Rare earth molten salt electrolysis intelligent regulation and control method based on infrared thermal imaging and AI prediction
By combining infrared thermal imaging with AI prediction, the problem of incomplete temperature monitoring in rare earth molten salt electrolysis has been solved, enabling intelligent control of rare earth molten salt electrolysis and improving production efficiency and product quality.
Patent Information
- Application Number
- CN202511355703.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-22
- Publication Date
- 2026-01-16
AI Technical Summary
In existing rare earth molten salt electrolysis production, temperature monitoring is not comprehensive enough, component detection is lagging, and control methods rely on human experience, resulting in low electrolysis efficiency, short equipment life and unstable product quality.
By combining infrared thermal imaging with AI prediction, a method is adopted to monitor temperature using thermocouples and infrared thermal imagers, and to build an AI prediction system based on GARCH and LSTM models. This enables real-time monitoring and accurate prediction of temperature and composition, and dynamic adjustment of electrolysis parameters.
It achieved 100% temperature monitoring coverage, reduced electrode ablation risk, 96.3% accuracy in predicting product carbon content, 10% increase in single-furnace metal output, and 12% reduction in unit energy consumption, significantly improving production stability and efficiency.
Smart Images
Figure CN121354698A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of rare earth molten salt electrolysis technology, specifically to a method for intelligent control of rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction. Background Technology
[0002] In current rare earth molten salt electrolysis production processes, factories generally rely on thermocouples for temperature measurement, manual sampling for composition analysis, and operators adjusting various process parameters based on experience. While these methods have been used for many years, with the expansion of production scale and the increasing demands for product quality, these traditional methods have revealed many problems. For example, temperature monitoring is not comprehensive enough, composition detection is too slow, and the control process relies on human experience, making it increasingly difficult to meet the needs of intelligent and efficient production.
[0003] The existing technology has the following shortcomings:
[0004] (1) Insufficient temperature monitoring coverage: The currently commonly used single-point thermocouple arrangement cannot fully reflect the temperature distribution of the entire electrolytic cell. In particular, the temperature changes are drastic but difficult to detect at the corners and around the electrodes, often resulting in localized crusting or electrode abnormalities, which can easily affect electrolysis efficiency and equipment lifespan. For example, the failure to monitor the edge temperature in the plant area in a timely manner led to the solidification of molten salt, resulting in a significant decline in annual output value.
[0005] (2) Delay in component detection: The original chemical analysis was mostly done by manual sampling and offline detection. This delay caused the control feedback to be delayed. Once the carbon content of a batch of raw materials fluctuates greatly, it may continuously affect the product quality of multiple batches, resulting in waste.
[0006] (3) The control method is relatively empirical: The parameter adjustment relies on the experience and historical records of the team workers, lacking real-time data support and accurate prediction mechanism, resulting in unstable energy consumption and uneven output. Summary of the Invention
[0007] The purpose of this invention is to provide an intelligent control method for rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction, which improves the spatial coverage and identification accuracy of molten surface temperature monitoring, constructs a closed-loop intelligent parameter control system, realizes dynamic optimization and efficient operation of the electrolysis process, and enables online prediction and trend identification of the carbon content of rare earth products.
[0008] To achieve the above objectives, this invention provides a method for intelligent control of rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction, comprising the following steps:
[0009] Step 1: Select locations on the object to be tested and place the thermocouple and infrared thermal imager respectively;
[0010] Step 2: Measure the temperature at a fixed location using an infrared thermal imager to obtain infrared imaging data;
[0011] Step 3: Compare the actual temperature measured by the thermocouple with the temperature measured by the infrared thermal imager;
[0012] Step 4: Calculate the emissivity correction factor based on the comparison results;
[0013] Step 5: Based on the emissivity correction factor, dynamically adjust the measurement settings of the infrared thermal imager to further acquire temperature data;
[0014] Step 6: Integrate GARCH and LSTM models to build an AI prediction model, input multidimensional data for collaborative prediction, and provide decision support for the control system.
[0015] Optionally, the temperature measured by the thermocouple in step 1 can be used as a known accurate temperature reference, and both the thermocouple and the infrared thermal imager can measure the temperature at the same location.
[0016] Optionally, the emissivity correction factor in step 4 is obtained by comparing the thermocouple data with the infrared imaging data, and the calculation formula is as follows:
[0017] ,
[0018] in, The temperature measured by the thermocouple. The temperature measured by an infrared thermal imager. This is the corrected temperature value.
[0019] Optionally, the multidimensional data in step 6 includes current, voltage, temperature, feed rate, and historical composition data.
[0020] Optionally, the AI prediction model adopts a cascaded fusion structure of LSTM and GARCH, wherein the LSTM model takes multi-dimensional time series data as input, learns long-term patterns in historical data through backpropagation algorithm, and outputs trend prediction results for future moments.
[0021] Optionally, the input of the GARCH model in the AI prediction model is the residual or error value predicted by the LSTM model, that is, the difference between the LSTM predicted value and the actual observed value, and the output is the volatility prediction value for a certain period of time in the future.
[0022] This invention provides an intelligent control method for rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction. Specifically, it compares thermocouple measurement data with infrared thermal imaging data and uses an emissivity correction algorithm to correct the data in real time, eliminating the influence of molten salt volatilization and surface state changes on temperature measurement. Then, it constructs an AI prediction model using GARCH and LSTM models. Multi-dimensional data such as current, voltage, raw material input, and historical test results are used as input, and an LSTM neural network model is used for fitting. At the same time, the GARCH method is referenced to supplement the data volatility, predict the indicators, and provide quantitative prediction of uncertainty, providing a basis for advance adjustment, thereby optimizing the production process. Attached Figure Description
[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0024] Figure 1 This is a schematic diagram illustrating the technical principle of a rare earth molten salt electrolysis intelligent control method based on infrared thermal imaging and AI prediction, according to the present invention.
[0025] Figure 2 This is a schematic diagram of the temperature measurement hardware deployment in a specific embodiment of the present invention. Detailed Implementation
[0026] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.
[0027] This invention provides a method for intelligent control of rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction, comprising the following steps:
[0028] Step 1: Select locations on the object to be tested and place the thermocouple and infrared thermal imager respectively;
[0029] Step 2: Measure the temperature at a fixed location using an infrared thermal imager to obtain infrared imaging data;
[0030] Step 3: Compare the actual temperature measured by the thermocouple with the temperature measured by the infrared thermal imager;
[0031] Step 4: Calculate the emissivity correction factor based on the comparison results;
[0032] Step 5: Based on the emissivity correction factor, dynamically adjust the measurement settings of the infrared thermal imager to further acquire temperature data;
[0033] Step 6: Integrate GARCH and LSTM models to build an AI prediction model, input multidimensional data for collaborative prediction, and provide decision support for the control system.
[0034] The technical principle of this invention is as follows: Figure 1 As shown, the following description, in conjunction with specific embodiments and implementation steps, will provide further details:
[0035] The hardware architecture in this embodiment includes: a high-resolution industrial-grade infrared thermal imager installed approximately 50cm directly above the electrolytic cell. The imager has a resolution of 640×512 pixels, a temperature measurement range from 200℃ to 1200℃, and a response time of less than 50ms, capable of comprehensively scanning the molten surface of the electrolytic cell. The outer shell is made of a special alloy material capable of withstanding high temperatures. The scanning angle is 120°, and the image resolution corresponds to an actual spatial resolution of 1cm×1cm. Emissivity correction based on thermocouples is performed hourly. To ensure accurate emissivity, three thermocouples are placed at the edge, center, and near the electrodes of the electrolytic cell for point-to-point comparison. The deployment is as follows... Figure 2 As shown.
[0036] Once deployed, the system will perform comprehensive temperature monitoring. Every hour, the system will execute steps 1 through 5 to correct the emissivity of the infrared thermal imager based on thermocouple data, ensuring data accuracy under different production environments. Through the system's automated correction and feedback mechanism, the real-time and accurate temperature monitoring can be ensured, avoiding potential risks caused by accumulated errors.
[0037] Furthermore, six months of data were collected from the enterprise, including infrared images every two seconds, thermocouple data, and carbon content measurement data. The data was divided into training, validation, and test sets. After training, the model's prediction error on the test set was <±3.5kg, meeting the requirements of practical applications.
[0038] In step 6, the AI prediction model adopts a cascaded fusion structure of LSTM and GARCH. Specifically, the LSTM model outputs a predicted future trend value, and its residuals are used as input to the GARCH model to model short-term volatility. Finally, the results of the two models are fused to achieve joint prediction of trends and volatility. This structure differs from the individual application of LSTM or GARCH in existing technologies. Instead, it achieves complementarity through residual coupling, significantly improving prediction accuracy and industrial applicability.
[0039] LSTM models excel at handling time series data with long-term dependencies, capturing trends and long-term changes in the input data. The processing procedure is as follows:
[0040] Input data: Multidimensional time series data (such as current, voltage, etc.) are used as input to the LSTM. The model's task is to predict future long-term trends based on historical data (e.g., predict temperature changes in the next 30 minutes).
[0041] Training process: In the LSTM model, the network learns long-term patterns in historical data (e.g., the trend of electrolyzer temperature variation under different operating conditions) through the backpropagation algorithm. The memory units of LSTM can help the model learn long-term dependencies and avoid the gradient vanishing problem in traditional RNN models.
[0042] Predictive output: LSTM outputs trend predictions for future time points, which will provide the basis for the next step of volatility modeling (processed by the GARCH model).
[0043] GARCH models are primarily used to model volatility in time series data. Unlike LSTM, GARCH focuses on the volatility of data in the short term, specifically the variation in variance. The processing procedure is as follows:
[0044] Input data: The input to the GARCH model is typically the residuals or error values predicted by the LSTM model (i.e., the difference between the LSTM predictions and the actual observations). These residuals reflect short-term fluctuations during the electrolysis process.
[0045] Volatility Modeling: GARCH models analyze the variance (volatility) of these error terms and capture patterns in volatility. GARCH models consider conditional heteroscedasticity, meaning that the volatility of data changes under certain conditions. By estimating the conditional variance of the residuals, GARCH can predict future volatility.
[0046] Output: GARCH outputs a volatility prediction for a future time period (e.g., current volatility over the next 30 minutes). This volatility prediction can help the electrolysis process control system identify potential volatility risks.
[0047] Furthermore, the LSTM model and the GARCH model are fused to optimize prediction and control during the electrolysis process.
[0048] Specifically, the LSTM model provides predictions of future trends, such as changes in temperature and carbon content. These predictions guide the long-term adjustments to the electrolysis process.
[0049] The GARCH model, based on the residuals output by the LSTM model, provides quantitative predictions of short-term fluctuations. This is crucial for dealing with sudden changes in the electrolysis process (such as equipment fluctuations, raw material fluctuations, etc.).
[0050] By combining long-term trends with short-term fluctuations in forecasting, the system can provide more comprehensive predictions of future production processes, ensuring the stability and accuracy of these processes.
[0051] Furthermore, during electrolysis, the outputs of LSTM and GARCH can provide decision support for the control system. For example, when LSTM predicts a gradual increase in temperature, GARCH can tell the system whether fluctuations will occur, thus determining whether the current or feed rate needs to be adjusted.
[0052] This invention is further illustrated with examples of field applications:
[0053] The system's deployment on a rare earth company's 5kA electrolysis production line has yielded significant results. Through thermal imaging, operators promptly detected an area of abnormally high temperature at the edge of the electrolytic cell. The system automatically adjusted the cooling fan speed and angle, reducing the temperature to normal within 10 minutes, thus preventing potential production accidents. Based on an AI predictive model, the system automatically optimized current and feed parameters. Single-furnace metal output increased from 28kg to 32.5kg, and carbon content decreased from 0.048% to 0.039%. Statistics show annual energy savings of approximately 1.2 million kWh, reduction of 45 tons of raw material waste, and direct economic benefits exceeding 3 million yuan. Simultaneously, production process stability has been significantly improved, and the scrap rate has been significantly reduced.
[0054] The beneficial effects are as follows:
[0055] (1) Temperature monitoring coverage is increased to approximately 100%, the risk of electrode ablation is reduced by more than 52%, and the equipment operation and maintenance cycle is extended by 30%;
[0056] (2) The accuracy rate of carbon content prediction for products reached 96.3%, and the rate of exceeding the standard decreased from 18% to 3.2%;
[0057] (3) The metal output per furnace increases by 10% to 15%, and the unit energy consumption decreases by 12% to 18%;
[0058] (4) The system achieves closed-loop operation, reducing manual intervention by more than 80%, and significantly improving production efficiency and stability.
[0059] The above description discloses only one or more preferred embodiments of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art can understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.
Claims
1. A method for intelligent control of rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction, characterized in that, Includes the following steps: Step 1: Select locations on the object to be tested and place the thermocouple and infrared thermal imager respectively; Step 2: Measure the temperature at a fixed location using an infrared thermal imager to obtain infrared imaging data; Step 3: Compare the actual temperature measured by the thermocouple with the temperature measured by the infrared thermal imager; Step 4: Calculate the emissivity correction factor based on the comparison results; Step 5: Based on the emissivity correction factor, dynamically adjust the measurement settings of the infrared thermal imager to further acquire temperature data; Step 6: Integrate GARCH and LSTM models to build an AI prediction model, input multidimensional data for collaborative prediction, and provide decision support for the control system.
2. The intelligent control method for rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction as described in claim 1, characterized in that, The temperature measured by the thermocouple in step 1 serves as a known accurate temperature reference. Both the thermocouple and the infrared thermal imager can measure the temperature at the same location.
3. The intelligent control method for rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction as described in claim 2, characterized in that, The emissivity correction factor in step 4 is obtained by comparing thermocouple data with infrared imaging data, and the calculation formula is as follows: ， ； in, The temperature measured by the thermocouple. The temperature measured by an infrared thermal imager. This is the corrected temperature value.
4. The intelligent control method for rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction as described in claim 3, characterized in that, The multidimensional data in step 6 includes current, voltage, temperature, feed rate, and historical composition data.
5. The intelligent control method for rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction as described in claim 4, characterized in that, The AI prediction model adopts a cascaded fusion structure of LSTM and GARCH. The LSTM model takes multi-dimensional time series data as input, learns long-term patterns in historical data through backpropagation algorithm, and outputs trend prediction results for future moments.
6. The intelligent control method for rare earth molten salt electrolysis based on infrared thermal imaging and AI prediction as described in claim 5, characterized in that, The input to the GARCH model in the AI prediction model is the residual or error value predicted by the LSTM model, that is, the difference between the LSTM prediction value and the actual observation value, and the output is the volatility prediction value for a certain period of time in the future.