed into a new n×J vector for predicting the blank holder force, and then these prediction results (ΔF) are used to construct the vector. p A new dataset (ΔF) is constructed by combining the measured blank holder force control amount ΔF with the actual measured blank holder force control amount ΔF. p ,ΔF).

[0123] (4) Finally, use the new dataset (ΔF) p The training set of the meta-model (ΔF) is used to complete the training of the edge pressure force prediction network.

[0124] (5) After training, the blank holder force prediction network is trained using the test set. The parameters of the network model with the best evaluation index are retained as the blank holder force prediction network required to predict the blank holder force control amount.

[0125] During the testing of the edge clamping force prediction network, the mean absolute percentage error e was used. MA and mean square error e MS Two metrics are used to evaluate the predictive performance of the meta-model, and e is selected. MA and e MS The smallest network model is used as the final meta-model; percentage error e MA and mean square error e MS The expression is as follows:

[0126]

[0127]

[0128] In the above formula, ΔF i ΔF represents the edge clamping force adjustment amount in the test dataset. p,i I represents the predicted edge pressure adjustment amount by the model; I represents the number of samples.

[0129] Example 2

[0130] Based on Embodiment 1, this embodiment provides a complex surface stamping processing control system based on process energy. This system is installed in a stamping machine and is used to evaluate the effect of each stamping action in the stamping process using the process energy-based complex surface stamping processing control method of Embodiment 1, and to provide feedback control to the stamping machine based on the evaluation results. Figure 4 As shown, the complex curved surface stamping processing control system based on process energy includes: a displacement sensor, a pressure sensor group, a data acquisition module, a data processing module, and a feedback control module.

[0131] Figure 5This is a simplified classic stamping die. In the figure, part A is the die cavity, and A1 is the blank holder ring of the die cavity; part B is the punch. In the complex curved surface stamping processing control system based on process energy in this embodiment, a displacement sensor is installed between the punch and die in the stamping equipment to measure the stamping depth h in real time. The pressure sensor group includes at least two pressure sensors, which are respectively installed below the blank holder ring and on the punch in the stamping equipment; the former is used to measure the real-time blank holder force during the stamping process, and the latter is used to measure the real-time stamping force during the stamping process. The data acquisition module is used to acquire the measurement signals from the displacement sensor and the pressure sensor group in real time and obtain the corresponding measurement data.

[0132] The data processing module includes a storage unit, a calculation unit, and a judgment unit. The storage unit stores the total stamping stroke h of the workpiece currently being processed. max The system includes a reference process energy curve for ideal processing conditions and a trained blank holder force prediction network. The computational unit calculates the corresponding real-time process energy E based on the measurement data acquired by the data acquisition module during each stamping operation. p,i And obtain the corresponding reference process energy E by querying the reference process energy curve. i Finally, the corresponding process energy deviation ΔE was calculated. p The judgment unit is used to first determine the current stamping stroke h. p Has the preset total stamping stroke h been reached? max When h p ≥h max Then the stamping task ends when h p <h max Then continue to determine the process energy deviation ΔE of the computing unit. p Does it exceed the preset safety range ΔE? pmax When ΔE p ≥ΔE pmax If ΔE is applied, the blank holder force for the next stamping cycle will be adjusted; when ΔE p <ΔE pmax Then wait to execute the next round of stamping.

[0133] The feedback control module is connected to the main control system of the stamping equipment. The feedback control module obtains the result of the judgment unit in the data processing module and makes the following decisions based on the judgment result: (1) When the judgment unit determines that the stamping task should be ended, a stop command is issued to the main control system of the stamping equipment. (2) When the judgment unit determines that the blank holder force of the next stamping action should be adjusted, the blank holder force prediction network in the storage unit is run and called to obtain a corresponding blank holder force adjustment amount ΔF. Then, an initial blank holder force adjustment command is issued to the main control system of the stamping equipment. The adjusted initial blank holder force is F. p,i+1 =F p,i+ΔF.

[0134] Example 3

[0135] Based on Examples 1 and 2, this embodiment provides a stamping equipment that employs a complex curved surface stamping processing control system based on process energy, as described in Example 2. This system enables online evaluation of processing quality during the stamping process and dynamic adjustment of the initial blank holder force during the stamping action based on deviations in process energy, thereby improving the stamping quality of the workpiece. After installing the complex curved surface stamping processing control system based on process energy, as... Figure 6 As shown, the operation process of the stamping equipment provided in this embodiment includes the following steps:

[0136] Step 1: Power on the system and set the total stamping stroke h. max .

[0137] Step 2: Set the reference process energy curve in the control program.

[0138] Step 3: Place the sheet metal on the pressure ring, and lower the die to the designated position to press the sheet metal down.

[0139] Step 4: Set the blank holder force for the stamping process in the main controller of the stamping equipment, and start processing;

[0140] Step 5: The data acquisition module acquires the detection signals from the displacement sensor and pressure sensor, and transmits the detection results of real-time displacement and measured blank holder force to the data processing module. The data processing module queries the preset reference process energy curve based on the real-time displacement and identifies the corresponding process energy E. i Used as a reference value.

[0141] Step 6: The data processing module transmits the real-time displacement and measured impact force measured by the displacement sensor and pressure sensor to the data processing module for real-time process energy E. p,i The calculation.

[0142] Step 7: The data processing module calculates the process energy deviation ΔE. p,i Calculate ΔE p,i =E p,i -E i ;

[0143] Step 8: Determine the deviation ΔE p,i Is it within the set adjustment range? If not, it indicates that the deviation is too large, the system will alarm and terminate the program. Wait for technicians to reset the stamping parameters.

[0144] If so, then continue to check whether h is satisfied. p,i <h max If so, the feedback control system will adjust the blank holder force F and the stamping depth h at this time.p and process energy deviation ΔE p The input is fed into the blank holder force prediction network, and the output is the blank holder force to be adjusted, ΔF. i And send it to the main controller of the stamping equipment to adjust the blank holder force to F for the next stamping cycle. i =F i +ΔF i ,

[0145] If h is not satisfied p,i <h max If the maximum stamping depth has been reached, the program ends and the stamping action is complete.

[0146] Step 9: The die rises to the designated position, the sheet metal is removed, and the task is completed.

[0147] The solution in this embodiment differs from that in Embodiment 1. In Embodiment 1, only the minimum value of the process energy deviation is set as the critical value for judging and executing different operations. If the real-time process energy deviation does not reach the minimum value, no adjustment is made in the next stamping process. If the process energy deviation exceeds the minimum value, the stamping parameters are adjusted in the next stamping process. In this embodiment, a dynamically adjustable range is selected as the criterion. If the detected real-time process energy deviation is within the range, an adjustment action of the stamping parameters is performed. If the detected real-time process energy deviation exceeds the adjustable range, no further adjustment of the stamping parameters is made; instead, a safety alarm is issued, requiring technicians to correct the stamping parameters. These two strategies are applicable to different scenarios. For example, the solution in Embodiment 1 is suitable for scenarios with relatively lenient process requirements, in which case this strategy can reduce the frequency of parameter adjustments by the main controller and improve processing efficiency. The latter is suitable for scenarios with more stringent processing quality requirements.

[0148] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for controlling the stamping process of complex curved surfaces based on process energy, characterized in that: It is used to dynamically adjust the impact blank holder force based on the process energy deviation in each stamping action of the stamping equipment, thereby improving the processing quality of the stamped parts; the complex curved surface stamping processing control method includes the following steps: S1: Construct and train a blank holder force prediction network based on an ensemble learning stacking framework. The input of the blank holder force prediction network is the stamping depth. h p Initial blank holder force F p Process energy deviation Δ E p The output is the blank holder force adjustment amount Δ F ; S2: Preset total stamping stroke of the workpiece h max Reference process energy curves under ideal processing conditions; The reference process energy curve includes a real-time stamping depth. h p,i Compared with reference process energy E i Mapping relationship between them; S3: Collect displacement and pressure signals at preset measurement points during the stamping process of the stamping equipment, and calculate the corresponding real-time process energy. E p,i And obtain the corresponding reference process energy by querying the reference process energy curve. E i , S4: Calculate the process energy deviation Δ corresponding to the i-th stamping action performed by the stamping equipment. E p,i Δ E p,i = E p,i - E i Based on the process energy deviation, the following decisions are made: S41: Determine Δ E p,i If the deviation exceeds the preset upper limit, proceed to the next step; otherwise, execute according to the preset stamping plan. S42: Determine the real-time stamping depth at the current moment. h p,i Does it exceed h max If the upper limit is reached, the stamping process ends; otherwise, proceed to the next step. S43: Real-time stamping depth h p,i Real-time blank holder force F p,i , and Δ E p,i The input is fed into the blank holder force prediction network, and the corresponding blank holder force adjustment amount Δ is output. F ; S44: Adjustment amount Δ based on blank holder force F Blanking force for the next stamping action to be performed F p,i+1 Adjustments will be made: F p,i+1 = F p,i +Δ F ; S5: After each stamping action, repeat steps S3-S4 until the stamped part is completed.

2. The method for controlling complex curved surface stamping based on process energy as described in claim 1, characterized in that: The blank holder force prediction network includes several baseline models and one meta-model; the input of the baseline model is the stamping depth. h p Initial blank holder force F p and process energy deviation Δ E p Output the predicted blank holder force offset Δ F p The meta-model is based on the blank holder force offset Δ predicted by each baseline model. F p Output a corresponding blank holder force adjustment amount Δ F .

3. The method for controlling complex curved surface stamping based on process energy as described in claim 2, characterized in that: The construction and training process of the edge pressure force prediction network includes the following steps: S01: Using SVM, Xgboost, LSTM and KNN as baseline models, construct a stacking framework based on ensemble learning to predict edge pressure force; S02: Collect the raw data corresponding to the stamping task in the simulation software. The raw data collection process is as follows: S021: Total stamping stroke for any stamping task h max Divided into equal parts m Part, No. i The stamping depth is expressed as h p,i ,in i =(1,2,…, m ); S022: The initial blank holder force for the stamping task. F p Divided into equal parts k Part, No. u The initial blank holder force is expressed as: F p,u ,in u =(1,2,…, k ); S023: Stamping simulation is performed using simulation software to calculate the stamping process under any conditions. h p,i and F p,u The process energy value under the given conditions is denoted as E p,i,u ; S024: Generate and record h p,i Under certain conditions, the blank holder force adjustment Δ compared to a specified initial blank holder force with different initial blank holder forces. F denoted as Δ F w , w The amount of data representing the adjustment of the blank holder force. w =(1,2,…, k -1); S025: Generate and record h p,i Conditions, process energy deviation Δ caused by different initial blank holder forces E p ; S03: Based on the raw data collected in the previous step, according to the process energy deviation Δ E p Initial blank holder force F p and blank holder force adjustment amount Δ F The mapping relationship between them is used to generate the corresponding original dataset. ; Δ E This is a matrix used to store and characterize process energy deviations; S04: Divide the original dataset to obtain the training and test sets for the training baseline model and the meta-model, and complete the training of the edge pressure force prediction network.

4. The method for controlling complex curved surface stamping based on process energy as described in claim 3, characterized in that: In step S01, the baseline model is selected as follows: (1) Select several different types of machine learning models, and set the input of the machine learning model as the stamping depth. h p Initial blank holder force F p and process energy deviation Δ E p The output is set to the blank holder force adjustment amount Δ F ; (2) Pre-train each machine learning model using the prepared dataset; (4) Perform Pearson correlation analysis on the output values ​​of each selected machine learning model, and calculate the Pearson correlation coefficient between any two machine learning models. r xy It is expressed as follows: In the above formula, Δ F p,i,x and Δ F p,i,y They are respectively from the first x The and the first y The output of the model is the first i One predicted value; Δ F p,x and Δ F p,y They are the first x The and the first y The average of the predictions from each model; I Indicates the number of samples; (5) Based on the analysis results of Pearson correlation coefficient, select a specified number of machine learning models with relatively low correlation coefficients as the required baseline models.

5. The method for controlling complex curved surface stamping based on process energy as described in claim 3, characterized in that: In step S04, the content of the edge clamping force prediction network during the training and testing phases is as follows: (1) Based on the number of baseline models n , the original dataset Divided into average n Subdatasets ; Where, Δ E For one 2×J The matrix used to store and characterize process energy deviations; Δ E i Corresponding Δ E The Middle i Elements of each subprocess; Δ F i The blank holder force control vector Δ F The Middle i The elements of each subprocess; J represents the number of columns in the matrix; (2) First, perform a test on each baseline model. n The training session will be held, and the first... i The training result of the first training iteration, i.e., the result of the second training iteration. i Subdatasets As a test set, keep the remaining ones n-1 Use one subset of the dataset as the training set; (3) n The prediction results Δ of the baseline model F p Constructing a new predictive edge clamping force n×J Vectors, and then these predictions Δ F p The measured blank holder force adjustment amount Δ F Construct a new dataset (Δ) F p ,Δ F ); (4) Finally, use the new dataset (Δ) F p ,Δ F This serves as the training set for the meta-model, and ultimately completes the training of the edge pressure force prediction network; (5) After training, the blank holder force prediction network is trained using the test set. The parameters of the network model with the best evaluation index are retained as the blank holder force prediction network required to predict the blank holder force control amount.

6. The method for controlling complex curved surface stamping based on process energy as described in claim 5, characterized in that: In the original dataset In the middle, Δ E For one 2×J The matrix used to characterize the energy deviation of the storage process is denoted as: In the above formula, the stamping depth h p For one 1×J The vector, denoted as . Process energy deviation Δ E p For one 1×J The vector, denoted as . ; definition F p For one 1×J The measured blank holder force vector is expressed as: Δ F for 1×J The blank holder force control vector is expressed as: Then, Δ E i , F p,i Δ F i Corresponding to 2×J, 1×J, 1×J Process energy deviation matrix Δ E , F p Δ F The Middle i Elements of a subprocess; Δ E i Represented as: 。 7. The method for controlling complex curved surface stamping based on process energy as described in claim 5, characterized in that: During the testing of the edge clamping force prediction network, the mean absolute percentage error e is used to measure the error. MA and mean square error e MS Two metrics are used to evaluate the predictive performance of the meta-model, and e is selected. MA and e MS The minimum network model serves as the final meta-model; percentage error e MA and mean square error e MS The expression is as follows: In the above formula, Δ F i This represents the amount of edge clamping force adjustment in the test dataset; Δ F p,i This represents the amount of blank holder force adjustment predicted by the model; I Indicates the number of samples.

8. The method for controlling complex curved surface stamping based on process energy as described in claim 1, characterized in that: In step S3, the real-time stamping depth of the stamping process is first calculated based on the displacement signal data. h p,i Then, the corresponding reference process energy is obtained by querying the reference process energy curve. E i Finally, by combining the displacement and pressure signals, the required real-time process energy is obtained by integrating the measured impact force with respect to the real-time displacement. E p,i Its expression is as follows: In the above formula, h p,i This refers to the real-time stamping depth during the stamping process. F(h) To characterize the impact depth during the stamping process of the current stamping equipment h Corresponding punching force F The function.

9. A control system for complex curved surface stamping based on process energy, characterized in that, It is installed in a stamping equipment and used to evaluate the effect of each stamping action in the stamping process using the energy-based complex curved surface stamping control method as described in any one of claims 1-8, and to provide feedback control to the stamping equipment based on the evaluation results; the energy-based complex curved surface stamping control system includes: A displacement sensor, installed between the punch and die in a stamping machine, is used to measure the stamping depth in real time during the stamping process. h ; A pressure sensor array is installed below the blank holder ring and on the punch in the stamping equipment; the pressure sensor array is used to measure the real-time blank holder force and the real-time stamping force during the stamping process. The data acquisition module is used to acquire the measurement signals of the displacement sensor and pressure sensor group in real time and obtain the corresponding measurement data; The data processing module includes a storage unit, a calculation unit, and a judgment unit; the storage unit is used to store the total stamping stroke of the workpiece to be processed. h max The system includes a reference process energy curve for ideal processing conditions and a trained blank holder force prediction network. The computing unit calculates the corresponding real-time process energy based on the measurement data acquired by the data acquisition module during each stamping operation. E p,i And obtain the corresponding reference process energy by querying the reference process energy curve. E i Finally, the corresponding process energy deviation Δ is calculated. E p The determination unit is used to first determine the stamping depth. h p Has the preset total stamping stroke been reached? h max :when h p ≥ h max Then the stamping task ends. h p < h max Then, the process energy deviation Δ calculated by the computing unit is further evaluated. E p Does it exceed the preset safety range? Δ E pmax When Δ E p ≥Δ E pmax If Δ is applied, the blank holder force for the next stamping cycle is adjusted; when Δ E p <Δ E pmax Then wait for the next round of stamping action; and The feedback control module is communicatively connected to the main control system of the stamping equipment. The feedback control module obtains the result of the judgment unit in the data processing module and makes the following decisions based on the judgment result: (1) When the judgment unit determines that the stamping task should be ended, it issues a stop command to the main control system of the stamping equipment; (2) When the judgment unit determines that the blank holder force of the next stamping action should be adjusted, it runs and calls the blank holder force prediction network in the storage unit to obtain a corresponding blank holder force adjustment amount Δ. F， Then, the main control system of the stamping equipment issues an adjustment command for the initial blank holder force, and the adjusted initial blank holder force is: F p,i+1 = F p,i +Δ F .

10. A stamping device, characterized in that: It employs the complex curved surface stamping control system based on process energy as described in claim 9.

Images