An automatic cooling method for die casting system based on deep learning

Abstract

The present invention discloses an automatic cooling method for a die-casting system based on deep learning. By introducing a convolutional neural network (CNN) and deep learning, feature extraction is performed on the thermal imaging image of the mold surface after the metal liquid solidifies to form a product and the thermal imaging image of the mold surface after the cooling system is finished working, so that the original input and output of the cooling system are changed from image form to mathematical matrix form, and the loss function can realize the calculation of the input and output images, and then obtain the ideal values ​​of the water flow rate and water flow time. In the present invention, the water flow rate and water flow time in the cooling system are determined by the CNN neural network to achieve an ideal cooling effect. Compared with the empirical method, the method of the present invention is more flexible in the adjustment process in actual production; the purpose of scientifically controlling the cooling system is achieved, and the goal of extending the service life of the mold and reducing the processing error can be achieved.

Description

Technical Field

[0001] The present invention relates to the field of die-casting technology, and in particular to an automatic cooling method for a die-casting system based on deep learning. Background Art

[0002] The die-casting process is a precision casting method that uses three major elements, namely machines, molds and alloys, to unify pressure, speed and time. It uses high pressure to force molten metal into a complex-shaped metal mold.

[0003] Die casting has the characteristics of high production precision, high material utilization rate and high production efficiency, and can manufacture metal parts with complex shapes, clear contours, thin walls and deep cavities, so it is increasingly widely used.

[0004] After the metal liquid is injected into the die casting, the mold temperature rises at the same time, and then the metal liquid solidifies to form the product, and the mold temperature also drops. In this process, due to the complex shape of the mold cavity and the large temperature difference between different positions of the mold, the mold is prone to micro deformation, causing micro cracks, resulting in mold failure and unqualified products. Although the existing mold is generally equipped with a cooling system consisting of several cooling paths to ensure that different positions of the mold can be cooled evenly, due to the lack of effective control, the current product qualification rate is not high and the mold life is short.

[0005] As for the cooling system of die-casting process, its cooling effect is determined by two factors: water flow rate and water flow time. In traditional cooling systems, in order to achieve the ideal cooling effect, the empirical method is adopted, and the water flow rate and water flow time are gradually changed according to the cooling effect in each actual production process. This empirical method often cannot quantitatively analyze the scientific water flow rate and water flow time, and will cause damage to the mold during the test process, affecting product processing. Summary of the invention

[0006] The purpose of the present invention is to provide an automatic cooling method for a die-casting system based on deep learning, which can solve one or more of the above-mentioned technical problems.

[0007] In order to achieve the above object, the technical solution proposed by the present invention is as follows:

[0008] A deep learning-based automatic cooling method for a die-casting system, comprising:

[0009] (1) The water valves in the cooling system of the die-casting machine are numbered from valve 1 to valve N; the water flow rate of each water valve is set to a1...a N , the water flow time of each water valve is b1...b N ;

[0010] (2) Establish a CNN convolutional neural network training database;

[0011] (21) The die-casting system is randomly executed T times. Each die-casting system has different water flow and water flow time. The water flow is Water flow time is

[0012] (22) Use a thermal imager to record the input image X1...X of each die casting system T , output image Y1...Y T ;

[0013] (23) According to the work of each die casting system, the corresponding water flow rate Water supply time Input image X1...X T , output image Y1...Y T Statistics form mold thermal image database;

[0014] (24) taking the cooling effect of the output image as a condition, selecting training data from the mold thermal image database of step (23) to form a training database, wherein the training data database includes water flow rates corresponding to each other Water supply time Input image X1...X M , output image Y1...Y M ;

[0015] (3) Training the CNN convolutional neural network

[0016] (31) Use CNN convolutional neural network to train the input image X1...X in the database in step (24) M With the output image Y1...Y M Perform feature extraction to generate training data and corresponding labels required for deep learning;

[0017] (32) setting a first loss function Loss_1 about the correlation between the input image, the output image, the water flow rate, and the water flow time, and training the CNN convolutional neural network in step (24) under the condition that the first loss function Loss_1 reaches a minimum to obtain the weights and biases of the CNN; to obtain the CNN1 convolutional neural network;

[0018] (4) Performing secondary training on the CNN1 convolutional neural network obtained in step (3);

[0019] (41) Establishment of secondary training database:

[0020] (411) Let the die-casting system perform L die-casting production again to obtain a new input image

[0021] (412) The output image Y1...Y in step (2) M The average output image is obtained from The average output image is the output image for secondary training;

[0022] (42) Taking the water flow rate and water flow time as unknown quantities, the input image in step (41) Average output image For known quantities, set the second loss function Loss_2 about the correlation between input and output images, output images, water flow rate, and water flow time;

[0023] (43) With the second loss function Loss_2 reaching its minimum value as a condition, the CNN1 convolutional neural network trained in step (3) is trained for the second time to obtain the CNN2 convolutional neural network, and the water flow a1...a is obtained. N The ideal value, water flow time b1...b N The ideal value of

[0024] (5) Using the water flow rate a1 ... a obtained in step (43) N And water flow time b1...b N , used in formal production.

[0025] Further: The process of generating the training data and corresponding labels required for deep learning in step (32) is: transform the input image X1...X M With the output image Y1...Y M The mold surface temperature distribution information contained in is extracted and converted into a one-dimensional array vector through a convolutional neural network.

[0026] Further: Here, the VGGNet network is used as the backbone network for feature extraction. The size of the input and output images is preprocessed to 224×224 pixels. The Layer1 convolution layer consists of 64 3×3 convolution kernels, using BN batch normalization, and the activation function is 'relu'; the Layer2 convolution layer consists of 64 3×3 convolution kernels, using BN batch normalization, the activation function is 'relu', and the maximum pooling with a step size of 2 is used; the Layer3 convolution layer consists of 128 3×3 convolution kernels, using BN batch normalization, and the activation function is The convolution layer of Layer 4 is composed of 128 convolution kernels of size 3×3, using BN batch standardization, the activation function is 'relu', and the maximum pooling with a step size of 2 is adopted; the convolution layer of Layer 5 is composed of 256 convolution kernels of size 3×3, using BN batch standardization, the activation function is 'relu'; the convolution layer of Layer 6 is composed of 256 convolution kernels of size 3×3, using BN batch standardization, the activation function is 'relu'; the convolution layer of Layer 7 is composed of 256 convolution kernels of size 1×, using BN batch standardization, the activation function is 'relu'. The number is 'relu', and the maximum pooling with a stride of 2 is used; the Layer8 convolution layer consists of 512 3×3 convolution kernels, using BN batch standardization, and the activation function is 'relu'; the Layer9 convolution layer consists of 512 3×3 convolution kernels, using BN batch standardization, and the activation function is 'relu'; the Layer10 convolution layer consists of 512 1×1 convolution kernels, using BN batch standardization, the activation function is 'relu', and the maximum pooling with a stride of 2 is used; the Layer11 convolution layer consists of 512 3×3 convolution kernels The convolution layer consists of 512 3×3 convolution kernels, uses BN batch standardization, and the activation function is 'relu'; the convolution layer consists of 512 1×1 convolution kernels, uses BN batch standardization, and the activation function is 'relu', and the maximum pooling with a step size of 2 is adopted; the fully connected layer consists of 512 neurons, and the activation function is 'relu'; the fully connected layer consists of 512 neurons, and the activation function is 'relu'.

[0027] Further: The training algorithm is the stochastic gradient descent algorithm, and the optimizer is the Adam algorithm.

[0028] Further: The first loss function in step (32) is: Loss_1 = ||A*CNN(X)-B-CNN(Y)|| 2 , where A is the water flow rate and B is the water flow time. X=[X1...X M ]、Y=[Y1...Y M ], CNN is a convolutional neural network.

[0029] Further: The second loss function in step (42) is: A is the water flow rate, B is the water flow time, is the average output image obtained in step (41), and CNN1 is the CNN1 convolutional neural network obtained in step (32).

[0030] The technical effects of the present invention are:

[0031] In the present invention, the water flow rate and water flow time in the cooling system are determined by the CNN neural network to achieve an ideal cooling effect. Compared with the empirical method, the method of the present invention is more flexible in the adjustment process in actual production; it achieves the purpose of scientifically controlling the cooling system, and can also achieve the goal of extending the service life of the mold and reducing the processing error. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings in the specification, which constitute a part of this application, are used to provide a further understanding of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0033] In the attached picture:

[0034] Figure 1 It is a schematic diagram of the overall structure of the present invention;

[0035] Figure 2 It is a flow chart of the working process of the present invention. DETAILED DESCRIPTION

[0036] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments, wherein the illustrative embodiments and descriptions are only used to explain the present invention but are not intended to be improper limitations of the present invention.

[0037] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0038] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0039] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0040] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0041] A deep learning-based automatic cooling method for a die-casting system, comprising:

[0042] (1) The water valves in the cooling system of the die-casting machine are numbered from valve 1 to valve N; the water flow rate of each water valve is set to a1...a N , the water flow time of each water valve is b1...b N ;

[0043] (2) Establish a CNN convolutional neural network training database;

[0044] (21) The die-casting system is randomly executed T times. Each die-casting system has different water flow and water flow time. The water flow is Water flow time is

[0045] (22) Use a thermal imager to record the input image X1...X of each die casting system T , output image Y1...Y T ;

[0046] (23) According to the work of each die casting system, the corresponding water flow rate Water supply time Input image X1...X T , output image Y1...Y T Statistics form mold thermal image database;

[0047] (24) taking the cooling effect of the output image as a condition, selecting training data from the mold thermal image database of step (23) to form a training database, wherein the training data database includes water flow rates corresponding to each other Water supply time Input image X1...X M , output image Y1...Y M .

[0048] (3) Training the CNN convolutional neural network

[0049] (31) Use CNN convolutional neural network to train the input image X1...X in the database in step (24) M With the output image Y1...Y M Perform feature extraction to generate training data and corresponding labels required for deep learning;

[0050] That is, the input image X1...X M With the output image Y1...Y M The mold surface temperature distribution information contained in is extracted and converted into a one-dimensional array vector through a convolutional neural network.

[0051] This method uses the VGGNet network as the backbone network for feature extraction, preprocesses the input and output images to 224×224 pixels, and the Layer1 convolution layer consists of 64 3×3 convolution kernels, using BN batch normalization, and the activation function is 'relu'; the Layer2 convolution layer consists of 64 3×3 convolution kernels, using BN batch normalization, the activation function is 'relu', and the maximum pooling with a step size of 2 is used; the Layer3 convolution layer consists of 128 3×3 convolution kernels, using BN batch normalization, and the activation function is 'relu'. lu'; Layer4 convolution layer consists of 128 3×3 convolution kernels, using BN batch standardization, the activation function is 'relu', and the maximum pooling with a step size of 2 is adopted; Layer5 convolution layer consists of 256 3×3 convolution kernels, using BN batch standardization, the activation function is 'relu'; Layer6 convolution layer consists of 256 3×3 convolution kernels, using BN batch standardization, the activation function is 'relu'; Layer7 convolution layer consists of 256 1× convolution kernels, using BN batch standardization, the activation function is 'relu', using the maximum pooling with a stride of 2; Layer8 convolution layer consists of 512 3×3 convolution kernels, using BN batch normalization, and the activation function is 'relu'; Layer9 convolution layer consists of 512 3×3 convolution kernels, using BN batch normalization, and the activation function is 'relu'; Layer10 convolution layer consists of 512 1×1 convolution kernels, using BN batch normalization, the activation function is 'relu', and the maximum pooling with a stride of 2; Layer11 convolution layer consists of 512 3×3 convolution kernels The convolution layer of Layer 12 is composed of 512 convolution kernels of size 3×3, and BN batch standardization is used, with the activation function of 'relu'; the convolution layer of Layer 13 is composed of 512 convolution kernels of size 1×1, and BN batch standardization is used, with the activation function of 'relu', and the maximum pooling with a step size of 2 is used; the fully connected layer of Layer 14 is composed of 512 neurons, with the activation function of 'relu'; the fully connected layer of Layer 15 is composed of 512 neurons, with the activation function of 'relu'.

[0052] The training algorithm is the stochastic gradient descent algorithm, and the optimizer is the Adam algorithm.

[0053] (32) Set the first loss function Loss_1 about the correlation between the input image, the output image, the water flow rate, and the water flow time. The first loss function is: Loss_1 = ||A*CNN(X)-B-CNN(Y)|| 2, where A is the water flow rate and B is the water flow time. In this step X=[X1...X M ]、Y=[Y1...Y M ], CNN is a convolutional neural network.

[0054] (33) The CNN convolutional neural network in step (24) is trained under the condition that the first loss function Loss_1 reaches the minimum to obtain the weights and biases of the CNN; to obtain the CNN1 convolutional neural network;

[0055] (4) Performing secondary training on the CNN1 convolutional neural network obtained in step (3);

[0056] (41) Establishment of secondary training database:

[0057] (411) Let the die-casting system perform L die-casting production again to obtain a new input image

[0058] (412) The output image Y1...Y in step (2) M The average output image is obtained from The average output image is the output image for secondary training;

[0059] (42) Taking the water flow rate and water flow time as unknown quantities, the input image in step (41) Average output image As known quantities, set the second loss function Loss_2 about the correlation between the input image, the output image, the water flow rate, and the water flow time; the second loss function in step (42) is:

[0060] in is the average output image obtained in step (31), CNN1 is the CNN1 convolutional neural network obtained in step (32); A is the water flow rate, and B is the water flow time.

[0061] (43) With the second loss function Loss_2 reaching its minimum value as a condition, the CNN1 convolutional neural network trained in step (3) is trained for the second time to obtain the CNN2 convolutional neural network, and the water flow a1...a is obtained. N The ideal value, water flow time b1...b N The ideal value of

[0062] (5) Using the water flow rate a1 ... a obtained in step (43) N And water flow time b1...b N , used in formal production.

[0063] In the present invention, by adjusting the water flow rate and water flow time of the cooling system, the cooling effect is ideal, and the cooling effect can be scientifically quantitatively analyzed to ensure that different positions of the mold can be evenly cooled, reduce point damage to the mold, and improve product processing quality.

[0064] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for automatic cooling of a die-casting system based on deep learning, characterized in that: include (1) The water valves in the cooling system of the die casting machine are numbered from valve 1 to valve N; the water flow rate of each water valve is set to a1...a N , the water flow time of each water valve is b1...b N ; (2) Establish a CNN convolutional neural network training database; (21) The die-casting system is randomly executed T times. Each die-casting system has different water flow and water flow time. The water flow is Water flow time is (22) Use a thermal imager to record the input image X1...X of each die casting system T , output image Y1...Y T ; (23) According to the work of each die casting system, the corresponding water flow rate Water supply time Input image X1...X T , output image Y1...Y T Statistics form mold thermal image database; (24) taking the cooling effect of the output image as a condition, selecting training data from the mold thermal image database of step (23) to form a training database, wherein the training data database includes water flow rates corresponding to each other Water supply time Input image X1...X M , output image Y1...Y M ; (3) Training the CNN convolutional neural network (31) Use CNN convolutional neural network to train the input image X1...X in the database in step (24) M With the output image Y1...Y M Perform feature extraction to generate training data and corresponding labels required for deep learning; (32) setting a first loss function Loss_1 about the correlation between the input image, the output image, the water flow rate, and the water flow time, and training the CNN convolutional neural network in step (24) under the condition that the first loss function Loss_1 reaches a minimum to obtain the weights and biases of the CNN; to obtain the CNN1 convolutional neural network; (4) Performing secondary training on the CNN1 convolutional neural network obtained in step (3); (41) Establishment of secondary training database: (411) Let the die-casting system perform L die-casting production again to obtain a new input image (412) The output image Y1...Y in step (2) M The average output image is obtained from The average output image is the output image for secondary training; (42) Taking the water flow rate and water flow time as unknown quantities, the input image in step (41) Average output image For known quantities, set the second loss function Loss_2 about the correlation among input image, output image, water flow rate and water flow time; (43) With the second loss function Loss_2 reaching its minimum value as a condition, the CNN1 convolutional neural network trained in step (3) is trained for the second time to obtain the CNN2 convolutional neural network, and the water flow a1...a is obtained. N The ideal value, water flow time b1...b N The ideal value of (5) Using the water flow rate a1 ... a obtained in step (43) N And water flow time b1...b N , used in formal production.

2. The automatic cooling method for die-casting system based on deep learning according to claim 1, characterized in that: The process of generating the training data and corresponding labels required for deep learning in step (32) is: transform the input image X1...X M With the output image Y1...Y M The mold surface temperature distribution information contained in is extracted and converted into a one-dimensional array vector through a convolutional neural network.

3. The automatic cooling method for a die-casting system based on deep learning according to claim 2, characterized in that: Here, the VGGNet network is used as the backbone network for feature extraction. The size of the input and output images is preprocessed to 224×224 pixels. The Layer1 convolution layer consists of 64 3×3 convolution kernels, which are processed by BN batch standardization and the activation function is 'relu'; the Layer2 convolution layer consists of 64 3×3 convolution kernels, which are processed by BN batch standardization and the activation function is 'relu', with a maximum pooling of 2 steps; the Layer3 convolution layer consists of 128 3×3 convolution kernels, which are processed by BN batch standardization and the activation function is 'relu'. lu'; Layer4 convolution layer consists of 128 3×3 convolution kernels, using BN batch standardization, the activation function is 'relu', and the maximum pooling with a step size of 2 is adopted; Layer5 convolution layer consists of 256 3×3 convolution kernels, using BN batch standardization, the activation function is 'relu'; Layer6 convolution layer consists of 256 3×3 convolution kernels, using BN batch standardization, the activation function is 'relu'; Layer7 convolution layer consists of 256 1× convolution kernels, using BN batch standardization, the activation function is 'relu', using the maximum pooling with a stride of 2; Layer8 convolution layer consists of 512 3×3 convolution kernels, using BN batch normalization, and the activation function is 'relu'; Layer9 convolution layer consists of 512 3×3 convolution kernels, using BN batch normalization, and the activation function is 'relu'; Layer10 convolution layer consists of 512 1×1 convolution kernels, using BN batch normalization, the activation function is 'relu', and the maximum pooling with a stride of 2; Layer11 convolution layer consists of 512 3×3 convolution kernels The convolution layer of Layer 12 is composed of 512 convolution kernels of size 3×3, and BN batch standardization is used, with the activation function of 'relu'; the convolution layer of Layer 13 is composed of 512 convolution kernels of size 1×1, and BN batch standardization is used, with the activation function of 'relu', and the maximum pooling with a step size of 2 is used; the fully connected layer of Layer 14 is composed of 512 neurons, with the activation function of 'relu'; the fully connected layer of Layer 15 is composed of 512 neurons, with the activation function of 'relu'.

4. The automatic cooling method for a die-casting system based on deep learning according to claim 3 is characterized in that: The training algorithm is the stochastic gradient descent algorithm, and the optimizer is the Adam algorithm.

5. The automatic cooling method for die casting system based on deep learning according to claim 1, characterized in that: The first loss function in step (32) is: Loss_1 = ||A*CNN(X)-B-CNN(Y)|| 2 ,in X=[X1...X M ]、Y=[Y1...Y M ], CNN is a convolutional neural network.

6. The automatic cooling method for die casting system based on deep learning according to claim 1, characterized in that: The second loss function in step (42) is: in is the average output image obtained in step (41), and CNN1 is the CNN1 convolutional neural network obtained in step (32).

Images