Blind-zone temperature prediction method and system for electric-assisted incremental forming based on generative adversarial network

Abstract

The present invention discloses a blind area temperature prediction method for electric-assisted incremental forming based on a generative adversarial network, which obtains a blind area temperature map of a sheet material processing area where a thermal imaging image is blocked during electric-assisted incremental forming of a sheet material, and obtains a non-blind area temperature map of a sheet material processing area where a thermal imaging image is not blocked; constructs a generative adversarial network model, whose input is a blind area temperature map, and whose output is a non-blind area temperature map; obtains a blind area temperature prediction model by training the blind area temperature map and the non-blind area temperature map obtained in step 1; and implements blind area temperature prediction by using the blind area temperature prediction model for the blind area temperature map to be predicted. The blind area temperature prediction is provided by a generative adversarial network, and the control accuracy of the forming temperature is improved, thereby improving the quality of the final product.

Description

Technical Field

[0001] The present invention relates to incremental forming processing, and in particular to a method and system for predicting temperature in blind areas of electrically assisted incremental forming based on a generative adversarial network. Background Art

[0002] Incremental forming is a flexible sheet metal forming technology that uses a simple geometric tool to gradually apply force to the sheet metal along a preset path, so that the material gradually deforms into the desired shape. The biggest advantage of this technology is that it can achieve moldless forming, greatly shortening production preparation time and reducing costs, while improving production flexibility, which is very suitable for small batch and multi-variety production mode. In addition, the incremental forming equipment has a simple structure, is easy to operate, and can respond quickly to market changes.

[0003] However, incremental forming still faces challenges when using materials with poor formability, such as titanium alloys, to manufacture parts. To solve this problem, the method of increasing the temperature is usually used to improve the formability of the material. The heating strategies include overall heating and local heating. Among them, electrically assisted incremental forming, as a local heating method, has shown significant advantages. Electrically assisted incremental forming technology can only locally heat the forming area without heating the entire workpiece, which not only saves energy, but also avoids unnecessary plastic deformation in the non-forming area due to excessive temperature, while reducing oxidation problems on the surface of the sheet.

[0004] This technology has been successfully applied to the forming of various metal materials such as magnesium alloys and aluminum alloys. In particular, titanium alloys such as TC4 alloy have become important materials in the fields of aerospace, automobiles, and medical due to their excellent fatigue strength, corrosion resistance, high strength-to-weight ratio, and good biocompatibility. However, its physical properties determine that it is extremely sensitive to temperature during the forming process. Therefore, precise control of the forming temperature has become one of the key factors to ensure the quality of the final product.

[0005] In terms of temperature monitoring, the traditional method is to use thermocouples to directly contact the plate or heating system to collect temperature data. Although this method is simple, due to the delay of heat conduction and heat loss, there is a deviation between the measurement result and the actual temperature. Infrared thermal imaging can measure the temperature of moving objects in real time, but its disadvantage is that the local area is easily affected by the obstruction of the processing equipment, especially near the tool, which will cause the temperature image to fail to reflect the temperature distribution of the processing deformation area. Summary of the invention

[0006] Purpose of the invention: In view of the above shortcomings, the present invention provides a method and system for predicting the temperature of blind areas of electrically assisted incremental forming based on a generative adversarial network, which can solve the problem that infrared thermal imaging cannot measure the temperature distribution of blocked blind areas.

[0007] Technical solution: To solve the above problems, the present invention adopts a blind area temperature prediction method of electric-assisted incremental forming based on a generative adversarial network, which includes the following steps:

[0008] Step 1: obtaining a temperature map of a blind area where a thermal imaging image of a sheet metal processing area is blocked during sheet metal electric-assisted incremental forming, and obtaining a temperature map of a non-blind area where a thermal imaging image of a sheet metal processing area is not blocked;

[0009] Step 2: Construct a generative adversarial network model, whose input is the blind spot temperature map and whose output is the non-blind spot temperature map; obtain a blind spot temperature prediction model by training the blind spot temperature map and the non-blind spot temperature map obtained in step 1;

[0010] Step 3: The blind spot temperature map to be predicted is used to predict the blind spot temperature through the blind spot temperature prediction model to realize the blind spot temperature prediction.

[0011] Furthermore, the method of obtaining an unobstructed temperature map without blind spots in the thermal imaging image of the sheet metal processing area includes obtaining the temperature map without blind spots by means of finite element simulation of electrically assisted single-point incremental forming, and obtaining the temperature map without blind spots by setting an infrared thermal imager on the opposite side where the sheet metal contacts the tool through an electrically assisted single-point incremental forming experiment.

[0012] Furthermore, the generative adversarial network model includes a generative network and an adversarial network. The generative network is used to generate a temperature map without blind spots based on the blind spot temperature map; the adversarial network is used to determine whether the input is a real temperature map without blind spots or a temperature map without blind spots generated by the generative network. The generative network and the adversarial network are trained alternately to obtain a blind spot temperature prediction model.

[0013] Furthermore, the acquired blind spot temperature map and non-blind spot temperature map are divided into a training data set and a test data set. The generative network and the adversarial network are alternately trained through the training data set. When the training reaches a preset number of times, the training is stopped to obtain a trained model. The trained model is tested through the test data set. If the test result meets the requirements, the trained model is used as the blind spot temperature prediction model; if the test result does not meet the requirements, the parameters of the generative network and the adversarial network are adjusted, and retraining is performed through the training data set.

[0014] Furthermore, the generation process of the generation network is specifically as follows:

[0015] The blind area temperature map is cropped, scaled, and the blind area is eliminated to obtain the feature map F1;

[0016] The feature map F1 is processed by the two-dimensional convolution module to obtain the feature map F2;

[0017] Feature map F2 is processed by the dilated convolution module to obtain feature map F3;

[0018] Feature map F3 is processed by the deconvolution module to obtain feature map F4;

[0019] The characteristic map F4 is concatenated with the characteristic map F1 to obtain the characteristic map F5, and the characteristic map F5 is used as the generated temperature map without blind spots.

[0020] Furthermore, the two-dimensional convolution module includes a first convolution pooling layer of 5×5 with a stride of 1×1, a second convolution pooling layer of 3×3 with a stride of 2×2, a third convolution pooling layer of 3×3 with a stride of 1×1, a fourth convolution pooling layer of 3×3 with a stride of 2×2, and a fifth convolution pooling layer of 3×3 with a stride of 1×1, which are connected in sequence.

[0021] Furthermore, the dilated convolution module includes a 3×3 first dilated convolution pooling layer with a stride of 1×1 and a dilation rate of 2, a 3×3 second dilated convolution pooling layer with a stride of 1×1 and a dilation rate of 4, a 3×3 third dilated convolution pooling layer with a stride of 1×1 and a dilation rate of 8, and a 3×3 fourth dilated convolution pooling layer with a stride of 1×1 and a dilation rate of 16, which are connected in sequence.

[0022] Furthermore, the deconvolution module includes a first convolution pooling layer of 3×3 with a step size of 1×1, a second convolution pooling layer of 5×5 with a step size of 1×1, a third deconvolution pooling layer of 4×4 with a step size of 1 / 2×1 / 2, a fourth convolution pooling layer of 3×3 with a step size of 1×1, a fifth deconvolution pooling layer of 4×4 with a step size of 1 / 2×1 / 2, and a sixth convolution pooling layer of 3×3 with a step size of 1×1.

[0023] Furthermore, the adversarial network includes five 5×5 convolutional pooling layers with a stride of 2×2 connected in sequence and a fully connected layer.

[0024] The present invention also adopts an electric-assisted incremental forming blind zone temperature prediction system based on a generative adversarial network, comprising:

[0025] A data acquisition module, used to acquire a temperature map of a blind area where the thermal imaging image of the sheet metal processing area is blocked during the electric-assisted incremental forming of the sheet metal, and to acquire a temperature map of a non-blind area where the thermal imaging image of the sheet metal processing area is not blocked;

[0026] A model building module is used to build a generative adversarial network model, whose input is a blind spot temperature map and whose output is a non-blind spot temperature map; a blind spot temperature prediction model is obtained by training the acquired blind spot temperature map and the non-blind spot temperature map;

[0027] The prediction module is used to predict the blind spot temperature through the blind spot temperature prediction model.

[0028] The blind area temperature prediction method of electric-assisted incremental forming based on generative adversarial network includes constructing data set, constructing generative network, constructing adversarial network, training model, testing model and practical model to obtain temperature image. The data set comes from the temperature distribution map obtained by finite element simulation and the reverse infrared temperature measurement map of electric-assisted single-point incremental forming. The training model obtains the temperature distribution law of the tool, workpiece and fixture when powered on, and calculates the temperature prediction distribution of the blind area in combination with the temperature map of the non-blind area. Finally, the predicted distribution is combined with the temperature map of the non-blind area for output display.

[0029] Beneficial effect: Compared with the prior art, the significant advantage of the present invention is that it can not only provide the ability to predict the blind area temperature through the generative adversarial network, but also make predictions based on the comprehensive finite element simulation data and the real data of the back surface temperature of the electrically assisted single-point incremental forming. Compared with the conventional prediction scheme based on single-dimensional finite element simulation data, the present invention has higher prediction accuracy and authenticity, which is beneficial to improving the control accuracy of the forming temperature, thereby improving the quality of the final product. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Schematic diagram of the prediction method of the present invention.

[0031] Figure 2 This is a schematic diagram of the electric assisted incremental forming which is the research object of the present invention.

[0032] Figure 3 This is a diagram of the electric-assisted single-point incremental forming experimental equipment in the present invention.

[0033] Figure 4 Schematic diagram of the construction process of the generative adversarial network model in the present invention.

[0034] Figure 5 Schematic diagram of the generation process of the generated network in the present invention.

[0035] Figure 6 It is a schematic diagram of the structure of the two-dimensional convolution module in the present invention.

[0036] Figure 7 Schematic diagram of the structure of the hollow convolution module in the present invention.

[0037] Figure 8 Schematic diagram of the structure of the deconvolution module in the present invention.

[0038] Fig. 9 It is a schematic diagram of the structure of the adversarial network in the present invention.

[0039] Fig.10 It is a real-time infrared temperature image of the electrically assisted incrementally formed sheet material that is blocked in the present invention.

[0040] Fig.11This is a temperature image of the unobstructed blind area of ​​the electric-assisted incremental forming in the present invention.

[0041] Fig.12 Schematic diagram of prediction error of the prediction model in the present invention. DETAILED DESCRIPTION

[0042] like Figure 1 As shown, a method for predicting the temperature of blind areas of electric-assisted incremental forming based on a generative adversarial network in this embodiment solves the problem that the existing technology cannot measure the temperature distribution of the blocked blind areas of electric-assisted incremental forming infrared thermal imaging, predicts the temperature distribution in the blocked blind areas, and facilitates the precise control of the forming temperature to ensure the quality of the final product. Using the image completion problem in image processing, a generative adversarial network is used to predict the temperature of the blind areas of electric-assisted incremental forming for the thermal imaging obtained by the infrared thermal imager. The following steps are included:

[0043] Step 1: Through the finite element simulation of electric-assisted single-point incremental forming and the electric-assisted single-point incremental forming experiment, the training data set and the test data set for the training of the electric-assisted incremental forming blind area temperature prediction model are constructed, including the blind area temperature map where the thermal imaging image of the sheet metal processing area is blocked, and the non-blind area temperature map where the thermal imaging image of the sheet metal processing area is not blocked. In the electric-assisted single-point incremental forming experiment, infrared thermal imagers are set on the same side (front side) where the sheet metal contacts the tool and the opposite side (back side) where the sheet metal contacts the tool, and the blind area temperature map and the non-blind area temperature map are obtained respectively.

[0044] Step 2: Construct a generative adversarial network model, whose input is the blind spot temperature map and whose output is the temperature map without blind spots; obtain the blind spot temperature prediction model by training the acquired blind spot temperature map and the temperature map without blind spots training data set. The generative adversarial network model includes a generative network and an adversarial network. The generative network is used to generate the temperature map without blind spots where the blind spots are not blocked according to the blind spot temperature map where the blind spots of the electric-assisted incremental forming are blocked; the adversarial network is used to determine whether the input temperature map without blind spots is a real temperature map or a machine-generated temperature map. The generative network and the adversarial network are alternately trained through the training data set. When the training reaches a certain number of times, the training is stopped to obtain the trained model; the trained model is tested with the test data set. If the test results meet the requirements, the model is used as the final electric-assisted incremental forming blind spot temperature prediction model; if the test results do not meet the requirements, the network parameters are adjusted, and the generative network and the adversarial network are retrained through the training data set.

[0045] like Figure 5As shown, the generation process of the generation network is specifically as follows: the blind spot temperature map is cropped, scaled, and the blind spot is eliminated to obtain a feature map F1, and the feature map F1 is passed through a two-dimensional convolution module to obtain a feature map F2; the feature map F2 is passed through a hole convolution module to obtain a feature map F3; the feature map F3 is passed through a deconvolution module to obtain a feature map F4; the feature map F4 is spliced ​​with the feature map F1 to obtain a feature map F5, and the feature map F5 is used as the generated temperature map without blind spots.

[0046] like Figure 6 As shown in the figure, the two-dimensional convolution module includes five parts connected in sequence. The first part includes a 5×5 convolution layer and a pooling layer with a stride of 1×1; the second and fourth parts include a 3×3 convolution layer and a pooling layer with a stride of 2×2; the third and fifth parts include a 3×3 convolution layer and a pooling layer with a stride of 1×1.

[0047] The use of a 2×2 convolution in a two-dimensional convolution module helps reduce the resolution of the temperature map to compress information. This enhances the efficiency of calculations in subsequent processing and produces better denoising effects.

[0048] like Figure 7 As shown in the figure, the dilated convolution module includes four parts connected in sequence. The first part includes a 3×3 dilated convolution layer and a pooling layer with a stride of 1×1 and a dilation rate of 2 in sequence; the second part includes a 3×3 dilated convolution layer and a pooling layer with a stride of 1×1 and a dilation rate of 4 in sequence; the third part includes a 3×3 dilated convolution layer and a pooling layer with a stride of 1×1 and a dilation rate of 8 in sequence; the fourth part includes a 3×3 dilated convolution layer and a pooling layer with a stride of 1×1 and a dilation rate of 16 in sequence.

[0049] The use of dilated convolution in the dilated convolution module enables the network to increase the receptive field, while trying to obtain image information in a wider range without losing additional information. The use of dilated convolution with different dilated rates allows the network to perceive the characteristics of images at different scales, while obtaining richer information, it helps to understand the texture structure of different scales in noisy images. This makes the entire network model more versatile and can be applied to temperature prediction tasks in scenarios with different process parameters.

[0050] like Figure 8 As shown in the figure, the deconvolution module includes six parts connected in sequence. The first, fourth and sixth parts include a 3×3 convolution layer and a pooling layer with a stride of 1×1 in sequence; the second part includes a 5×5 convolution layer and a pooling layer with a stride of 1×1 in sequence; the third and fifth parts include a 4×4 deconvolution layer and a pooling layer with a stride of 1 / 2×1 / 2 in sequence.

[0051] like Fig. 9As shown in Figure 1, the adversarial network consists of five 5×5 convolutional layers and pooling layers with a stride of 2×2, connected sequentially, and a fully connected layer.

[0052] Step 3: Use the blind spot temperature prediction model to predict the blind spot temperature map to achieve blind spot temperature prediction. Fig.10 The real-time infrared temperature image of the electric-assisted incremental forming shown in FIG. 1 is input into the blind zone temperature prediction model to obtain the following Fig.11 As shown in the figure, the real-time temperature image of the unobstructed blind area of ​​the electric-assisted incremental forming, the prediction error of the prediction model is as follows Fig.12 shown.

Claims

1. A blind area temperature prediction method for electric-assisted incremental forming based on generative adversarial networks, characterized in that: The following steps are involved: Step 1: obtaining a blind area temperature map of a sheet material processing area where the thermal imaging image is blocked during the sheet material electric-assisted incremental forming, and obtaining a blind area temperature map of the sheet material processing area where the thermal imaging image is not blocked; obtaining the blind area temperature map of the sheet material processing area where the thermal imaging image is not blocked includes obtaining the blind area temperature map through electric-assisted single-point incremental forming finite element simulation, and obtaining the blind area temperature map by setting an infrared thermal imager on the opposite side of the sheet material contacting the tool through the electric-assisted single-point incremental forming experiment; Step 2: construct a generative adversarial network model, whose input is a blind spot temperature map and whose output is a temperature map without blind spots; obtain a blind spot temperature prediction model by training the blind spot temperature map and the temperature map without blind spots obtained in step 1; the generative adversarial network model includes a generative network and an adversarial network, and the generative network is used to generate a temperature map without blind spots according to the blind spot temperature map; the adversarial network is used to determine whether the input is a real temperature map without blind spots or a temperature map without blind spots generated by the generative network, and the generative network and the adversarial network are alternately trained to obtain a blind spot temperature prediction model; The generation process of the generation network is specifically as follows: The blind area temperature map is cropped, scaled, and the blind area is eliminated to obtain the feature map F1; The feature map F1 is processed by the two-dimensional convolution module to obtain the feature map F2; Feature map F2 is processed by the dilated convolution module to obtain feature map F3; Feature map F3 is processed by the deconvolution module to obtain feature map F4; The characteristic graph F4 is concatenated with the characteristic graph F1 to obtain the characteristic graph F5, and the characteristic graph F5 is used as the generated temperature graph without blind areas; Step 3: The blind spot temperature map to be predicted is used to predict the blind spot temperature through the blind spot temperature prediction model to realize the blind spot temperature prediction.

2. The method for predicting temperature in blind areas of electrically assisted incremental forming according to claim 1, characterized in that: The acquired blind spot temperature map and non-blind spot temperature map are divided into a training data set and a test data set. The generative network and the adversarial network are alternately trained through the training data set. When the training reaches a preset number of times, the training is stopped to obtain a trained model. The trained model is tested through the test data set. If the test result meets the requirements, the trained model is used as the blind spot temperature prediction model. If the test results do not meet the requirements, the parameters of the generative network and the adversarial network are adjusted and retrained using the training data set.

3. The method for predicting temperature in blind areas of electrically assisted incremental forming according to claim 2, characterized in that: The two-dimensional convolution module includes a first 5×5 convolution pooling layer with a step size of 1×1, a second 3×3 convolution pooling layer with a step size of 2×2, a third 3×3 convolution pooling layer with a step size of 1×1, a fourth 3×3 convolution pooling layer with a step size of 2×2, and a fifth 3×3 convolution pooling layer with a step size of 1×1, which are connected in sequence.

4. The method for predicting temperature in blind area of ​​electric assisted incremental forming according to claim 2, characterized in that: The dilated convolution module includes a 3×3 first dilated convolution pooling layer with a stride of 1×1 and a dilation rate of 2, a 3×3 second dilated convolution pooling layer with a stride of 1×1 and a dilation rate of 4, a 3×3 third dilated convolution pooling layer with a stride of 1×1 and a dilation rate of 8, and a 3×3 fourth dilated convolution pooling layer with a stride of 1×1 and a dilation rate of 16, which are connected in sequence.

5. The method for predicting temperature in blind areas of electrically assisted incremental forming according to claim 2, characterized in that: The deconvolution module includes a first 3×3 convolution pooling layer with a step size of 1×1, a second 5×5 convolution pooling layer with a step size of 1×1, a third 4×4 deconvolution pooling layer with a step size of 1 / 2×1 / 2, a fourth 3×3 convolution pooling layer with a step size of 1×1, a fifth 4×4 deconvolution pooling layer with a step size of 1 / 2×1 / 2, and a sixth 3×3 convolution pooling layer with a step size of 1×1.

6. The method for predicting temperature in blind areas of electric-assisted incremental forming according to claim 4, characterized in that: The adversarial network includes five 5×5 convolutional pooling layers with a stride of 2×2 connected in sequence and a fully connected layer.

7. An electric-assisted incremental forming blind zone temperature prediction system based on generative adversarial networks, characterized in that: include: A data acquisition module, used for acquiring a temperature map of a blind area where a thermal imaging image of a sheet material processing area is blocked during electric-assisted incremental forming of the sheet material, and acquiring a temperature map of a non-blind area where the thermal imaging image of the sheet material processing area is not blocked; the acquisition of a temperature map of a non-blind area where the thermal imaging image of the sheet material processing area is not blocked includes acquiring a temperature map of a non-blind area through finite element simulation of electric-assisted single-point incremental forming, and acquiring a temperature map of a non-blind area by setting an infrared thermal imager on the opposite side where the sheet material contacts the tool through an electric-assisted single-point incremental forming experiment; A model building module is used to build a generative adversarial network model, whose input is a blind spot temperature map and whose output is a temperature map without blind spots; a blind spot temperature prediction model is obtained by training the acquired blind spot temperature map and the temperature map without blind spots; the generative adversarial network model includes a generative network and an adversarial network, the generative network is used to generate a temperature map without blind spots according to the blind spot temperature map; the adversarial network is used to determine whether the input is a real temperature map without blind spots or a temperature map without blind spots generated by the generative network, and the generative network and the adversarial network are alternately trained to obtain a blind spot temperature prediction model; The generation process of the generation network is specifically as follows: The blind area temperature map is cropped, scaled, and the blind area is eliminated to obtain the feature map F1; The feature map F1 is processed by the two-dimensional convolution module to obtain the feature map F2; Feature map F2 is processed by the dilated convolution module to obtain feature map F3; Feature map F3 is processed by the deconvolution module to obtain feature map F4; The characteristic graph F4 is concatenated with the characteristic graph F1 to obtain the characteristic graph F5, and the characteristic graph F5 is used as the generated temperature graph without blind areas; The prediction module is used to predict the blind spot temperature through the blind spot temperature prediction model.

Images