Workpiece local heat control method based on deep learning

Abstract

The invention discloses a workpiece local heat control method based on deep learning, and the method comprises the following steps: 1, preparing neural network data: employing a thermal imager to shoot a detected object, and obtaining an overall temperature distribution diagram of a system; step 2, training a neural network: training a dual-output deep learning model by taking the control parameter v as input and taking the labels (Ts1, Ts2) as output; according to the scheme, a deep learning model with a seven-layer network structure is adopted; and step 3, randomly generating control parameters, and selecting optimal parameters according to a prediction result of the neural network: during small-range temperature regulation and control, firstly recording the temperatures Ts1, pre and Ts2, pre of the s1 region and the s2 region at the moment. According to the workpiece local heat control method based on deep learning, an innovative control strategy is provided, control parameters are optimized through a neural network, precise regulation and control of the temperature of a specific area are achieved, meanwhile, the temperature stability of other areas is effectively maintained, and therefore the requirement for high-precision temperature control in the die casting and injection molding process is met.

Description

Technical Field

[0001] The present invention relates to the technical field of workpiece heat control, and specifically to a workpiece local heat control method based on deep learning. Background Art

[0002] In the die-casting and injection molding production processes, temperature control is crucial to improving production efficiency, ensuring product quality and equipment safety, especially the distribution of workpiece and mold surface temperature. Due to the differences in temperature distribution of each workpiece and mold, it is often necessary to quickly and accurately adjust the temperature of certain areas to ensure the stability of the production process and the consistency of the product. However, in a complex production environment, how to accurately adjust the control parameters based on the temperature data collected by the thermal imager to ensure that the temperature of a specific area meets the requirements while maintaining the stability of the temperature in other areas is still a challenge. At the same time, with the development of artificial intelligence and machine learning technologies, adaptive control methods based on neural networks have gradually become a research hotspot in the field of temperature control.

[0003] Traditional temperature control methods usually rely on experience or rule control, but when faced with complex temperature distribution and multi-variable interactions, it is often difficult to provide sufficient accuracy and flexibility. For example, in some high-precision die-casting production, the temperature control accuracy of traditional methods can only reach ±5°C, while the actual production accuracy required is ±2°C, which leads to a product defective rate of more than 10%. At the same time, traditional workpiece local heat control methods mainly rely on empirical formulas and manual adjustments, which are difficult to achieve precise control, and there are also problems such as low control accuracy, slow response speed, and poor adaptability to workpiece shape and material. Therefore, we propose a workpiece local heat control method based on deep learning to solve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide a local heat control method for a workpiece based on deep learning, so as to solve the problem that the traditional temperature control method proposed in the above background technology usually relies on experience or rule control, but when faced with complex temperature distribution and the interaction of multiple variables, it is often difficult to provide sufficient accuracy and flexibility. At the same time, the traditional local heat control method of the workpiece mainly relies on empirical formulas and manual adjustment, which is difficult to achieve precise control, and there are also problems such as low control accuracy, slow response speed, and poor adaptability to workpiece shape and material.

[0005] To achieve the above object, the present invention provides the following technical solution: A method for local heat control of a workpiece based on deep learning, comprising the following steps: Step 1, preparing neural network data: using a thermal imager to photograph the object to be inspected to obtain the overall temperature distribution diagram of its system; Step 2, training the neural network: taking the control parameter v as input, and using its label (T s1 , T s2) as output, train a dual-output deep learning model; this scheme recommends using a 7-layer network structure deep learning model; Step 3, randomly generate control parameters, and select the optimal parameters based on the prediction results of the neural network: When performing small-scale temperature control, first record the temperature T of the s1 area and the s2 area at this time s1，pre and T s2，pre , and then randomly generate 1000 sets of control parameters v random = {v 1 , v 2 , ..., v 1000}, these parameters are random within a given control range and cover possible temperature control scenarios; Step 4, thermal control strategy formulation: feedforward control based on model prediction, feedback control and closed-loop optimization, and intelligent decision-making and multi-objective optimization; Step 5, real-time monitoring and adjustment: monitoring system construction, anomaly detection and processing, and online optimization and update.

[0006] Preferably, in step 1, the thermal imager is aimed at a local area of ​​the workpiece to be measured, ensuring that the lens is kept at an appropriate distance from the target to obtain a clear thermal image, and the shooting button of the thermal imager is pressed to collect a thermal image of the local area of ​​the workpiece. At this time, the thermal imager will automatically record the temperature data of each point in the image, and at the same time, according to specific needs, the thermal image is divided into two areas: area s1 and area s2, area s1 is an area where the temperature needs to be precisely controlled; area s2 is an area where a stable temperature is expected; on this basis, the average temperature values ​​T of the two areas are calculated respectively. s1 and T s2 , and use it as the temperature value of each area; at the same time, record the corresponding hot and cold water pipe control parameter set: v = {(q i , p i , t i ,θ i )|i∈{1, 2, ..., n}} where q represents flow rate, p represents pressure, t represents temperature, θ represents valve opening, and n represents the number of hot and cold water pipes involved in the regulation.

[0007] Preferably, the flow, pressure, temperature and valve opening of each hot and cold water pipe in step 1 will affect the temperature distribution of the target area; since multiple hot and cold water pipes work together in the area, the temperature of the target area is the result of the comprehensive effect of the control parameters of these pipes; therefore, under different control parameter combinations, the temperature changes of the target area are recorded, and 20,000 sets of data are collected to generate the corresponding data set: where v i represents the i-th group of control parameters, and They respectively represent the temperature values ​​of the s1 area and the s2 area corresponding to the i-th group.

[0008] Preferably, the following architecture is adopted in the step 2: the first layer: convolution layer, including 30 one-dimensional convolution kernels of size 3, and the activation function is ReLU; the second layer: convolution layer, including 60 one-dimensional convolution kernels of size 6, and the activation function is ReLU; the third layer: pooling layer, using maximum pooling, the kernel size is 2, and the step size is 3; the fourth layer: convolution layer, including 120 one-dimensional convolution kernels of size 3, and the activation function is ReLU; the fifth layer: fully connected layer, the output is 128 neurons, and the activation function is ReLU; the sixth layer: fully connected layer, the output is 2 neurons, and the activation function is linear.

[0009] Preferably, during the training process of step 2, the loss function is defined as follows: Depending on the prediction effect of the model on the test set, it may be necessary to adjust the network structure and optimizer to further improve the model performance and ultimately complete the training of the neural network.

[0010] Preferably, in step 3, the randomly generated control parameter v random Input the trained neural network model to get the corresponding temperature prediction value T s1，post and T s2，post , select the control parameter v that minimizes the following objective function opt : Using v opt Adjust the working state of the control system so that the temperature of the s1 area is as close to the desired target value as possible, while keeping the temperature of the s2 area stable.

[0011] Preferably, in step four, the trained deep learning model is used to predict the temperature distribution of the workpiece at a future moment, and the control parameters such as the power of the heating equipment and the flow rate of the cooling system are adjusted in advance according to the prediction results to achieve early intervention of the local heat of the workpiece; the actually measured temperature data is compared with the model prediction results, the deviation value is calculated, and the control parameters are adjusted according to the deviation value to form a closed-loop feedback system, so that the local temperature of the workpiece is always kept within the target range through continuous feedback and adjustment; considering multiple optimization objectives, such as ensuring that the local temperature of the workpiece meets the process requirements while minimizing energy consumption and equipment loss, the deep learning model is combined with a multi-objective optimization algorithm, such as a genetic algorithm, a particle swarm optimization algorithm, etc., to formulate an optimal heat control strategy.

[0012] Preferably, in step five, a real-time monitoring platform is established to transmit the data collected by the temperature sensor to the monitoring center in real time, and the temperature distribution of the workpiece, control parameter changes and other information are displayed in a graphical interface, so that the operator can intuitively understand the thermal state of the workpiece; the temperature data is analyzed in real time using a deep learning model to promptly detect abnormal temperature changes, such as local overheating, excessive temperature fluctuations, etc. Once an abnormality is detected, the system automatically issues an alarm and takes corresponding treatment measures according to preset strategies, such as reducing heating power, increasing cooling intensity, etc.; as the production process proceeds, new temperature data and control effect data are continuously collected, and the deep learning model is updated and optimized online, so that the model can adapt to the influence of factors such as changes in workpiece material properties and equipment aging, and maintain good thermal control performance.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: the local heat control method of the workpiece based on deep learning proposes an innovative control strategy, which optimizes the control parameters through neural networks to achieve precise control of the temperature in a specific area, while effectively maintaining the temperature stability of other areas, thereby meeting the requirements of die casting and injection molding processes for high-precision temperature control. The specific contents are as follows:

[0014] (1) In a complex production environment, the control system needs to adjust the flow, pressure and other control parameters of multiple hot and cold water pipes. Optimizing these control parameters through neural networks can effectively avoid excessive heating or cooling, thereby reducing energy waste. The intelligently optimized temperature control system can automatically adjust according to specific temperature requirements, avoiding excessive operation or repeated adjustments, reducing energy consumption in the production process, and indirectly reducing production costs.

[0015] (2) High-precision temperature control can ensure the temperature stability of the mold and workpiece surface, avoiding product defects caused by temperature fluctuations, such as deformation and cracks. The temperature control system based on neural network can monitor and optimize the temperature distribution in real time, ensuring that each production batch of products maintains consistency in temperature control, thereby significantly improving product quality and precision, reducing scrap rate, and ensuring that each workpiece in the production process meets high standards.

[0016] (3) Use the trained deep learning model to predict the temperature distribution of the workpiece at a future time. According to the prediction results, adjust the control parameters such as the power of the heating equipment and the flow rate of the cooling system in advance to achieve early intervention of the local heat of the workpiece. Compare the actual measured temperature data with the model prediction results, calculate the deviation value, and adjust the control parameters according to the deviation value to form a closed-loop feedback system. Through continuous feedback and adjustment, the local temperature of the workpiece is always kept within the target range.

[0017] (4) Establish a real-time monitoring platform to transmit the data collected by the temperature sensor to the monitoring center in real time, and use a graphical interface to display information such as the temperature distribution of the workpiece and changes in control parameters, so that operators can intuitively understand the thermal status of the workpiece. At the same time, use a deep learning model to analyze the temperature data in real time and detect abnormal temperature changes in a timely manner, such as local overheating and excessive temperature fluctuations. Once an abnormality is detected, the system automatically issues an alarm and takes corresponding treatment measures according to the preset strategy, such as reducing the heating power and increasing the cooling intensity. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the local heat control process of the workpiece of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] See also Figure 1 The present invention provides a technical solution: a method for controlling local heat of a workpiece based on deep learning, comprising the following steps: Step 1, preparing neural network data: using a thermal imager to photograph the object to be tested, and obtaining the overall temperature distribution diagram of its system. In step 1, aiming the thermal imager at a local area of ​​the workpiece to be tested, ensuring that the lens and the target are kept at an appropriate distance to obtain a clear thermal image, and pressing the shooting button of the thermal imager to collect a thermal image of the local area of ​​the workpiece, at this time, the thermal imager will automatically record the temperature data of each point in the image, and at the same time, according to specific needs, divide the thermal image into two areas: area s1 and area s2, area s1 is the area where the temperature needs to be precisely controlled; area s2 is the area where it is expected to maintain a stable temperature; on this basis, the average temperature values ​​T of the two areas are calculated respectively. s1 and T s2 , and use it as the temperature value of each area; at the same time, record the corresponding hot and cold water pipe control parameter set: v = {(q i , p i , t i ,θ i)|i∈{1, 2, ..., n}}where q represents flow, p represents pressure, t represents temperature, θ represents valve opening, and n represents the number of hot and cold water pipes involved in the regulation. The flow, pressure, temperature and valve opening of each hot and cold water pipe in step 1 will affect the temperature distribution of the target area. Since multiple hot and cold water pipes work together in this area, the temperature of the target area is the result of the combined effect of the control parameters of these pipes. Therefore, under different combinations of control parameters, the temperature changes of the target area are recorded, and 20,000 sets of data are collected to generate the corresponding data set: where v i represents the i-th group of control parameters, and Respectively represent the temperature values ​​of the s1 region and the s2 region corresponding to the i-th group; Step 2, training the neural network: take the control parameter v as input, and use its label (T s1 , T s2 ) is the output, and a dual-output deep learning model is trained; this solution recommends a deep learning model with a 7-layer network structure. The following architecture is used in step 2: the first layer: convolution layer, containing 30 one-dimensional convolution kernels of size 3, and the activation function is ReLU; the second layer: convolution layer, containing 60 one-dimensional convolution kernels of size 6, and the activation function is ReLU; the third layer: pooling layer, using maximum pooling, kernel size of 2, and step size of 3; the fourth layer: convolution layer, containing 120 one-dimensional convolution kernels of size 3, and the activation function is ReLU; the fifth layer: fully connected layer, the output is 128 neurons, and the activation function is ReLU; the sixth layer: fully connected layer, the output is 2 neurons, and the activation function is linear. During the training process of step 2, the loss function is defined as follows: According to the prediction effect of the model on the test set, it may be necessary to adjust the network structure and optimizer to further improve the model performance and finally complete the training of the neural network; Step 3: Randomly generate control parameters and select the optimal parameters according to the prediction results of the neural network: When performing small-scale temperature control, first record the temperature T of the s1 area and the s2 area at this time s1，pre and T s2，pre , and then randomly generate 1000 sets of control parameters v random = {v 1 , v 2 , ..., v 1000}, these parameters are random within the given control range and cover possible temperature control scenarios. In step 3, the randomly generated control parameters v random Input the trained neural network model to get the corresponding temperature prediction value T s1，post and T s2，post , select the control parameter v that minimizes the following objective function opt : Using vopt Adjust the working state of the control system so that the temperature of the s1 area is as close to the desired target value as possible, while keeping the temperature of the s2 area stable; Step 4, formulation of heat control strategy: feedforward control, feedback control and closed-loop optimization, intelligent decision-making and multi-objective optimization based on model prediction. In step 4, the trained deep learning model is used to predict the temperature distribution of the workpiece at the future moment. According to the prediction results, the power of the heating equipment, the flow rate of the cooling system and other control parameters are adjusted in advance to achieve early intervention of the local heat of the workpiece; the actual measured temperature data is compared with the model prediction results, the deviation value is calculated, and the control parameters are adjusted according to the deviation value to form a closed-loop feedback system. Feedback and adjustment are performed to keep the local temperature of the workpiece within the target range. Multiple optimization objectives are considered, such as ensuring that the local temperature of the workpiece meets the process requirements while minimizing energy consumption and equipment loss. The deep learning model is combined with multi-objective optimization algorithms, such as genetic algorithms and particle swarm optimization algorithms, to develop the optimal heat control strategy. Step 5: Real-time monitoring and adjustment: monitoring system construction, anomaly detection and processing, and online optimization and update. In step 5, a real-time monitoring platform is established to transmit the data collected by the temperature sensor to the monitoring center in real time, and a graphical interface is used to display the temperature distribution of the workpiece, control parameter changes and other information, so that operators can intuitively understand the thermal status of the workpiece. Using deep The deep learning model analyzes temperature data in real time and promptly detects abnormal temperature changes, such as local overheating and excessive temperature fluctuations. Once an abnormality is detected, the system automatically issues an alarm and takes corresponding treatment measures according to the preset strategy, such as reducing heating power and increasing cooling intensity. As the production process proceeds, new temperature data and control effect data are continuously collected, and the deep learning model is updated and optimized online so that the model can adapt to the influence of factors such as changes in workpiece material properties and equipment aging, and maintain good heat control performance. Therefore, in a complex production environment, the control system needs to adjust the control parameters such as flow and pressure of multiple hot and cold water pipes. Through the neural network The network optimizes these control parameters, which can effectively avoid overheating or cooling, thereby reducing energy waste. The intelligently optimized temperature control system can automatically adjust according to specific temperature requirements to avoid over-operation or repeated adjustments, reducing energy consumption in the production process and indirectly reducing production costs. High-precision temperature control can ensure the temperature stability of the mold and workpiece surface, avoiding product defects caused by temperature fluctuations, such as deformation and cracks. The temperature control system based on the neural network can monitor and optimize the temperature distribution in real time to ensure that each production batch of products maintains consistency in temperature control, thereby significantly improving product quality and precision, reducing scrap rate, and ensuring that each workpiece in the production process meets high standards.

[0021] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for controlling local heat of a workpiece based on deep learning, characterized in that: The following steps are involved: Step 1: Prepare neural network data: Use a thermal imager to shoot the object to be tested and obtain the overall temperature distribution map of its system; Step 2: Train the neural network: Take the control parameter v as input and use its label (T s1 , T s2 ) as output, train a dual-output deep learning model; this scheme recommends using a 7-layer network structure deep learning model; Step 3, randomly generate control parameters, and select the optimal parameters based on the prediction results of the neural network: When performing small-scale temperature control, first record the temperature T of the s1 area and the s2 area at this time s1，pre and T s2，pre , and then randomly generate 1000 sets of control parameters v random = {v1, v2, ..., v 1000 }, these parameters are random within a given control range and cover possible temperature control scenarios; Step 4, thermal control strategy formulation: feedforward control based on model prediction, feedback control and closed-loop optimization, and intelligent decision-making and multi-objective optimization; Step 5, real-time monitoring and adjustment: monitoring system construction, anomaly detection and processing, and online optimization and update.

2. The method for controlling local heat of a workpiece based on deep learning according to claim 1, characterized in that: In the step 1, the thermal imager is aimed at the local area of ​​the workpiece to be measured, ensuring that the lens and the target are kept at an appropriate distance to obtain a clear thermal image, and the shooting button of the thermal imager is pressed to collect the thermal image of the local area of ​​the workpiece. At this time, the thermal imager will automatically record the temperature data of each point in the image, and at the same time, according to specific needs, the thermal image is divided into two areas: area s1 and area s2. Area s1 is the area where the temperature needs to be precisely controlled; area s2 is the area where the temperature is expected to be kept stable; on this basis, the average temperature values ​​T of the two areas are calculated respectively. s1 and T s2 , and use it as the temperature value of each area; at the same time, record the corresponding hot and cold water pipe control parameter set: v = {(q i , p i , t i ,θ i )|i∈{1, 2, ..., n}} where q represents flow rate, p represents pressure, t represents temperature, θ represents valve opening, and n represents the number of hot and cold water pipes involved in the regulation.

3. The method for controlling local heat of a workpiece based on deep learning according to claim 1, characterized in that: The flow, pressure, temperature and valve opening of each hot and cold water pipe in step 1 will affect the temperature distribution of the target area; since multiple hot and cold water pipes work together in the area, the temperature of the target area is the result of the comprehensive effect of the control parameters of these pipes; therefore, under different control parameter combinations, the temperature changes of the target area are recorded, and 20,000 sets of data are collected to generate the corresponding data set: where v i represents the i-th group of control parameters, and They respectively represent the temperature values ​​of the s1 area and the s2 area corresponding to the i-th group.

4. The method for controlling local heat of a workpiece based on deep learning according to claim 1, characterized in that: The following architecture is used in step 2: the first layer: a convolutional layer, including 30 one-dimensional convolution kernels of size 3, and the activation function is ReLU; The second layer: convolution layer, containing 60 one-dimensional convolution kernels of size 6, and the activation function is ReLU; the third layer: pooling layer, using maximum pooling, kernel size is 2, and step size is 3; the fourth layer: convolution layer, containing 120 one-dimensional convolution kernels of size 3, and the activation function is ReLU; the fifth layer: fully connected layer, the output is 128 neurons, the activation function is ReLU; the sixth layer: fully connected layer, the output is 2 neurons, and the activation function is linear.

5. The method for controlling local heat of a workpiece based on deep learning according to claim 1, characterized in that: During the training process of step 2, the loss function is defined as follows: Depending on the prediction effect of the model on the test set, it may be necessary to adjust the network structure and optimizer to further improve the model performance and ultimately complete the training of the neural network.

6. The method for controlling local heat of a workpiece based on deep learning according to claim 1, characterized in that: In step 3, the randomly generated control parameter v random Input the trained neural network model to get the corresponding temperature prediction value T s1，post and T s2，post , select the control parameter v that minimizes the following objective function opt : Using v opt Adjust the working state of the control system so that the temperature of the s1 area is as close to the desired target value as possible, while keeping the temperature of the s2 area stable.

7. The method for controlling local heat of a workpiece based on deep learning according to claim 1, characterized in that: In step 4, the trained deep learning model is used to predict the temperature distribution of the workpiece at a future moment, and the control parameters such as the power of the heating equipment and the flow rate of the cooling system are adjusted in advance according to the prediction results to achieve early intervention of the local heat of the workpiece; the actually measured temperature data is compared with the model prediction results, the deviation value is calculated, and the control parameters are adjusted according to the deviation value to form a closed-loop feedback system. Through continuous feedback and adjustment, the local temperature of the workpiece is always kept within the target range; considering multiple optimization objectives, such as ensuring that the local temperature of the workpiece meets the process requirements while minimizing energy consumption and equipment loss, the deep learning model is combined with a multi-objective optimization algorithm, such as a genetic algorithm, a particle swarm optimization algorithm, etc., to develop an optimal heat control strategy.

8. The method for controlling local heat of a workpiece based on deep learning according to claim 1, characterized in that: In the step five, a real-time monitoring platform is established to transmit the data collected by the temperature sensor to the monitoring center in real time, and the temperature distribution of the workpiece, the change of control parameters and other information are displayed in a graphical interface, so that the operator can intuitively understand the thermal status of the workpiece; the temperature data is analyzed in real time by using a deep learning model to detect abnormal temperature changes in time, such as local overheating, excessive temperature fluctuations, etc. Once an abnormality is detected, the system automatically issues an alarm and takes corresponding treatment measures according to the preset strategy, such as reducing the heating power, increasing the cooling intensity, etc.; as the production process proceeds, new temperature data and control effect data are continuously collected, and the deep learning model is updated and optimized online, so that the model can adapt to the influence of factors such as changes in the material properties of the workpiece and aging of the equipment, and maintain good thermal control performance.

Images