A method for predicting temperature of products in the transfer stage before plastic processing based on digital twins

Through digital twin technology and neural network, the temperature of heat-treated workpieces is predicted and controlled in real time, which solves the problem of temperature not meeting requirements during transportation and improves product quality and energy efficiency.

CN116776675BActive Publication Date: 2024-11-26YANSHAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310663216.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-06-06
Publication Date
2024-11-26
Estimated Expiration
2043-06-06

AI Technical Summary

Technical Problem

Before the product is plasticized, the product temperature cannot be monitored and controlled in real time during the transfer process, resulting in the temperature not meeting the requirements and affecting product quality and energy efficiency.

Method used

Using a digital twin method, a temperature prediction model is established through a finite element model and neural network, combining temperature sensors, speed sensors and position sensors to predict the temperature when the heat-treated workpiece arrives at the processing table in real time, and make a decision based on the prediction results. .

Benefits of technology

Accurate prediction and control of product temperature is achieved, ensuring that the temperature before processing meets the requirements, improving product quality and energy utilization efficiency, and reducing unnecessary heat treatment links.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116776675B_ABST
    Figure CN116776675B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for predicting the temperature of a product in the transfer stage before plastic processing based on digital twins, which belongs to the field of heat treatment technology and includes the following steps: step 10, establishing a finite element model of a heat-treated workpiece; step 20, training the temperature prediction model and verifying the temperature prediction model using training data; step 30, establishing a digital twin system; step 40, predicting the temperature of the heat-treated workpiece when the transfer device reaches the processing position by using the trained temperature prediction model. The present invention establishes a finite element model based on the principle of thermodynamics and optimizes the model by comparing it with the measured data. The ambient temperature, time and minimum temperature of the product obtained by finite element simulation are used as training data. A prediction model is established by machine learning (including but not limited to neural networks). The prediction model is combined with software and hardware to establish a digital twin system, which can predict the temperature of the product in the transfer stage before plastic processing in real time and make intelligent decisions on the transfer behavior.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of heat treatment, and relates to a method for temperature prediction and transportation behavior decision-making in a transportation stage before plastic processing of a product. Background Art

[0002] The production process of forgings and rolled products is generally heat treatment, transportation, forging / rolling. In the entire process of product production, temperature is an important parameter. Too high temperature will cause problems such as coarse grains, overburning and overheating, which will damage the quality of the product. Too low temperature during forging / rolling will cause problems such as difficult processing and cracking. The problem of too high temperature can be solved by optimizing the heat treatment process and controlling the maximum temperature during heat treatment. In order to solve the problem of too low rolling / forging temperature, it is necessary to control the temperature at the beginning of rolling / forging so that it is not lower than the temperature allowed at the beginning of processing.

[0003] In actual production, it is impossible to know whether the temperature of the product meets the requirements during the transfer process before plastic processing and when the product arrives at the processing table after transfer, so it is impossible to control product quality and waste energy. Summary of the invention

[0004] The problem to be solved by the present invention is to provide a method for temperature prediction in the transfer stage before plastic processing of products based on digital twins. Through finite element models, neural networks, digital twins and other technologies, the temperature value of the heat-treated workpiece when it arrives at the processing table is predicted, thereby deciding the behavior of the transfer device.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: a method for predicting the temperature of the product in the transfer stage before plastic processing based on digital twin, comprising the following steps: step 10, establishing a finite element model of a heat-treated workpiece, and obtaining the temperature values ​​of the heat-treated workpiece at different transfer times under different transfer ambient temperatures by inputting different transfer ambient temperatures and the furnace-out temperatures of the heat-treated workpiece into the finite element model; step 20, taking different transfer ambient temperatures and the corresponding different transfer times obtained by the finite element model in step 10 as input parameters, and the temperature values ​​of the heat-treated workpiece at different transfer times under different transfer ambient temperatures as output parameters, the input parameters and the output parameters constitute training data, and the training data is used to train the temperature prediction model and verify the temperature prediction model; step 30, establishing a digital twin system by means of a temperature sensor for collecting the temperature in the workshop, a speed sensor for collecting the transfer speed, a position sensor for detecting the distance between the transfer device and the processing position, and a background server; step 40, obtaining the ambient temperature in the workshop, the transfer speed of the transfer device, and the distance between the transfer device and the processing table, and calculating the transfer time through the background server, and then predicting the temperature of the heat-treated workpiece when the transfer device arrives at the processing position through the trained temperature prediction model.

[0006] Furthermore, in step 10, the simulation time in the finite element model is t, the range of the simulated transfer environment temperature is from the lowest temperature to the highest temperature of the production workshop throughout the year and the simulation is performed with a gradient of 3°C, and the value range of t is 0-7200s.

[0007] Furthermore, measured temperature values ​​of at least five positions in the heat-treated workpiece are collected and compared with simulated temperature values ​​of corresponding positions in the finite element model, and the finite element model is optimized until the difference between the simulated temperature value and the measured temperature value is less than 2%.

[0008] Furthermore, in step 20, the input parameters and the output parameters constitute the training data and the training data is normalized, and 85% of the training data is divided into a training set and 15% as a test set, wherein the training set is used to train the temperature prediction model, and the test set is used to detect the accuracy of the temperature prediction model.

[0009] Further, in step 20, the temperature prediction model includes a neural network or a BP neural network. When a neural network is used, the hidden layer in the neural network is 2 layers, the input layer has 2 nodes, the hidden layer has 20 nodes, the transfer function is a sigmoid and ReLU function, the gradient descent algorithm is a stochastic gradient descent algorithm, the loss function is a mean square error loss function, the initial learning rate is 0.1, the normalization method is not limited to the maximum and minimum normalization method, the learning rate adopts a decay strategy, and the number of training times epoach is 1000000 times.

[0010] Further, in step 40, all sensors collect data every 5 seconds, and after the transfer time is calculated by the background server, the transfer environment temperature in the current workshop and the calculated transfer time are input into the trained temperature prediction model to predict the temperature of the heat-treated workpiece when the transfer device arrives at the processing position, and then judge whether the temperature of the heat-treated workpiece meets the requirements when the heat-treated workpiece arrives at the processing position: if the heat-treated workpiece is transferred at the current travel speed and is less than the specified temperature when it arrives at the processing position, it is judged whether the speed of the transfer device has reached the maximum transfer speed. If so, the transfer device drives the heat-treated workpiece back to the heat treatment furnace for heat treatment. Otherwise, the arrival time is calculated according to the maximum transfer speed, the arrival temperature is predicted, and it is judged whether the temperature of the heat-treated workpiece is greater than or equal to the specified temperature when it arrives at the processing position. If so, the transfer is carried out at the maximum speed. Otherwise, the transfer device drives the heat-treated workpiece back to the heat treatment furnace for heat treatment; if the heat-treated workpiece is transferred at the current travel speed and is greater than or equal to the specified temperature when it arrives at the processing position, the transfer device moves at the current travel speed.

[0011] The present invention establishes a finite element model based on the principles of thermodynamics and optimizes the model by comparing it with measured data. The ambient temperature, time and minimum temperature of the product obtained by finite element simulation are used as training data, and a prediction model is established through machine learning (including but not limited to neural networks). The prediction model is combined with software and hardware to establish a digital twin system, which can predict the temperature of the product in the transfer stage before plastic processing in real time and make intelligent decisions on the transfer behavior.

[0012] The present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 Flowchart of the present invention;

[0014] Figure 2 Digital twin system composition diagram;

[0015] Figure 3 Working principle diagram of digital twin system;

[0016] Figure 4 The product temperature-time curves of the selected 7 working conditions at an ambient temperature of 280.15-298.15K of the embodiment of the present invention;

[0017] Figure 5 Comparison curve between the neural network prediction result and the actual result in the embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present invention provides a method for predicting the temperature of a product in a transportation stage before plastic processing based on digital twin, which comprises the following steps.

[0019] Step 10, see attached Figure 1 , a finite element model of the heat-treated workpiece is established, and the temperature values ​​of the heat-treated workpiece at different transfer times under different transfer ambient temperatures are obtained by inputting different transfer ambient temperatures and the furnace-out temperatures of the heat-treated workpiece into the finite element model.

[0020] The temperature transfer process of the heat-treated workpiece during transportation can be considered as a natural convection process. The factors affecting the temperature of the heat-treated workpiece mainly include the initial temperature of the heat-treated workpiece, the ambient temperature, and the transportation time.

[0021] The finite element model needs to establish two areas, the solid area of ​​the heat-treated workpiece and the fluid area that wraps the heat-treated workpiece. The size of the fluid area is 10 times the size of the solid area. The initial temperature of the workpiece, that is, the furnace temperature, is the temperature of the heat-treated workpiece when it leaves the heat treatment furnace according to the actual production process on site in the production workshop. The incompressible model and k-ε model are used. The pressure adopts the body force weight method and the transient calculation method. The range of simulation time t is 0-7200s, and the selected t is much larger than the actual transportation time of the production workshop. The initial temperature of the heat-treated workpiece in the finite element model needs to be determined according to the actual production process on site in the production workshop, which is the temperature of the heat-treated workpiece when it leaves the heat treatment furnace. The ambient temperature is the lowest temperature to the highest temperature of the production workshop throughout the year, and the simulation is performed with a gradient of 3°C. Select one of the working conditions for simulation, obtain the temperature values ​​of 5 different positions on the surface of the heat-treated workpiece at different times under this working condition, and then conduct experimental verification to obtain the actual measured values ​​of these 5 positions at different times under this working condition, compare the actual measured values ​​with the temperature values ​​obtained by simulation, and optimize the finite element model until the difference between the temperature value obtained by simulation and the temperature value obtained by actual measurement is less than 2%, at which time the model optimization is completed; use the optimized finite element model to simulate all the working conditions of the design, and obtain the minimum temperature value of the heat-treated workpiece at different times under different ambient temperatures. The time here is from 0s, with an interval of 0.2s.

[0022] Step 20: different transfer environment temperatures and corresponding different transfer moments obtained by the finite element model in step 10 are used as input parameters, and the temperature values ​​of the heat-treated workpieces at different transfer environment temperatures and different transfer moments are used as output parameters. The input parameters and the output parameters constitute training data, and the training data is used to train the temperature prediction model and verify the temperature prediction model.

[0023] In this step, the input parameters and output parameters form the training data and the training data is normalized. 85% of the training data is divided into a training set and 15% as a test set. The training set is used to train the temperature prediction model, and the test set is used to test the accuracy of the temperature prediction model. The temperature prediction model is trained with the training set, and the temperature prediction model is verified with the verification set until the temperature difference between the temperature prediction value and the test set is less than 2°C.

[0024] The temperature prediction model can use a neural network or a BP neural network. When a neural network is used, the hidden layer in the neural network is 2 layers, the input layer has 2 nodes, the hidden layer has 20 nodes, the transfer function is sigmoid and ReLU functions, the gradient descent algorithm is a stochastic gradient descent algorithm, the loss function is a mean square error loss function, the initial learning rate is 0.1, the normalization method is not limited to the maximum and minimum normalization method, the learning rate adopts a decay strategy, and the number of training times epoach is 1000000 times.

[0025] Step 30, see attached Figure 2 A digital twin system is established through 6 temperature sensors installed in the workshop for collecting the temperature in the workshop, a speed sensor installed on the transfer device for collecting the transfer speed, a position sensor for detecting the distance between the transfer device and the processing position, and a background server.

[0026] Step 40, see attached Figure 3 , obtain the ambient temperature in the workshop, the transfer speed of the transfer device, and the distance between the transfer device and the processing table, and calculate the transfer time through the background server. Then, the temperature of the heat-treated workpiece when the transfer device arrives at the processing position is predicted through the trained temperature prediction model. The next behavior of the transfer device is determined based on the comparison result between the predicted temperature and the required temperature.

[0027] In detail, all sensors collect data every 5 seconds. After the transfer time is calculated by the background server, the transfer environment temperature in the current workshop and the calculated transfer time are input into the trained temperature prediction model to predict the temperature of the heat-treated workpiece when the transfer device arrives at the processing position, and then determine whether the temperature of the heat-treated workpiece meets the requirements when the heat-treated workpiece arrives at the processing position:

[0028] If the heat-treated workpiece is transported at the current speed and the temperature is less than the specified temperature when it arrives at the processing position, it is determined whether the speed of the transport device has reached the maximum transport speed. If so, the transport device drives the heat-treated workpiece back to the heat treatment furnace for heat treatment. Otherwise, the arrival time is calculated according to the maximum transport speed, the arrival temperature is predicted, and it is determined whether the temperature of the heat-treated workpiece is greater than or equal to the specified temperature when it arrives at the processing position. If so, it is transported at the maximum speed. Otherwise, the transport device drives the heat-treated workpiece back to the heat treatment furnace for heat treatment.

[0029] If the heat treatment workpiece is transported at the current travel speed and the temperature reaches the processing position, the transport device moves to the processing position at the current travel speed.

[0030] See attached Figure 4 and 5 The present invention is described in detail below by taking a titanium alloy cylindrical billet as an example. In the finite element model, the furnace temperature is 1273.15K, the simulation time is 600s, and the ambient temperature is 280.15-298.15K.

[0031] (1) According to the temperature range, the working conditions of 280.15K, 283.15K, 286.15K, 289.15K, 292.15K, 295.15K and 298.15K were selected, and the data corresponding to the minimum temperature and time of the product under the seven working conditions were calculated by the finite element model.

[0032] (2) The input parameters are the ambient temperature and the time interval of 0.2s starting from 0. The output parameter is the minimum temperature of the titanium alloy cylindrical billet corresponding to the current time. The data corresponding to the minimum temperature of the seven working conditions and the time and ambient temperature are converted into a tabular data file. These data are used as training data.

[0033] The training data is divided into a training set and a validation set. The training set is 85% of the total data, and the test set is 15% of the total data. That is, the data of the six working conditions with ambient temperatures of 280.15K, 283.15K, 289.15K, 292.15K, 295.15K, and 298.15K are used as the training set, and the data of the working condition with an ambient temperature of 286.15K is used as the test set. The training data is normalized using the maximum-minimum normalization method. The formula of the maximum-minimum normalization method is as follows: a =(xx min ) / (x max -x min ), where x a is the normalized value, x is the value to be normalized, and x max is the largest value in the original number set, x min is the smallest value in the original set.

[0034] A neural network model was established with 2 hidden layers, 2 nodes in the input layer, 20 nodes in the hidden layer, sigmoid and ReLU functions as the transfer functions, stochastic gradient descent (SGD) as the gradient descent algorithm, mean square error loss function as the loss function, the initial learning rate was 0.1, the learning rate adopted a decay strategy, and the number of training times was 1,000,000.

[0035] The neural network model is trained with the training set and trained repeatedly until the set training target is reached and the difference between the predicted value and the simulated value is less than 2°C.

[0036] Use the test set to test the prediction accuracy of the neural network. Figure 5 It can be seen that the prediction performance of the neural network is good, with a maximum error of 1.76K in 0-1.6s and a maximum error of less than 1K in 1.6-600s.

[0037] (3) Build a digital twin system solution based on the prediction model and software and hardware facilities (temperature sensors, speed sensors, position sensors, computing platforms, clients, transfer devices, etc.). The twin system consists of two parts: software and hardware. The software includes various software, systems and prediction models required by the computer. The hardware includes temperature sensors, speed sensors, position sensors, transfer devices, clients and other hardware facilities. The temperature sensor is used to obtain the workshop temperature, the speed sensor is used to obtain the travel speed of the transfer device, and the position sensor is used to obtain the distance between the loading device and the processing position. The time required to reach the processing position is equal to the current distance obtained divided by the current travel speed of the transfer device plus the transfer time.

[0038] (4) Based on the digital twin system, the temperature of the transfer process is predicted and the transfer behavior is decided. The working path is as follows: 6 temperature sensors obtain the workshop temperature, remove the extreme values, take the average value and write it into the computer table file. The transfer speed and the distance between the transfer device and the processing table obtained by the speed sensor and the position sensor are also written into the local table of the computer. A program is written to calculate the transfer time. The total transfer time = distance / transfer speed + used transfer time. The total transfer time is written into the table file where the workshop temperature is located. The obtained ambient temperature and transfer time are normalized. The sensor collects data every 5 seconds. The digital twin system determines whether the product temperature meets the requirements when it arrives at the processing table every 5 seconds. The prediction model reads the ambient temperature and the total transfer time to output the product temperature at the time of arrival. This temperature value needs to be denormalized to determine whether the predicted temperature is greater than or equal to the specified temperature. If so, the product is transferred at the current speed. If it is less than the specified temperature, the product is transferred at the maximum transfer speed. In the next 5 seconds, it is determined whether the product temperature is greater than or equal to the specified temperature when it arrives at the processing table. If so, it is transferred at the maximum speed. If it is lower than the specified temperature, it is returned to the heating furnace for heat treatment.

[0039] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or some technical features can be replaced by equivalents without departing from the spirit of the technical solution of the present invention, which should be included in the scope of the technical solution for protection of the present invention.

Claims

1. A method for predicting the temperature of products in the transfer stage before plastic processing based on digital twins, It is characterized in that The following steps are involved: Step 10, establishing a finite element model for the heat-treated workpiece, and obtaining temperature values ​​of the heat-treated workpiece at different transporting times under different transporting ambient temperatures by inputting different transporting ambient temperatures and the out-of-furnace temperatures of the heat-treated workpiece into the finite element model; Step 20, using different transfer environment temperatures and corresponding different transfer times obtained by the finite element model in step 10 as input parameters, and the temperature values ​​of the heat-treated workpiece at different transfer environment temperatures and different transfer times as output parameters, the input parameters and the output parameters constitute training data, and using the training data to train the temperature prediction model and verify the temperature prediction model; Step 30: Establish a digital twin system by using a temperature sensor for collecting the temperature in the workshop, a speed sensor for collecting the transfer speed, a position sensor for detecting the distance between the transfer device and the processing position, and a background server; Step 40, obtain the ambient temperature in the workshop, the transfer speed of the transfer device, and the distance between the transfer device and the processing table, and calculate the transfer time through the background server, and then predict the temperature of the heat-treated workpiece when the transfer device arrives at the processing position through the trained temperature prediction model.

2. The method for predicting the temperature of a product in the transportation stage before plastic processing based on digital twin according to claim 1, It is characterized in that In step 10, the simulation time in the finite element model is t, the range of the simulated transfer environment temperature is from the lowest temperature to the highest temperature of the production workshop throughout the year, and the simulation is performed with a gradient of 3°C, and the value range of t is 0-7200s.

3. The method for predicting the temperature of a product in the transportation stage before plastic processing based on digital twin according to claim 1, It is characterized in that At least five measured temperature values ​​at positions in the heat-treated workpiece are collected and compared with the simulated temperature values ​​at corresponding positions in the finite element model, and the finite element model is optimized until the difference between the simulated temperature value and the measured temperature value is less than 2%.

4. The method for predicting the temperature of a product in the transportation stage before plastic processing based on digital twin according to claim 1, It is characterized in that In step 20, the input parameters and the output parameters constitute the training data and the training data is normalized, and 85% of the training data is divided into a training set and 15% as a test set, wherein the training set is used to train the temperature prediction model, and the test set is used to detect the accuracy of the temperature prediction model.

5. The method for predicting the temperature of a product in the transportation stage before plastic processing based on digital twin according to claim 1, It is characterized in that In step 20, the temperature prediction model includes a neural network or a BP neural network.

6. The method for predicting the temperature of the product in the transportation stage before plastic processing based on digital twin according to claim 5, It is characterized in that The neural network has 2 hidden layers, 2 nodes in the input layer, 20 nodes in the hidden layer, the transfer function is sigmoid and ReLU function, the gradient descent algorithm is the stochastic gradient descent algorithm, the loss function is the mean square error loss function, the initial learning rate is 0.1, the normalization method is not limited to the maximum and minimum normalization method, the learning rate adopts a decay strategy, and the number of training times epoach is 1000000 times.

7. The method for predicting the temperature of a product in the transportation stage before plastic processing based on digital twin according to any one of claims 1 to 6, It is characterized in that In step 40, all sensors collect data every 5 seconds. After the transfer time is calculated by the background server, the transfer environment temperature in the current workshop and the calculated transfer time are input into the trained temperature prediction model to predict the temperature of the heat-treated workpiece when the transfer device arrives at the processing position, and then determine whether the temperature of the heat-treated workpiece meets the requirements when the heat-treated workpiece arrives at the processing position: If the heat-treated workpiece is transported at the current speed and the temperature is less than the specified temperature when it arrives at the processing position, it is determined whether the speed of the transport device has reached the maximum transport speed. If so, the transport device drives the heat-treated workpiece back to the heat treatment furnace for heat treatment. Otherwise, the arrival time is calculated according to the maximum transport speed, the arrival temperature is predicted, and it is determined whether the temperature of the heat-treated workpiece is greater than or equal to the specified temperature when it arrives at the processing position. If so, it is transported at the maximum speed. Otherwise, the transport device drives the heat-treated workpiece back to the heat treatment furnace for heat treatment. If the heat-treated workpiece is transported at the current travel speed and the temperature is greater than or equal to the specified temperature when it reaches the processing position, the transport device moves at the current travel speed.