A method for controlling heat treatment of large castings based on CNN neural network model

Abstract

The present invention relates to a method for controlling the heat treatment of large castings based on a CNN neural network model, and belongs to the field of casting processing technology. The method comprises the steps of: constructing a CNN neural network model, setting the casting characteristics, water injection port location, water flow rate, and medium temperature data in the heat treatment system as input layer neurons, and the surface temperature gradient as output layer neurons; collecting the temperature gradient data corresponding to the input and output during the heat treatment of large castings as a training set for the neural network for learning and training; and after the training is completed, using the neural network model to predict and set heat treatment process parameters. The device comprises an equipment control module, a flow rate control module, a medium temperature control module, a temperature acquisition module, a feedback module, and a model database module. The method, combined with the device, can flexibly adjust the cooling rate during the heat treatment process, simplifying the process flow, shortening the production cycle, reducing production costs, and effectively implementing heat treatment parameters that meet the corresponding performance indicators of large castings.

Description

Technical Field

[0001] The present invention belongs to the technical field of casting processing, and in particular relates to a method for controlling heat treatment of large castings based on a CNN neural network model. Background Art

[0002] Precise control of quenching and cooling has long been one of the most critical issues in heat treatment processes. Adjusting the mechanical properties of castings by controlling the heat treatment process is a common practice among manufacturers. Traditional heat treatment processes are typically performed by process technicians based on their experience, resulting in a waste of materials, energy, and time. With the advancement of technology, process line designers are prioritizing intelligent, precise, and sustainable approaches. Predicting heat treatment process parameters through the establishment of neural network models has become a key research and application area in my country and many other countries.

[0003] Heat treatment, the second step in the casting process, is a crucial and crucial step in the entire casting manufacturing process. Precisely controlling heat treatment process parameters and ensuring temperature uniformity significantly impact the mechanical properties and quality of the final casting. Therefore, effectively predicting and modifying the process parameters required to achieve target mechanical properties during heat treatment has become a pressing technical challenge for major steel mills. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for controlling the heat treatment of large castings based on a CNN neural network model, aiming to solve the technical problem of predicting and correcting heat treatment process parameters in real time; constructing a CNN neural network model, setting the casting characteristics, water injection port position, water flow rate and medium temperature data in the heat treatment system as input layer neurons, and the surface temperature gradient as output layer neurons; collecting the temperature gradient data corresponding to the input and output during the heat treatment process of large castings as a training set for neural network learning and training; after the training is completed, the neural network model is used to predict and set the heat treatment process parameters.

[0005] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0006] (1) Construct a CNN neural network model and establish a multi-layer neural network topology through the convolution algorithm; use the casting characteristics, water injection port position, water flow rate and medium temperature in the heat treatment system as input neurons, and the surface temperature gradient as the output layer neuron; based on the established neural network topology, use the SoftMax method to perform modified linear unit classification on the input neurons, use Max Pooling as the weight to construct the CNN artificial neural network model, and use the momentum-adaptive learning rate adjustment algorithm to quickly learn and train the network; collect the temperature gradient data corresponding to the input and output during the heat treatment process of large castings as the training set of the neural network for learning and training; after training, the neural network model is used to predict and set the heat treatment process parameters.

[0007] (2) The process parameters obtained in step (1) are input into the equipment control module. The equipment control module uses a single-chip microcomputer system and is connected to the flow rate control module and the medium temperature control module respectively. The flow rate control module controls the medium flow rate by controlling the degree of closure of the electromagnetic switch; the medium temperature control module controls the medium temperature by controlling the heating of the heating coil and the cooling of the one-way refrigeration plate. Combined with the prediction parameters of the neural network model, real-time control is performed through the equipment control system.

[0008] (3) The temperature of large castings during the heat treatment process is collected through the temperature acquisition module (infrared thermal imager), and the data is sent to the feedback module in real time; the hot spots in the temperature gradient data are located, and the equipment control system is adjusted in real time to make the overall temperature gradient uniform and ensure the heat treatment quality of the castings; at the same time, the temperature gradient data is uploaded to the model database module, combined with the corresponding neuron input data to process it into a training set, and the neural network model is further trained to achieve a continuous optimization effect.

[0009] Preferably, the casting characteristics of the present invention include size specifications, content of main components, type of electric heating furnace used and holding time.

[0010] Preferably, the feedback module of the present invention performs grayscale correction on the temperature data graph collected by the temperature acquisition module, determines the hot spot position and feeds it back to the equipment control system, and adjusts the setting of process parameters in real time; the error detection and correction unit analyzes the influence of material properties, water injection port position, water flow rate and medium temperature process parameters on the temperature gradient graph, and uploads the data to the model database module.

[0011] Preferably, the model database module of the present invention processes the different hot spots that appear in the temperature gradient diagram due to changes in material properties, water injection port position, water flow rate and medium temperature process parameters into a training set, and further trains the neural network model to achieve a continuous optimization effect.

[0012] Beneficial effects of the present invention:

[0013] (1) The method of the present invention links the setting and correction of heat treatment process parameters, on-site production and neural network prediction models to form a closed-loop system, which supervises and corrects each other. At the same time, it provides accurate reference data and direction for the setting and adjustment of the heat treatment temperature model, thereby realizing autonomous regulation of the process line.

[0014] (2) The method of the present invention has been tested and found that the neural network can effectively predict the material properties achieved under different heat treatment parameters with a high prediction accuracy. It can make the material meet the material performance requirements after one heat treatment based on the set heat treatment parameters. Compared with the existing technology, it simplifies the process flow, shortens the production cycle, and reduces the production cost. It is a more effective prediction technology solution with certain practicality.

[0015] (3) The method of the present invention includes an equipment control module, a flow rate control module, a medium temperature control module, a temperature acquisition module, a feedback module, and a model database module. During the heat treatment process, the cooling rate can be flexibly adjusted, and heat treatment parameters that meet the corresponding performance indicators of large castings can be effectively implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings are used to provide a further understanding of the present invention. The schematic diagrams and descriptions of the present invention are used to explain the present invention and do not constitute an improper limitation of the present invention. Among them:

[0017] Figure 1 Schematic diagram of the topological structure of the multi-layer neural network model of the present invention;

[0018] Figure 2 A schematic diagram of a calculation method for determining weights using the neural network softmax method of the present invention;

[0019] Figure 3 A schematic diagram of the basic process of neural network learning of the present invention;

[0020] Figure 4 This is a schematic diagram of the basic operation process of the device of the present invention;

[0021] Figure 5 Schematic diagram of the device module of the present invention.

[0022] Figure 6 Temperature gradient diagram;

[0023] Figure 7 Adjusted temperature gradient map;

[0024] Figure 8 Homogenized Bema complex structure.

[0025] Among them, 1-temperature acquisition module; 2-equipment control module; 3-flow rate control module; 4-cooling medium tank; 5-medium temperature control module. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Therefore, the following detailed description of the embodiments of the present application is not intended to limit the scope of the application for protection, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0027] The heat treatment mentioned in the present invention refers to methods such as tempering and solid solution.

[0028] In the embodiment of the present invention, a convolution algorithm is used to establish a multi-layer neural network. The topological structure of the multi-layer neural network model is as follows: Figure 1 As shown in Figure 2; it consists of an input layer, a hidden layer, and an output layer.

[0029] Figure 2 The softmax method is used for neural network to classify data, and the results of multi-classification are presented in the form of probability. The converted results are normalized so that the neural network model can output the set data more quickly and accurately.

[0030] The temperature gradient data corresponding to the input and output during the heat treatment of large castings are collected as the training set of the neural network for learning and training. Figure 3 This is a basic flow chart of neural network learning. The material properties, water flow rate, water injection port position and medium temperature in the heat treatment cooling system are set as input neurons, and the temperature gradient data is set as output neurons. The hidden layer uses the softmax method to filter features and set weights to train the neural network model. The process parameters and corresponding temperature data in actual production are collected to establish a secondary training set to further train and optimize the neural network.

[0031] The device used in the method of the embodiment of the present invention is as follows Figure 5 As shown, the system includes a temperature acquisition module 1, a device control module 2, a flow rate control module 3, a cooling medium tank 4, and a medium temperature control module 5. The temperature acquisition module 1 is located above the quenching water tank, the flow rate control module 3 is located at one end of the water pipe, and the medium temperature control module 5 is located outside the water pipe and includes a heating coil and a one-way cooling fin. The device control module 2 is located on one side of the water tank and is connected to the flow rate control module and the medium temperature control module.

[0032] Specific implementation application:

[0033] The product that undergoes heat treatment is a chromium-molybdenum alloy casting liner with the following composition:

[0034] element C Si Mn Cr Ni Mo S P 100% (%) 0.458 0.433 0.513 3.438 0.628 0.307 0.021 0.012

[0035] 50 sample data of heat treatment temperature of lining plates with similar components were collected and divided into 5 groups according to the medium flow rate and temperature during heat treatment to establish a training set. The data were input into the neural network model and a multi-layer mapping convolutional network (convolution-excitation-pooling-full connection) was used. SoftMax normalization was performed and the momentum-adaptive learning rate adjustment algorithm was used to quickly train the network model. During the training process, the typical network parameters were selected as follows: the learning process display frequency d f =10, maximum number of training steps m e =8000 steps, error index e g =0.02, learning rate l r =0.08. After 852 training cycles, the network converged to an expected error of 0.02 and was able to predict process parameters.

[0036] The temperature gradient data for homogenizing the structure is set, and the trained neural network model is used to predict the process parameter requirements to achieve the target, specifically: water flow rate of 0.25 m / s, medium temperature of 35°C, and water injection port opened 1 / 4.

[0037] After waiting for the water flow to run smoothly, the casting liner is taken out of the trolley-type resistance furnace (300kw, 1600℃) at 860℃ and sent into the water pool. The infrared temperature collector collects temperature data of the casting liner in the water pool. The temperature gradient diagram is as follows: Figure 6 shown.

[0038] The temperature gradient map is transmitted to the feedback module, the hotspot of the temperature gradient data is located, and the position is fed back to the equipment control module. 1 / 2 of the water inlet is opened and the water flow rate is adjusted to 0.3m / s.

[0039] After adjustment, the temperature gradient image collected by the temperature collector is as follows Figure 7 The results show that the hot spot temperature is significantly reduced, ensuring the temperature uniformity of the casting.

[0040] The feedback module transmits the hotspot position and uploads the temperature gradient data to the model database module. After 10 data collections, it is processed into a secondary training set and the neural network is trained again to achieve a more accurate prediction model for the heat treatment process parameters of the subsequent batches of products. Finally, a uniform Bema complex phase structure is obtained, such as Figure 8 shown.

[0041] In this embodiment, the heat treatment process parameters are predicted by a neural network prediction model, the corresponding set parameters are controlled by the equipment control module, and the process parameters are timely adjusted during the process through the temperature acquisition and feedback module. The collected temperature data is used to perform secondary training on the neural network model to obtain more accurate process parameters. Analysis of the casting products that have been applied by the system shows that the structure is more uniform, which simplifies the entire process flow, reduces manual operation, and improves the production efficiency of the enterprise.

Claims

1. A method for controlling heat treatment of large castings based on a CNN neural network model, characterized in that: include: (1) Construct a CNN neural network model and establish a multi-layer neural network topology through the convolution algorithm; use the casting characteristics, water injection port position, water flow rate and medium temperature in the heat treatment system as input neurons, and the surface temperature gradient as the output layer neuron; Based on the established neural network topology, use the SoftMax method to perform modified linear unit classification on the input neurons, use Max Pooling as the weight to construct the CNN artificial neural network model, and use the momentum-adaptive learning rate adjustment algorithm to quickly learn and train the network; Collect the temperature gradient data corresponding to the input and output during the heat treatment process of large castings as the training set for neural network learning and training; After training, the neural network model is used to predict and set the heat treatment process parameters; (2) Inputting the process parameters obtained in step (1) into the equipment control module, the equipment control module uses a single-chip microcomputer system and is connected to the flow rate control module and the medium temperature control module respectively. The flow rate control module controls the medium flow rate by controlling the degree of closure of the electromagnetic switch; the medium temperature control module controls the medium temperature by controlling the heating of the heating coil and the cooling of the one-way refrigeration plate, and combines the prediction parameters of the neural network model to achieve real-time control through the equipment control system; (3) The temperature of large castings during the heat treatment process is collected through the temperature acquisition module, and the data is sent to the feedback module in real time; the hot spots in the temperature gradient data are located, and the equipment control system is adjusted in real time to make the overall temperature gradient uniform and ensure the heat treatment quality of the castings; at the same time, the temperature gradient data is uploaded to the model database module, combined with the corresponding neuron input data to process it into a training set, and the neural network model is further trained to achieve a continuous optimization effect.

2. The method for controlling heat treatment of large castings based on a CNN neural network model according to claim 1, characterized in that: The casting characteristics include size specifications, main component content, electric heating furnace model used and holding time.

3. The method for controlling heat treatment of large castings based on a CNN neural network model according to claim 1, characterized in that: The feedback module performs grayscale correction on the temperature data graph collected by the temperature acquisition module, determines the hotspot position and feeds it back to the equipment control system to adjust the setting of process parameters in real time; The error detection and correction unit analyzes the influence of material properties, water injection port position, water flow rate and medium temperature process parameters on the temperature gradient diagram, and uploads the data to the model database module.

4. The method for controlling heat treatment of large castings based on a CNN neural network model according to claim 1, characterized in that: The model database module processes the different hot spots that appear in the temperature gradient graph as material properties, water injection port position, water flow rate and medium temperature process parameters change into a training set, and further trains the neural network model to achieve continuous optimization effects.

Images