Method and device for predicting and controlling in-furnace temperature of ultra-fast high-temperature sintering furnace in real time

Abstract

The invention discloses an in-furnace temperature real-time prediction and control method and device for an ultra-fast high-temperature sintering furnace, and the method comprises the steps: collecting in-furnace technological parameters and temperature data in real time, and carrying out the preprocessing, and obtaining a training data set; introducing a multi-task learning framework and an attention mechanism into the MLP model, and constructing an improved MLP model; training the improved MLP model by adopting the training data set to obtain a trained improved MLP model; the preprocessed real-time data are input into an improved MLP model, the temperature in the furnace is predicted in real time through the improved MLP model, and the predicted temperature is obtained; and the closed-loop control module monitors the temperature in the furnace in real time according to the predicted temperature. The in-furnace temperature real-time prediction and control method for the ultra-fast high-temperature sintering furnace can solve the problems that a traditional temperature prediction method is low in precision, low in response speed and lack of closed-loop control under complex process conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial temperature control, and in particular relates to a method and device for real-time prediction and control of the temperature in an ultra-fast high-temperature sintering furnace. Background Art

[0002] With the accelerating pace of industrialization, efficient and precise production processes have become crucial requirements for all types of industrial equipment, especially high-temperature sintering equipment. In fields such as metals, ceramics, and powder metallurgy, the sintering process, as a crucial manufacturing step, directly impacts the quality of the final product and production efficiency. Precise control of the sintering temperature is a key factor affecting sintering quality, efficiency, and energy consumption.

[0003] Traditional sintering furnace temperature control methods typically rely on temperature sensors, thermocouples, and other measuring devices for temperature monitoring. These devices have slow response times in high-temperature environments and their measurement results are subject to errors. Furthermore, temperature adjustments using traditional methods often lag, preventing precise real-time control. This leads to accumulated temperature control errors, which in turn impact sintering quality and can even cause low production efficiency and energy waste.

[0004] As an emerging sintering method, UHS (ultra-fast sintering) technology has shown good application prospects in the field of high-temperature processing due to its advantages such as high efficiency, rapid heating, energy conservation and emission reduction. However, in actual applications, UHS ultra-fast sintering devices face the challenge of large temperature fluctuations due to their high-speed heating characteristics. Due to the rapid changes in the temperature in the furnace during the sintering process, traditional temperature control methods are difficult to adapt to the dynamic characteristics of this type of equipment, making it difficult to monitor and adjust the temperature in the furnace in real time, which in turn affects the stability of the sintering process and product quality. Especially during the rapid heating stage, the temperature fluctuates violently, which places higher requirements on equipment control and product quality control.

[0005] In recent years, deep learning technology has been gradually introduced into industrial temperature control systems. Multilayer perceptrons (MLPs), a type of deep learning, have been widely used in prediction and control due to their excellent nonlinear fitting capabilities and ability to capture complex relationships. Compared to traditional statistical methods, MLP models can accurately predict future furnace temperature changes based on the characteristic relationships of historical data, thus avoiding the lag and inaccuracy of traditional methods and achieving more precise and real-time temperature control.

[0006] Although some studies have attempted to apply MLP to temperature prediction, the existing research results still have certain limitations due to the high temperature characteristics and complex process parameters of UHS ultra-fast sintering devices. Many existing models fail to fully consider the impact of process data such as process number, current intensity, and current duration on temperature changes, and these factors play a vital role in the ultra-fast sintering process. For example, the intensity and duration of the current directly affect the heating rate in the furnace, and the introduction of the process number can help the model better distinguish data under different process conditions, so that the model can adapt to different temperature change patterns according to different process characteristics. These factors interact with each other during the temperature change process, thereby affecting the accuracy of temperature prediction. Therefore, how to effectively and comprehensively consider the complex relationship between multi-dimensional process parameters and furnace temperature in the MLP model is the key to further improving the prediction accuracy and control accuracy of sintering temperature.

[0007] Furthermore, existing technologies often separate temperature prediction and control, lacking a real-time linkage mechanism. This separate design leads to lag in temperature control, making it difficult to adapt to the rapidly changing temperature environment of UHS ultrafast sintering equipment. Therefore, combining temperature prediction with closed-loop control to achieve real-time monitoring and dynamic regulation of furnace temperature is an urgent issue. Summary of the Invention

[0008] In order to overcome the shortcomings of the existing technology, the present invention provides a real-time prediction and control method for the temperature in an ultra-fast high-temperature sintering furnace, which is used to solve the problems of low accuracy, slow response speed and lack of closed-loop control of traditional temperature prediction methods under complex process conditions.

[0009] The second object of the present invention is to provide a device for real-time prediction and control of the temperature inside an ultra-fast high-temperature sintering furnace.

[0010] The technical solution of the present invention to solve the above technical problems is:

[0011] A method for real-time prediction and control of the temperature in an ultra-fast high-temperature sintering furnace comprises the following steps:

[0012] S1: Real-time collection of process parameters and temperature data in the furnace, and pre-processing of the collected process parameters and temperature data to obtain a training data set;

[0013] S2: Introducing a multi-task learning framework and attention mechanism into the MLP model to construct an improved MLP model;

[0014] S3: Using the training data set in step S1 to train the improved MLP model in step S2, to obtain a trained improved MLP model;

[0015] S4: Input the preprocessed real-time data into the improved MLP model, and use the trained improved MLP model to predict the furnace temperature in real time to obtain the predicted temperature;

[0016] S5: Based on the predicted temperature output by the improved MLP model, the closed-loop control module monitors the temperature inside the furnace in real time.

[0017] Preferably, in step S1, the process parameters and temperature data in the furnace are collected in real time through sensors; the process parameters include time series characteristics, process number, physical characteristics and environmental parameters, wherein the time series characteristics include the current time and running time; the physical parameters include current intensity and voltage intensity; the environmental parameters include furnace pressure and atmosphere composition.

[0018] Preferably, in step S1, the steps of preprocessing the collected process parameters and temperature data are:

[0019] S101: storing the collected process parameters and temperature data in a database or temporary storage, preliminarily organizing the process parameters and temperature data, and unifying the data format;

[0020] S102: Cleaning the collected process parameter and temperature data to remove noise and outliers, eliminate erroneous data caused by sensor failure or environmental interference, and fill in or annotate missing data;

[0021] S103: Standardizing and normalizing the process parameters and temperature data, and constructing dynamic features based on the time series characteristics of the process parameters;

[0022] S104: Perform numerical encoding and embedding processing on the process number, and map it into a continuous vector space representation.

[0023] Preferably, the steps of step S103 are:

[0024] Obtain the minimum value of the process number and subtract the minimum value from all process numbers to ensure that the process numbers are in the range of 0 to N-1, where N is the number of process number types;

[0025] For the data of other process parameters and temperature data except process number, the standardization method is used to make the mean of each data 0 and the standard deviation 1;

[0026] Each process number is assigned an embedding vector and concatenated with the original input features to form the final input of the model.

[0027] Preferably, in step S2, the steps of optimizing the traditional MLP model are:

[0028] S201: Based on the traditional MLP model, design a network architecture of an improved MLP model, wherein the network architecture of the improved MLP model includes an input layer, a hidden layer, and an output layer;

[0029] S202: Introducing a process number embedding layer into the input layer to map the process number into a low-dimensional continuous vector to capture the potential association between different process numbers;

[0030] S203: Introducing a multi-task learning framework that combines the temperature prediction task with an auxiliary task for capturing the correlation between process parameters and temperature;

[0031] S204: Add a self-attention layer to the hidden layer to dynamically calculate the weights of the input features;

[0032] S205: Define a weighted average loss function to balance the learning objectives of the temperature prediction task and the process parameter association analysis task, and use an adaptive learning rate scheduler to optimize the training process.

[0033] Preferably, in step S4, the collected real-time data is fed into the trained improved MLP model, and the input features are dynamically adjusted based on the changing trend of the process parameters; the prediction results output by the improved MLP model are denormalized to obtain the final predicted temperature.

[0034] Preferably, in step S5, the specific steps of the closed-loop control module performing real-time monitoring of the temperature in the furnace are:

[0035] Set a safety threshold. When the predicted temperature exceeds the safety threshold, the closed-loop control module will automatically trigger an alarm and stop the operation of the ultra-fast high-temperature sintering furnace; at the same time, it will record the triggering time of the alarm, the corresponding predicted temperature value, and related process parameters.

[0036] A device for real-time prediction and control of the temperature in an ultra-fast high-temperature sintering furnace, comprising:

[0037] Data acquisition module: used to collect process parameters and temperature data in the furnace in real time;

[0038] Data processing module: used for data cleaning and preprocessing of collected process parameters and temperature data;

[0039] Model prediction module: Based on the process parameters and temperature data processed by the data processing module, the improved MLP model installed inside the module predicts the temperature inside the furnace and performs denormalization on the prediction results to obtain the predicted temperature;

[0040] Closed-loop control module: Based on the predicted temperature output by the model prediction module, the ultra-fast high-temperature sintering furnace is monitored in real time.

[0041] An electronic device includes a central processing unit and a memory, wherein the central processing unit is used to call and run a computer program stored in the memory to execute the steps of the method for real-time prediction and control of the temperature in the ultra-fast high-temperature sintering furnace.

[0042] A computer-readable storage medium stores a computer program implemented according to the method for real-time prediction and control of the temperature in an ultra-fast high-temperature sintering furnace in the form of computer-readable instructions. When the computer program is called and executed by a computer, the steps included in the corresponding method are executed.

[0043] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0044] 1. The improved MLP model in the real-time prediction and control method for the furnace temperature of an ultra-fast high-temperature sintering furnace of the present invention optimizes the model structure through multi-task learning and attention mechanism, so that the improved MLP model can more efficiently capture the complex nonlinear relationship between the furnace temperature and multi-dimensional process parameters, significantly improving the accuracy and response speed of temperature prediction.

[0045] 2. The real-time prediction and control method for the temperature inside the ultra-fast high-temperature sintering furnace of the present invention uses an improved MLP model to accurately predict the temperature inside the furnace without relying on sensors, thereby greatly reducing equipment maintenance costs and improving system reliability.

[0046] 3. The method for real-time prediction and control of the temperature in the ultra-fast high-temperature sintering furnace of the present invention reduces energy waste and improves production efficiency through precise temperature control.

[0047] 4. The closed-loop control module in the real-time prediction and control method for the temperature inside the ultra-fast high-temperature sintering furnace of the present invention can stop the operation of the ultra-fast high-temperature sintering furnace in time when the temperature is too high, thereby effectively preventing damage to the ultra-fast high-temperature sintering furnace and the occurrence of safety accidents.

[0048] 5. The real-time prediction and control method of the temperature inside the ultra-fast high-temperature sintering furnace of the present invention is not only applicable to UHS ultra-fast sintering devices, but can also be extended to temperature management scenarios of other high-temperature processing equipment (such as metal heat treatment furnaces, ceramic sintering furnaces, etc.).

[0049] 6. The real-time prediction and control method for the temperature inside the ultra-fast high-temperature sintering furnace of the present invention provides an innovative solution for the intelligent temperature control of ultra-fast sintering equipment, significantly improves the stability and efficiency of the sintering process, and has important industrial application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The figure is a flow chart of the method for real-time prediction and control of the temperature in the ultra-fast high-temperature sintering furnace of the present invention.

[0051] Figure 2 Schematic diagram of the network architecture of the improved MLP model.

[0052] Figure 3 A diagram of the data set.

[0053] Figure 4 A schematic diagram of the prediction results.

[0054] Figure 5 Comparison of time-temperature curves between the improved MLP model and the COMSOL simulation model.

[0055] Figure 6 Comparison of the time-temperature curves of the improved MLP model and the infrared thermometer.

[0056] Figure 7 Comparison of the time-temperature curves of the improved MLP model and infrared thermal imaging. DETAILED DESCRIPTION

[0057] The present invention will be described in further detail below with reference to the embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0058] Example 1

[0059] See also Figures 1-4 The method for real-time prediction and control of the temperature in an ultra-fast high-temperature sintering furnace of the present invention comprises the following steps:

[0060] S1: Real-time collection of process parameters and temperature data in the furnace, and pre-processing of the collected process parameters and temperature data to obtain a training data set;

[0061] In this embodiment, the process parameters and temperature data in the furnace are collected in real time by sensors, wherein the data collection frequency needs to meet the demand of capturing the rapid changes in the temperature in the furnace; the process parameters include time series characteristics, process numbers, physical characteristics and environmental parameters, wherein the time series characteristics include the current time and running time, etc.; the process number is used to identify different current step process types (numbered in sequence such as A, B, C, etc.); the physical parameters include current intensity and voltage intensity; the environmental parameters include furnace pressure and atmosphere composition.

[0062] Next, the collected process parameter and temperature data are cleaned and preprocessed to ensure data quality. The specific steps are as follows:

[0063] S101: storing the collected process parameters and temperature data in a database or temporary storage, and preliminarily organizing the process parameters and temperature data to ensure that the data format is unified and the time sequence is complete, and to prepare for subsequent processing;

[0064] S102: Clean the collected process parameter and temperature data to remove noise and outliers. Statistical analysis or machine learning-based methods are used to identify and eliminate abnormal data points. For example, erroneous data caused by sensor failure or environmental interference is eliminated. Missing data is properly filled or annotated.

[0065] S103: Standardize (Z-score standardization) and normalize (Min-Max normalization) the process parameters and temperature data to make them evenly distributed and eliminate dimensional differences to ensure the stability of model training. Based on the time series characteristics of the process parameters, dynamic features are constructed to enhance the model's ability to capture rapid changes in furnace temperature. In this embodiment, the dynamic features include sliding windows and time difference features. The sliding window is used to divide the time series data into sliding windows of fixed length (such as 10 time steps) to extract the change trend of historical process parameters. The time difference feature is used to calculate the difference between the current moment and the previous moments (such as Δ current intensity, Δ time, etc.) to capture the change rate of the process parameters.

[0066] S104: Perform numerical encoding and embedding processing on the process number, and map it into a continuous vector space representation.

[0067] During the data collection and preprocessing process, all input data must conform to specific formatting requirements. For example, time data uses a standard timestamp format, current intensity and duration are floating-point values, and process numbers are represented by integer mappings. Specifically, process numbers are alphabetic inputs. To accommodate the input of the improved MLP model, they are first converted to numeric codes. For example, the letter "A" is mapped to 0, "B" to 1, and so on.

[0068] S2: Introduce the multi-task learning framework and attention mechanism into the MLP model to build an improved MLP model. Specifically:

[0069] S201: Based on the traditional MLP model, design a network architecture of an improved MLP model, wherein the network architecture of the improved MLP model includes an input layer, a hidden layer, and an output layer;

[0070] S202: Introducing a process ID embedding layer into the input layer, mapping the process ID (process_id) into a continuous vector representation to capture the potential associations between different process IDs;

[0071] S203: Introducing a multi-task learning framework to combine the temperature prediction task with other related tasks (such as process parameter correlation analysis), share feature representations, and improve the model's ability to capture complex nonlinear relationships;

[0072] S204: Add a self-attention layer to the hidden layer to dynamically calculate the importance weights of the input features to enhance the model's ability to focus on key process parameters.

[0073] S205: Define a weighted average loss function to balance the learning objectives of the temperature prediction task and the process parameter association analysis task, and use an adaptive learning rate scheduler to optimize the training process.

[0074] Therefore, the improved MLP model in the present invention includes an input layer, a process number embedding layer, a multi-task learning framework, a hidden layer and an output layer, wherein,

[0075] The input layer is used to receive the standardized and normalized process parameters and the embedded process numbers;

[0076] In this embodiment, the input layer is used to receive real-time data from the furnace, including time, current, temperature, and other data. This data is loaded and preprocessed using a CSV file. The input features (X) and outputs (y, including temperature and process parameters) are standardized using a StandardScaler. During training, the standardizer processes the training data. During testing, the trained standardizer is used to perform the same processing on the data to ensure consistency between the test data and the training data.

[0077] The input layer processes the input data as follows:

[0078] The minimum process ID is obtained and subtracted from all process IDs to ensure that the process IDs are within the range of 0 to N-1, where N is the number of process ID types. Input features such as time, current, and temperature are normalized. A standardization method (StandardScaler) is used to ensure that each input feature has a mean of 0 and a standard deviation of 1, thereby eliminating the impact of different dimensions on model training. Fitting is performed on the training set, and the same transformation is applied to the test set. Since the process ID (process_id) is embedded as a discrete feature, an embedding vector is assigned to each process ID and concatenated with the original input features (time, current, temperature) to form the final input for the improved MLP model. The dimension of the embedding layer is determined by the hyperparameter embedding_dim, and the parameters of the embedding layer are obtained through training. Finally, the data format of the dataset is converted to a PyTorch tensor to meet the input requirements of the improved MLP model.

[0079] The multi-task learning framework is used to combine the temperature prediction task with other auxiliary tasks (such as process parameter correlation analysis) and adopt a shared feature representation method. This not only improves the accuracy of temperature prediction, but also uses process parameter information to assist the learning of the improved MLP model.

[0080] The self-attention layer dynamically calculates the importance weights of input features and adopts a query-key-value mechanism to enhance the improved MLP model's ability to focus on key process parameters. During training, the loss weights of temperature data and process parameters are adjusted according to the weight changes, thereby affecting the final loss calculation and helping the improved MLP model to achieve a better balance between different tasks.

[0081] The hidden layer (fully connected layer) includes at least two fully connected layers and uses a ReLU activation function to enhance nonlinear expression capabilities;

[0082] In this embodiment, the number of layers and nodes in the hidden layer is set by hyperparameters. The expressive power of the improved MLP model is generally improved by increasing the network depth. Each hidden layer is nonlinearly transformed using a ReLU activation function. The role of these hidden layers is to extract complex feature patterns from the input features and the embedded process number to capture the regularity of temperature changes in the furnace.

[0083] The output layer includes two nodes: one for temperature prediction value and one for process parameter prediction value. In this embodiment, the output layer uses a linear activation function to ensure that the prediction result is continuous, so as to be suitable for regression tasks such as temperature.

[0084] In addition, the weighted average loss function is defined as:

[0085]

[0086] in, is the loss function (mean square error MSE) of the temperature prediction task, is the loss function of the auxiliary task (classification loss or regression loss), and α is the weight coefficient (set to 0.7).

[0087] S3: Use the training data set in step S1 to train the improved MLP model in step S2 to obtain a trained improved MLP model. The specific training process is as follows:

[0088] The training dataset is divided into a training set and a test set to prevent overfitting and ensure that the improved MLP model is evaluated on unseen data to verify the generalization ability of the model.

[0089] During the training phase, the input features are normalized using the training set data and the normalizers (scaler_X and scaler_y) are saved. During the testing phase, the saved normalizers are applied to transform the test set to ensure that the same preprocessing method is used during training and testing.

[0090] The preprocessed data is imported into the improved MLP model for training. After every 100 epochs, the current total loss is output to facilitate monitoring of training progress. If the current total loss is less than the previous best loss, the state of the improved MLP model is saved to ensure that the best improved MLP model is obtained through training. During the training process, the Adam optimizer is used to adaptively adjust the learning rate, which helps to accelerate training and avoid overfitting. The improved MLP model outputs the predicted furnace temperature in the test set, and the predicted result is fed back to the sintering furnace control system. The temperature is monitored based on the predicted result, and adjustments are made based on the deviation between the actual temperature and the predicted temperature. If the predicted temperature of the improved MLP model deviates significantly, the process parameters are adjusted through the feedback mechanism or the improved MLP model is retrained to optimize the temperature control during the sintering process.

[0091] During training, an adaptive learning rate scheduler (cosine annealing strategy) is used to dynamically adjust the learning rate to accelerate convergence and improve prediction accuracy. The mean square error (MSE) is used as the loss function to calculate the losses of temperature and process parameters separately. The Adam optimizer is selected as the optimizer because of its adaptive learning rate, which can effectively accelerate model training and prevent gradient explosion or vanishing problems. During training, the loss calculation will update the model weights through backpropagation. The temperature loss and process parameter loss are weighted and merged with the dynamic weights to calculate the total loss. The dynamic weights can adjust the weight contribution of different tasks based on the effects observed during training.

[0092] The improved MLP model in this embodiment combines advanced methods such as multi-task learning, self-attention mechanism, and dynamic loss weight adjustment, so that the model can not only predict temperature more accurately, but also correlate and analyze process parameters, thereby significantly improving the prediction effect.

[0093] S4: Input the preprocessed real-time data into the improved MLP model, and use the trained improved MLP model to predict the furnace temperature in real time to obtain the predicted temperature;

[0094] In this embodiment, a trained improved MLP model is deployed into a real-time prediction system. The model inputs are preprocessed process parameters (including dynamic features), and the outputs are predicted values for the furnace temperature and process parameters. The trained normalizers and model weights are loaded simultaneously. Pre-saved normalizers (such as scaler_X and scaler_y) and model weights (using the torch.load .pth file) are used to restore the model parameters obtained during training. Real-time furnace data (such as time, current, and process number) is then normalized to match the training data. This preprocessed real-time data is then input into the prediction system for the trained improved MLP model. The standardized and formatted real-time data is then passed as input to the prediction system for forward propagation calculations. The temperature predictions output by the model are denormalized and restored to actual temperature units (Celsius). The model input features are dynamically adjusted (e.g., by updating the sliding window data) based on the changing trends of the process parameters, improving the real-time and accuracy of the predictions.

[0095] S5: Based on the predicted temperature output by the improved MLP model, the closed-loop control module monitors the temperature inside the furnace in real time.

[0096] The closed-loop control module monitors the temperature prediction output by the improved MLP model in real time. When the predicted temperature exceeds a preset safety threshold (e.g., 2300°C), an alarm is triggered and the following actions are performed:

[0097] Alarm triggering: alerts operators through audible alarms, visual prompts or remote notifications;

[0098] Shutdown: Automatically stop the operation of the UHS ultra-fast high-temperature sintering furnace to prevent damage to the furnace or other safety accidents caused by over-temperature.

[0099] Record the trigger time of the alarm, the current predicted temperature value and related process parameters (such as current intensity, process number, etc.), and dynamic characteristics (such as sliding window data); the recorded data can be used for subsequent analysis and optimization, such as improving model performance or adjusting safety thresholds.

[0100] In addition, the closed-loop control module supports dynamic adjustment of the safety threshold to adapt to temperature control requirements under different process conditions, for example: setting a higher safety threshold in high-temperature processes; setting a lower safety threshold in sensitive processes.

[0101] The system in the ultra-fast high-temperature sintering furnace of the present invention supports multiple alarm triggering methods, including but not limited to sound alarms, visual prompts and remote notifications; after the alarm is triggered, the system automatically records relevant data (such as trigger time, temperature value, process parameters, etc.) and generates a detailed report for subsequent analysis.

[0102] In order to verify the technical effects of the present invention, experimental tests were carried out in an actual production environment; the following is a detailed description of the experimental results and effects.

[0103] (1) Experimental data

[0104] The furnace temperature and process parameter data for the past three months were selected as the training and validation sets. The data included key parameters such as time, current intensity, process number, and furnace temperature.

[0105] (2) Experimental equipment

[0106] The experiment was conducted on a UHS ultra-fast high-temperature sintering furnace, using an improved MLP model for temperature prediction and closed-loop control;

[0107] (3) Comparison method

[0108] The improved MLP model is compared with the traditional MLP model to evaluate the superiority of the present invention.

[0109] (4) Model performance comparison

[0110] Improved MLP model: Mean absolute error (MAE): 11°C, root mean square error (RMSE): 1.6°C, prediction response time: less than 0.1 seconds;

[0111] Traditional MLP model: Mean absolute error (MAE): 232°C, root mean square error (RMSE): 8.1°C, prediction response time: 0.2 seconds.

[0112] Through experimental data comparison and verification, the prediction accuracy of the improved MLP model is improved by about 30% compared with the traditional method, and the temperature fluctuation amplitude is reduced by more than 20%; the improved MLP model in the present invention is significantly superior to the traditional MLP model in both temperature prediction accuracy and response speed.

[0113] Experimental results show that the present invention can significantly optimize the stability and efficiency of the sintering process, while reducing energy consumption and improving production quality.

[0114] In addition, during the experiment, the closed-loop control module successfully triggered the alarm five times and stopped the furnace operation in time, avoiding safety accidents caused by excessive temperature; the system supports dynamic adjustment of safety thresholds and shows good adaptability under different process conditions; after the alarm is triggered, the system automatically records the trigger time, temperature value and related process parameters, and generates a detailed report for subsequent analysis.

[0115] (5) Comprehensive benefit analysis

[0116] a. Scientific research benefits: Using the improved MLP to predict temperature is beneficial to improving scientific research efficiency. The prediction and monitoring of the intracranial temperature of complex sintering furnaces can gradually move away from traditional temperature sensors (thermocouples, infrared thermometers) and thermoelectric simulation.

[0117] b. Economic benefits: By reducing furnace damage and safety accidents caused by overheating, maintenance costs can be saved by approximately RMB 500,000 per year.

[0118] c. Industrial application value: The present invention provides a reliable temperature prediction and control method for UHS ultrafast sintering equipment, which is applicable to industrial scenarios under various complex process conditions.

[0119] Compared with traditional temperature monitoring methods based on physical models, the real-time prediction and control method for the temperature inside an ultra-fast high-temperature sintering furnace of the present invention can provide more accurate and real-time temperature prediction, especially during the ultra-fast sintering process, and can accurately capture complex temperature variation patterns. By inputting real-time collected process data such as current intensity, duration and process number into the improved MLP model and combining it with historical temperature data, the improved MLP model can adaptively learn and adjust to ensure consistent and high-precision temperature prediction under different working conditions. Therefore, the present invention can effectively solve the shortcomings of traditional temperature control methods in ultra-fast sintering, meet the strict requirements for temperature prediction accuracy and response speed, and has high practicality and application value.

[0120] Example 2

[0121] See also Figure 5-Figure 7 The real-time prediction and control method of the temperature inside the ultra-fast high-temperature sintering furnace of the present invention is compared with the traditional simulation and infrared thermometer and infrared thermal imaging temperature measurement methods:

[0122] In order to verify the technical effects of the present invention, experimental tests were carried out in an actual production environment; the following is a detailed description of the experimental results and effects.

[0123] (1) Experimental data

[0124] The furnace temperature and process parameter data for the past three months were selected as the training and validation sets. The data included key parameters such as time, current intensity, process number, and furnace temperature.

[0125] (2) Experimental equipment

[0126] The experiment was carried out in a UHS ultra-fast high-temperature sintering furnace. The sintered sample was a cylindrical SiC block with a diameter of 50 mm and a thickness of 15 mm. The sintering process was 200°C / min. At the same time, a variety of temperature calibration materials were selected for temperature calibration, including silver (melting point 961°C), copper (melting point 1083°C), nickel (melting point 1455°C), palladium (melting point 1554°C), platinum (melting point 1768°C), chromium (melting point 1872°C), as well as aluminum oxide (melting point 2054°C) and boron carbide (melting point 2350°C). While ensuring the measurement stability of the sintering temperature in the UHS ultra-fast high-temperature sintering furnace, the calibration temperature fitting curve was used as a reference for temperature measurement accuracy.

[0127] (3) Experimental methods

[0128] The experiment was conducted on a UHS ultra-fast high-temperature sintering furnace, using an improved MLP model for temperature prediction and closed-loop control;

[0129] The experiment built a UHS ultra-fast high-temperature sintering furnace model on the COMSOL simulation platform, and obtained the simulated temperature through thermal field simulation analysis of thermal dielectrics;

[0130] The experiment was carried out in a UHS ultra-fast high-temperature sintering furnace, and a domestic infrared thermometer (Spirit Optoelectronics, model STRONG-SR-7032, with a range of 700-3200°C) was used to collect temperature;

[0131] The experiment was carried out in a UHS ultra-fast high-temperature sintering furnace, and the temperature was collected using an infrared thermal imaging instrument (Shenzhen Hongjia Technology, IS80, with a range of 900-3000°C).

[0132] (4) Comparison method

[0133] The time-temperature curves obtained using the improved MLP model under this process were compared with those obtained from a COMSOL simulation model, a traditional infrared thermometer, and infrared thermal imaging to evaluate the superiority of the present invention in terms of temperature prediction speed and accuracy. Due to the limited range of physical temperature measuring instruments, the temperature data obtained from the simulation was used to replace the temperature data from room temperature to 700°C or 900°C that the physical temperature measuring instruments could not collect, in order to draw a complete time-temperature curve.

[0134] (5) Comparison of experimental results

[0135] Based on the melting of the temperature calibration material, it can be seen that the temperature at each time point in the sintering process basically matches the corresponding coordinate points on the time-temperature curve of the improved MLP model, and the predicted result curve is basically consistent with the fitting curve. Comparing the improved MLP model with the COMSOL simulation results, the time-temperature curves are basically consistent, indicating that the results of the COMSOL simulation model are relatively accurate and the two curves are close to fitting;

[0136] Comparing the time-temperature curves of the temperature prediction results of the improved MLP model with the temperature measurement results of the infrared thermometer and infrared thermal imaging, it can be seen that the temperature change trends over time of the three curves are the same, but the temperature measurement effect of the physical temperature measuring instrument has a certain deviation; at the same time, there is a systematic offset along the temperature axis of the curve, that is, there is a fixed or regular difference in the temperature measurement values corresponding to the same time point. This phenomenon is a manifestation of the hysteresis of the physical temperature measuring instrument in reading and measuring temperature; finally, the physical temperature measuring instrument can be interfered with to varying degrees by external factors during the temperature measurement process, resulting in certain temperature curve fluctuations, which cannot be completely avoided.

[0137] (6) Analysis of experimental results

[0138] Experimental results show that the temperature prediction results of the improved MLP model and the temperature prediction results of COMSOL simulation are close to the actual values, and the difference between the two is not large. However, due to the large amount of preparation work required for simulation, the requirements for computer configuration and the time spent on the simulation process, the improved MLP model has more advantages.

[0139] Compared to traditional physical temperature measurement instruments, the improved MLP model demonstrates a natural advantage in real-time performance. Furthermore, the prediction process is virtually immune to external factors found in real-world experimental environments. Traditional physical temperature measurement instruments, however, suffer from hysteresis and fluctuations in temperature measurement, as well as temperature deviations between different instruments. Furthermore, they are relatively expensive. Therefore, the improved MLP model offers significant advantages in both time and cost. Furthermore, its real-time prediction results are stable and accurate, a level not found in traditional methods.

[0140] Example 3

[0141] The device for real-time prediction and control of the temperature in the ultra-fast high-temperature sintering furnace of the present invention comprises:

[0142] Data acquisition module: used to collect process parameters and temperature data in the furnace in real time;

[0143] Data processing module: used to remove noise and outliers, fill missing values, and perform standardization and normalization on the collected process parameter and temperature data;

[0144] Model prediction module: Based on the process parameters and temperature data processed by the data processing module, the furnace temperature is predicted through the built-in improved MLP model, and the prediction results are denormalized to obtain the predicted temperature;

[0145] Closed-loop control module: performs real-time monitoring of the ultra-fast high-temperature sintering furnace based on the predicted temperature of the model prediction module.

[0146] Example 4

[0147] The electronic device of the present invention includes a central processing unit and a memory, and the central processing unit is used to call and run a computer program stored in the memory to execute the steps of the real-time prediction and control method for the temperature inside the ultra-fast high-temperature sintering furnace as described in Example 1.

[0148] Example 5

[0149] The computer-readable storage medium of the present invention stores a computer program implemented by the method for real-time prediction and control of the temperature inside the ultra-fast high-temperature sintering furnace described in Example 1 in the form of computer-readable instructions. When the computer program is called and executed by a computer, the steps included in the corresponding method are executed.

[0150] The above is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited to the above content. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A method for real-time prediction and control of the temperature in an ultra-fast high-temperature sintering furnace, characterized in that: The following steps are involved: S1: Real-time collection of process parameters and temperature data in the furnace, and pre-processing of the collected process parameters and temperature data to obtain a training data set; S2: Introducing a multi-task learning framework and attention mechanism into the MLP model to construct an improved MLP model; S3: Using the training data set in step S1 to train the improved MLP model in step S2, to obtain a trained improved MLP model; S4: Input the preprocessed real-time data into the improved MLP model, and use the trained improved MLP model to predict the furnace temperature in real time to obtain the predicted temperature; S5: Based on the predicted temperature output by the improved MLP model, the closed-loop control module monitors the temperature inside the furnace in real time.

2. The method for real-time prediction and control of the temperature in an ultra-fast high-temperature sintering furnace according to claim 1, characterized in that: In step S1, the process parameters and temperature data in the furnace are collected in real time through sensors; the process parameters include time series characteristics, process number, physical characteristics and environmental parameters, wherein the time series characteristics include the current time and running time; the physical parameters include current intensity and voltage intensity; the environmental parameters include furnace pressure and atmosphere composition.

3. The method for real-time prediction and control of the temperature in an ultra-fast high-temperature sintering furnace according to claim 2, characterized in that: In step S1, the steps of pre-processing the collected process parameters and temperature data are as follows: S101: storing the collected process parameters and temperature data in a database or temporary storage, preliminarily organizing the process parameters and temperature data, and unifying the data format; S102: Cleaning the collected process parameter and temperature data to remove noise and outliers, eliminate erroneous data caused by sensor failure or environmental interference, and fill in or annotate missing data; S103: Standardizing and normalizing the process parameters and temperature data, and constructing dynamic features based on the time series characteristics of the process parameters; S104: Perform numerical encoding and embedding processing on the process number, and map it into a continuous vector space representation.

4. The method for real-time prediction and control of the temperature in an ultra-fast high-temperature sintering furnace according to claim 3, characterized in that: The steps of step S103 are: Obtain the minimum value of the process number and subtract the minimum value from all process numbers to ensure that the process numbers are in the range of 0 to N-1, where N is the number of process number types; For the data of other process parameters and temperature data except process number, the standardization method is used to make the mean of each data 0 and the standard deviation 1; Each process number is assigned an embedding vector and concatenated with the original input features to form the final input of the model.

5. The method for real-time prediction and control of the temperature in an ultra-fast high-temperature sintering furnace according to claim 4, characterized in that: In step S2, the steps for optimizing the traditional MLP model are: S201: Based on the traditional MLP model, design a network architecture of an improved MLP model, wherein the network architecture of the improved MLP model includes an input layer, a hidden layer, and an output layer; S202: Introducing a process number embedding layer into the input layer to map the process number into a low-dimensional continuous vector to capture the potential association between different process numbers; S203: Introducing a multi-task learning framework that combines the temperature prediction task with an auxiliary task for capturing the correlation between process parameters and temperature; S204: Add a self-attention layer to the hidden layer to dynamically calculate the weights of the input features; S205: Define a weighted average loss function to balance the learning objectives of the temperature prediction task and the process parameter association analysis task, and use an adaptive learning rate scheduler to optimize the training process.

6. The method for real-time prediction and control of the temperature in an ultra-rapid high-temperature sintering furnace according to claim 5, characterized in that: In step S4, the collected real-time data is fed into the trained improved MLP model, and the input features are dynamically adjusted based on the changing trends of the process parameters; The prediction results output by the improved MLP model are denormalized to obtain the final predicted temperature.

7. The method for real-time prediction and control of the temperature in an ultra-rapid high-temperature sintering furnace according to claim 6, characterized in that: In step S5, the specific steps of the closed-loop control module to monitor the temperature in the furnace in real time are as follows: Set a safety threshold. When the predicted temperature exceeds the safety threshold, the closed-loop control module will automatically trigger an alarm and stop the operation of the ultra-fast high-temperature sintering furnace; at the same time, it will record the triggering time of the alarm, the corresponding predicted temperature value, and related process parameters.

8. A device for real-time prediction and control of the temperature in an ultra-rapid high-temperature sintering furnace used in the method for real-time prediction and control of the temperature in an ultra-rapid high-temperature sintering furnace according to any one of claims 1 to 7, characterized in that: include: Data acquisition module: used to collect process parameters and temperature data in the furnace in real time; Data processing module: used for data cleaning and preprocessing of collected process parameters and temperature data; Model prediction module: Based on the process parameters and temperature data processed by the data processing module, the improved MLP model installed inside the module predicts the temperature inside the furnace and performs denormalization on the prediction results to obtain the predicted temperature; Closed-loop control module: Based on the predicted temperature output by the model prediction module, the ultra-fast high-temperature sintering furnace is monitored in real time.

9. An electronic device comprising a central processing unit and a memory, characterized in that: The central processing unit is used to call and run the computer program stored in the memory to execute the steps of the method for real-time prediction and control of the temperature in the ultra-fast high-temperature sintering furnace according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that It stores a computer program implemented according to the method for real-time prediction and control of the temperature inside the ultra-fast high-temperature sintering furnace as described in any one of claims 1 to 7 in the form of computer-readable instructions. When the computer program is called and run by a computer, the steps included in the corresponding method are executed.

Images