Intelligent optimization control system for coil curing process
By combining real-time infrared thermal imaging with deep learning, a multi-region adaptive control system was developed to solve the problem of uneven temperature distribution during the curing process of high-voltage transformer coils. This system achieved optimized temperature uniformity and energy efficiency, and improved insulation performance and equipment reliability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHANGCHUN YUEQUAN BIONIC TECHNOLOGY CO LTD
- Filing Date
- 2025-12-15
- Publication Date
- 2026-04-14
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional control methods are difficult to effectively address the nonlinear heat conduction during the curing process of high-voltage transformer coils, resulting in uneven temperature distribution and affecting insulation performance and long-term reliability.
A multi-region adaptive control system combining real-time infrared thermal imaging and deep learning is adopted. Through data acquisition, preprocessing, temperature prediction and optimization algorithms, the heater power is dynamically adjusted to achieve temperature uniformity.
It achieves precise temperature control during the coil curing process, reduces local overheating or underheating, improves insulation performance and long-term reliability, and reduces energy consumption and equipment heat load.
Smart Images
Figure CN121857291A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of coil curing technology, and specifically relates to an intelligent optimization control system for the coil curing process. Background Technology
[0002] The high-voltage transformer coil is the core component of the transformer. It is precisely wound from conductive windings and multiple layers of insulating materials and undertakes the key functions of electromagnetic energy conversion and voltage transformation. Its structure usually includes multiple coils, interlayer insulation and cooling channels. The whole coil is vacuum pressure impregnated and filled with insulating resin and then cured at high temperature to form a solid whole with high mechanical strength and stable electrical insulation performance. The uniformity and control precision of this curing process directly determine the long-term electrical insulation strength, mechanical stability and short-circuit withstand capability of the coil.
[0003] During the curing process of high-voltage transformer coils, the complex structure of the coil's internal multilayer insulation material interlaced with the conductor makes it difficult for traditional control methods based on fixed programs or simple feedback to effectively handle nonlinear heat conduction behavior. This leads to significant temperature unevenness on the coil surface and inside the coil during the curing stage. This unevenness can cause localized overheating or insufficient curing, resulting in thermal stress concentration and differences in curing degree in the insulation material. This directly affects the insulation performance and long-term reliability of the final product, becoming a key issue restricting the manufacturing of high-quality coils. Therefore, the following solutions are proposed to address these issues. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent optimization control system for the coil curing process. By combining real-time infrared thermal imaging with deep learning prediction in a multi-region adaptive control method, it is possible to achieve precise temperature control and dynamic optimization of the coil curing process, thus solving the problem of uneven temperature distribution caused by nonlinear heat conduction in existing technologies.
[0005] To solve the above-mentioned technical problems, the present invention provides an intelligent optimization control system for the coil curing process, the control system comprising: Data acquisition module: responsible for driving the infrared thermal imager to acquire raw temperature distribution data of the coil surface in real time; Data preprocessing module: Cleans, denoises, and performs coordinate mapping on the raw temperature data, transforming it into a regular and reliable temperature matrix; Temperature prediction module: Using a deep learning model, it predicts the temperature distribution of the coil at future times based on current temperature data, and anticipates changes in thermal behavior; Control decision module: Based on the predicted temperature distribution, the optimal power adjustment of each heater is calculated through an optimization algorithm to achieve the goal of uniform curing; Control execution module: Sends the power commands generated by the decision module to the heater for execution, and monitors the execution status in real time to ensure that the control is in place; System Management Module: As the central hub of the system, it is responsible for parameter configuration, model updates, process scheduling and security monitoring, ensuring the coordinated and stable operation of the entire system; The workflow of the control system is as follows: Step S1: System initialization: Load the pre-trained deep learning model, set the target temperature and other control parameters, and initialize the infrared thermal imager and multi-area heater controller. Step S2, Real-time data acquisition: Use an infrared thermal imager to acquire the surface temperature distribution of the coil, and perform preprocessing such as noise reduction and alignment; Step S3, Temperature Distribution Prediction: Input real-time temperature data into the model to predict the temperature distribution of each region of the coil at future times; Step S4, Control Decision Generation: Based on the predicted temperature distribution, the optimal power adjustment value for each heater is calculated using an optimization algorithm; Step S5, Control Execution: Send the power adjustment command to the heater controller, execute the power adjustment, and monitor the execution status; Step S6, Iterative Optimization and Termination: Repeat the above acquisition, prediction and control steps until solidification is complete, and update the model periodically to adapt to process changes.
[0006] Step S1, system initialization, specifically includes the following steps: Step S11: Load the pre-trained deep learning model, which is trained based on historical high-voltage transformer coil solidification data and is used to predict temperature distribution; the model input is the coil surface temperature matrix and the output is the temperature distribution at future time steps. Step S12: Set control parameters, including target curing temperature. Predicting time steps Heater power range and optimizing weight parameters ; Step S13: Initialize the infrared thermal imager and multi-zone heater controller to ensure normal communication between the device and the main control system; set the resolution of the infrared thermal imager to at least 640×480 pixels to cover coil surface details.
[0007] Step S2, real-time data acquisition, specifically includes the following steps: Step S21: Activate the infrared thermal imager and acquire real-time temperature distribution data of the high-voltage transformer coil surface at a rate of 1 frame per second; the data format is a two-dimensional matrix. ,in, For pixel coordinates, t Indicates the current time; Step S22: Preprocess the acquired temperature data, including noise reduction (using a median filter) and coordinate alignment, mapping the temperature matrix to the physical space region of the coil to ensure that each pixel corresponds to a specific heating control area.
[0008] Step S3, temperature distribution prediction, specifically includes the following steps: Step S31: The preprocessed real-time temperature matrix Input a pre-trained deep learning model, which is based on a convolutional neural network structure, and output the future time step. Predicted temperature distribution The prediction process takes into account the thermal conductivity characteristics of the coil material and historical temperature changes, and the model update frequency is synchronized with data acquisition. Step S32: Verify the rationality of the prediction results. By comparing the gradient change between the current temperature and the predicted temperature, if the predicted value exceeds the physical range (such as being lower than the ambient temperature or exceeding the material's heat resistance limit), then the backup interpolation algorithm is activated for correction.
[0009] Step S4, the control decision generation, specifically includes the following steps: Step S41: Based on the predicted temperature distribution Calculate the control adjustment for each heater zone and define the optimization objective function: In the formula, J The objective function value to be minimized in the optimization process. N This represents the number of temperature monitoring points (corresponding to pixel areas). The index variable is used for summation to iterate through all temperature monitoring points. For the first Predicted temperature values at each temperature monitoring point The preset, desired target curing temperature, These are the weighting coefficients. M For the number of heaters, j The index variable is used to sum and iterate through all heaters. P j For the first j The power of each heater; Step S42: Solve for min using the gradient descent algorithm. J The optimal power adjustment value for each heater is obtained. The solution process is constrained by Ensure that the power is within a safe range; Step S43: Adjust the power value It is converted into a control signal and formatted into a sequence of instructions that the heater controller can recognize.
[0010] Furthermore, step S5, the control execution specifically includes the following steps: Step S51: Send the generated control commands to the multi-zone heater controller via a communication interface (such as Modbus protocol) to adjust the power output of each heater in real time. P j ; Step S52: Monitor the execution results and verify the temperature change after power adjustment using an infrared thermal imager. If a communication failure or execution abnormality is detected, trigger an alarm and switch to safe mode.
[0011] Furthermore, step S6, loop optimization and termination, specifically includes the following steps: Step S61: Repeat steps S2 to S5 until the curing process reaches the preset total time (e.g., 30 minutes) or the temperature uniformity index stabilizes (e.g., the temperature difference between all areas and the target temperature is less than 100°C). ); Step S62: During the loop, the deep learning model is updated periodically (e.g., every 5 minutes) to fine-tune the model parameters using the latest acquired temperature data to adapt to the aging of the coil material or changes in the environment; Step S63: After the curing process is complete, save all control logs and temperature data for subsequent analysis and model optimization.
[0012] The present invention has the following beneficial effects: 1. This invention acquires coil surface temperature distribution data through real-time infrared thermal imaging and uses a deep learning model to predict future temperature changes, thereby dynamically adjusting the power of multi-zone heaters. This prediction-based control strategy can identify temperature unevenness trends in advance and calculate power adjustment amounts through optimization algorithms, ensuring that the temperature in each region of the coil always approaches the target value. Compared with traditional fixed-program control, this method effectively reduces local overheating or underheating, and avoids material stress concentration and insulation performance degradation caused by temperature gradients. The entire process does not rely on empirical parameters but achieves precise balance through data-driven methods, improving the stability and consistency of curing quality while reducing the risk of product defects caused by temperature unevenness.
[0013] 2. This invention integrates real-time data acquisition with a deep learning model fine-tuning mechanism, enabling it to continuously adapt to changes in coil material properties and environmental interference. During the curing process, the model periodically updates parameters to reflect the impact of material aging or fluctuations in external conditions, thereby dynamically optimizing control decisions. This adaptive capability allows the system to maintain stable performance even when facing non-uniform heat conduction or sudden disturbances, avoiding the failure problems caused by fixed control logic in traditional methods. Furthermore, the combination of predictive models and real-time feedback enhances the system's ability to identify and correct abnormal states early, strengthens the robustness of the overall process, and ensures long-term operational reliability.
[0014] 3. This invention balances temperature uniformity and energy consumption through an objective function, and introduces power constraints and optimization algorithms into the control decision-making process to dynamically calculate the minimum necessary power adjustment for each heater. This avoids the overheating or power waste common in traditional control, and achieves precise energy allocation. Because the system responds to temperature changes in advance based on a predictive model, unnecessary power fluctuations are reduced, thereby reducing overall energy consumption while maintaining curing quality. This energy efficiency optimization not only reduces operating costs but also alleviates equipment heat load and extends heater life, meeting the requirements of sustainable manufacturing. Attached Figure Description
[0015] Figure 1 This is a schematic diagram of the intelligent optimization control system for the coil curing process of the present invention. Detailed Implementation
[0016] Please see Figure 1 The image shown is an embodiment of the present invention.
[0017] This embodiment describes an intelligent optimization control system for the coil curing process. The control system includes: Data acquisition module: responsible for driving the infrared thermal imager to acquire raw temperature distribution data of the coil surface in real time; Data preprocessing module: Cleans, denoises, and performs coordinate mapping on the raw temperature data, transforming it into a regular and reliable temperature matrix; Temperature prediction module: Using a deep learning model, it predicts the temperature distribution of the coil at future times based on current temperature data, and anticipates changes in thermal behavior; Control decision module: Based on the predicted temperature distribution, the optimal power adjustment of each heater is calculated through an optimization algorithm to achieve the goal of uniform curing; Control execution module: Sends the power commands generated by the decision module to the heater for execution, and monitors the execution status in real time to ensure that the control is in place; System Management Module: As the central hub of the system, it is responsible for parameter configuration, model updates, process scheduling and security monitoring, ensuring the coordinated and stable operation of the entire system; The control system's workflow is as follows: Step S1: System initialization: Load the pre-trained deep learning model, set the target temperature and other control parameters, and initialize the infrared thermal imager and multi-area heater controller. Step S1, system initialization specifically includes the following steps: Step S11: Load the pre-trained deep learning model. The deep learning model is trained based on historical high-voltage transformer coil solidification data and is used to predict the temperature distribution. The model input is the coil surface temperature matrix, and the output is the temperature distribution at future time steps. Step S12: Set control parameters, including target curing temperature. Predicting time steps Heater power range and optimizing weight parameters ; Step S13: Initialize the infrared thermal imager and multi-zone heater controller to ensure normal communication between the device and the main control system; set the resolution of the infrared thermal imager to at least 640×480 pixels to cover coil surface details.
[0018] Step S2, Real-time data acquisition: Use an infrared thermal imager to acquire the surface temperature distribution of the coil, and perform preprocessing such as noise reduction and alignment; Step S2, real-time data acquisition specifically includes the following steps: Step S21: Activate the infrared thermal imager and acquire real-time temperature distribution data of the high-voltage transformer coil surface at a rate of 1 frame per second; the data format is a two-dimensional matrix. ,in, For pixel coordinates, Indicates the current time; Step S22: Preprocess the acquired temperature data, including noise reduction (using a median filter) and coordinate alignment, mapping the temperature matrix to the physical space region of the coil to ensure that each pixel corresponds to a specific heating control area.
[0019] Step S3, Temperature Distribution Prediction: Input real-time temperature data into the model to predict the temperature distribution of each region of the coil at future times; Step S3, temperature distribution prediction specifically includes the following steps: Step S31: The preprocessed real-time temperature matrix Input a pre-trained deep learning model, which is based on a convolutional neural network structure, and output the future time step. Predicted temperature distribution The prediction process takes into account the thermal conductivity characteristics of the coil material and historical temperature changes, and the model update frequency is synchronized with data acquisition. Step S32: Verify the rationality of the prediction results. By comparing the gradient change between the current temperature and the predicted temperature, if the predicted value exceeds the physical range (such as being lower than the ambient temperature or exceeding the material's heat resistance limit), then the backup interpolation algorithm is activated for correction.
[0020] Step S4, Control Decision Generation: Based on the predicted temperature distribution, the optimal power adjustment value for each heater is calculated using an optimization algorithm; Step S4, the control decision generation specifically includes the following steps: Step S41: Based on the predicted temperature distribution Calculate the control adjustment for each heater zone and define the optimization objective function: In the formula, The objective function value to be minimized in the optimization process. This represents the number of temperature monitoring points (corresponding to pixel areas). The index variable is used for summation to iterate through all temperature monitoring points. For the first Predicted temperature values at each temperature monitoring point The preset, desired target curing temperature, These are the weighting coefficients. For the number of heaters, The index variable is used to sum and iterate through all heaters. For the first The power of each heater; Step S42: Solve using the gradient descent algorithm The optimal power adjustment value for each heater is obtained. The solution process is constrained by Ensure that the power is within a safe range; Step S43: Adjust the power value It is converted into a control signal and formatted into a sequence of instructions that the heater controller can recognize.
[0021] Step S5, Control Execution: Send the power adjustment command to the heater controller, execute the power adjustment, and monitor the execution status; Step S5, the control execution specifically includes the following steps: Step S51: Send the generated control commands to the multi-zone heater controller via a communication interface (such as Modbus protocol) to adjust the power output of each heater in real time. ; Step S52: Monitor the execution results and verify the temperature change after power adjustment using an infrared thermal imager. If a communication failure or execution abnormality is detected, trigger an alarm and switch to safe mode.
[0022] Step S6, Iterative Optimization and Termination: Repeat the above acquisition, prediction and control steps until solidification is complete, and update the model periodically to adapt to process changes.
[0023] Step S6, loop optimization and termination, specifically includes the following steps: Step S61: Repeat steps S2 to S5 until the curing process reaches the preset total time (e.g., 30 minutes) or the temperature uniformity index stabilizes (e.g., the temperature difference between all areas and the target temperature is less than 100°C). ); Step S62: During the loop, the deep learning model is updated periodically (e.g., every 5 minutes) to fine-tune the model parameters using the latest acquired temperature data to adapt to the aging of the coil material or changes in the environment; Step S63: After the curing process is complete, save all control logs and temperature data for subsequent analysis and model optimization.
[0024] One specific application of this embodiment is: This embodiment focuses on the curing process of a high-voltage transformer coil of model HT-500. This coil is composed of multiple layers of insulating paper and interwoven copper wire. During the curing process, localized overheating can easily lead to degradation of the insulating material, and traditional control methods struggle to ensure temperature uniformity. The system achieves adaptive local temperature control through real-time infrared thermal imaging and deep learning prediction. The specific steps are as follows: System Configuration Coil parameters: HT-500 high-voltage transformer coil, diameter 0.5 meters, height 1 meter, surface divided into 10 independent heating zones (each zone corresponds to one heater); Infrared thermal imager: 640×480 pixels resolution, 1 frame / second frame rate, covering the entire coil surface; Heaters: 10 independently controlled heaters, with a power range of 0-1000W; Control parameter: Target curing temperature Predicted time step Seconds, optimize weight parameters ; Deep learning model: pre-trained convolutional neural network (CNN), with a 640×480 temperature matrix as input and a temperature distribution matrix for the next 10 seconds as output.
[0025] Implementation steps: Step S1: System Initialization Step S11: Load the pre-trained CNN model, which is trained based on historical HT-500 coil solidified data, and read the model weights from the storage module; Step S12: Set control parameters: target curing temperature Predicted time step seconds, heater power range Optimize weights ; Step S13: Initialize the infrared thermal imager and heater controller: Connect the device via Ethernet communication protocol, execute the self-test program to ensure that the thermal imager can acquire temperature data of 640×480 pixels and that each heater responds to control commands.
[0026] Step S2: Real-time data acquisition Step S21: Start the infrared thermal imager and acquire coil surface temperature data at a rate of 1 frame per second to obtain a real-time temperature matrix. In the formula, and For pixel coordinates, The current time; Step S22: Preprocess the temperature matrix: Apply a medium filter (window size 3×3) to remove noise, and map the pixel coordinates to 10 heating regions, each region corresponding to a 64×48 pixel block, and calculate the real-time average temperature of each region. for .
[0027] Step S3: Temperature Distribution Prediction Step S31: The preprocessed real-time temperature matrix Input a CNN model, and the model outputs a predicted temperature distribution matrix. ; Step S32: Verify the prediction results: Calculate the average predicted temperature for each region in the prediction matrix. If any area If the temperature is below the ambient temperature (assuming 25°C) or exceeds the material's heat resistance limit of 200°C, then bilinear interpolation is used to correct the value from the neighboring region.
[0028] Step S4: Control Decision Generation Step S41: Predict temperature based on regional average Define the optimization objective function: In the formula, It is the first The power of each heater; Step S42: Minimize using gradient descent Initialize current power Set the learning rate Iterate 5 times, updating the power in each iteration: In the formula, The gradient is obtained from the backpropagation of the CNN model and is approximately a constant of 0.5 (based on the thermal conductivity of the coil). Step S43: Constrain the power range: Calculate the adjustment amount and cut to make To obtain the final power value .
[0029] Step S5, Control Execution Step S51: Transfer the power adjustment command \( P_j^ The command is sent to the heater controller via the Modbus protocol. The command format is hexadecimal code, which includes the heater ID and power value. Step S52: Monitor execution results: Read heater feedback signals. If any heater does not acknowledge power adjustment within 2 seconds, mark the area as faulty and skip the adjustment in the next control cycle.
[0030] Step S6: Loop Optimization and Termination Step S61: Repeat steps 2 to 5, executing the control loop once every 10 seconds; Step S62: Update the CNN model every 5 minutes (i.e. after 30 loops): Fine-tune the model parameters using the temperature data from the most recent 5 minutes, set the learning rate to 0.001, and save the updated model version. Step S63: Termination condition: When the total curing time reaches 30 minutes, or the real-time temperature of all areas reaches a certain level. and The difference lasted for 5 minutes and was less than When the time comes, stop the control process and save all temperature data and control logs to the database.
[0031] This embodiment demonstrates the application of the system in the curing scenario of high-voltage transformer coils through specific parameters and operational details, ensuring temperature uniformity.
Claims
1. An intelligent optimization control system for the coil curing process, characterized in that: The control system includes: Data acquisition module: responsible for driving the infrared thermal imager to acquire raw temperature distribution data of the coil surface in real time; Data preprocessing module: Cleans, denoises, and performs coordinate mapping on the raw temperature data, transforming it into a regular and reliable temperature matrix; Temperature prediction module: Using a deep learning model, it predicts the temperature distribution of the coil at future times based on current temperature data, and anticipates changes in thermal behavior; Control decision module: Based on the predicted temperature distribution, the optimal power adjustment of each heater is calculated through an optimization algorithm to achieve the goal of uniform curing; Control execution module: Sends the power commands generated by the decision module to the heater for execution, and monitors the execution status in real time to ensure that the control is in place; System Management Module: As the central hub of the system, it is responsible for parameter configuration, model updates, process scheduling and security monitoring, ensuring the coordinated and stable operation of the entire system; The workflow of the control system is as follows: Step S1: System initialization: Load the pre-trained deep learning model, set the target temperature and other control parameters, and initialize the infrared thermal imager and multi-area heater controller. Step S2, Real-time data acquisition: Use an infrared thermal imager to acquire the surface temperature distribution of the coil, and perform preprocessing such as noise reduction and alignment; Step S3, Temperature Distribution Prediction: Input real-time temperature data into the model to predict the temperature distribution of each region of the coil at future times; Step S4, Control Decision Generation: Based on the predicted temperature distribution, the optimal power adjustment value for each heater is calculated using an optimization algorithm; Step S5, Control Execution: Send the power adjustment command to the heater controller, execute the power adjustment, and monitor the execution status; Step S6, Iterative Optimization and Termination: Repeat the above acquisition, prediction and control steps until solidification is complete, and update the model periodically to adapt to process changes.
2. The intelligent optimization control system for the coil curing process according to claim 1, characterized in that: Step S1, system initialization, specifically includes the following steps: Step S11: Load the pre-trained deep learning model, which is trained based on historical high-voltage transformer coil solidification data and is used to predict temperature distribution; the model input is the coil surface temperature matrix and the output is the temperature distribution at future time steps. Step S12: Set control parameters, including target curing temperature. Predicting time steps Heater power range and optimizing weight parameters ; Step S13: Initialize the infrared thermal imager and multi-zone heater controller to ensure normal communication between the device and the main control system; set the resolution of the infrared thermal imager to at least 640×480 pixels to cover coil surface details.
3. The intelligent optimization control system for the coil curing process according to claim 1, characterized in that: Step S2, real-time data acquisition, specifically includes the following steps: Step S21: Activate the infrared thermal imager to acquire real-time temperature distribution data on the surface of the high-voltage transformer coil; the data format is a two-dimensional matrix. ,in, For pixel coordinates, Indicates the current time; Step S22: Preprocess the collected temperature data, including noise reduction and coordinate alignment, and map the temperature matrix to the physical space area of the coil to ensure that each pixel corresponds to a specific heating control area.
4. The intelligent optimization control system for the coil curing process according to claim 1, characterized in that: Step S3, temperature distribution prediction, specifically includes the following steps: Step S31: The preprocessed real-time temperature matrix Input a pre-trained deep learning model, which is based on a convolutional neural network structure, and output the future time step. Predicted temperature distribution The prediction process takes into account the thermal conductivity characteristics of the coil material and historical temperature changes, and the model update frequency is synchronized with data acquisition. Step S32: Verify the rationality of the prediction results. By comparing the gradient changes of the current temperature and the predicted temperature, if the predicted value exceeds the physical range, the backup interpolation algorithm is activated for correction.
5. The intelligent optimization control system for the coil curing process according to claim 1, characterized in that: Step S4, the control decision generation, specifically includes the following steps: Step S41: Based on the predicted temperature distribution Calculate the control adjustment for each heater zone and define the optimization objective function: In the formula, J The objective function value to be minimized in the optimization process. N This represents the number of temperature monitoring points (corresponding to pixel areas). The index variable is used for summation to iterate through all temperature monitoring points. For the first Predicted temperature values at each temperature monitoring point The preset, desired target curing temperature, These are the weighting coefficients. M For the number of heaters, j The index variable is used to sum and iterate through all heaters. P j For the first j The power of each heater; Step S42: Solve for min using the gradient descent algorithm. J The optimal power adjustment value for each heater is obtained. The solution process is constrained by Ensure that the power is within a safe range; Step S43: Adjust the power value It is converted into a control signal and formatted into a sequence of instructions that the heater controller can recognize.
6. The intelligent optimization control system for the coil curing process according to claim 1, characterized in that: Step S5, the control execution specifically includes the following steps: Step S51: Send the generated control commands to the multi-zone heater controller via the communication interface to adjust the power output of each heater in real time. P j ; Step S52: Monitor the execution results and verify the temperature change after power adjustment using an infrared thermal imager. If a communication failure or execution abnormality is detected, trigger an alarm and switch to safe mode.
7. The intelligent optimization control system for the coil curing process according to claim 1, characterized in that: Step S6, loop optimization and termination, specifically includes the following steps: Step S61: Repeat steps S2 to S5 until the curing process reaches the preset total time or temperature uniformity index and stabilizes. Step S62: During the loop, the deep learning model is updated periodically, and the model parameters are fine-tuned using the latest collected temperature data to adapt to the aging of the coil material or changes in the environment. Step S63: After the curing process is complete, save all control logs and temperature data for subsequent analysis and model optimization.