A vertical furnace body temperature layout recommendation method

Abstract

The application provides a vertical furnace body temperature layout recommendation method, and belongs to the technical field of intelligent recommendation, and comprises the following steps: collecting the internal temperature of a furnace body and current production process parameters in real time by using sensors distributed at different positions of the furnace body, performing cleaning and preprocessing operations on the collected data, and dividing the preprocessed data into a training set and a test set according to a preset proportion; combining a back propagation algorithm and a gradient descent optimization algorithm to enable a model to learn the complex mapping relationship between various factors inside the furnace body and temperature layout; extracting features of the test set data by using the trained model, and obtaining key features of the internal temperature distribution of the furnace body through multi-layer convolution and pooling operations of the model; and performing optimization calculation on the furnace body temperature layout by using an intelligent optimization algorithm according to the key features and a preset temperature layout optimization target, to generate a recommended temperature layout scheme. The temperature control precision of the furnace body is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent recommendation technology, and in particular to a method for recommending the temperature layout of a vertical furnace body. Background Technology

[0002] In integrated circuit manufacturing, the furnace body, as a crucial heating element in vertical furnaces, directly impacts product quality and production efficiency due to the rationality of its internal temperature layout. Traditional furnace temperature control methods often rely on operator experience, making it difficult to accurately and in real-time acquire the internal temperature distribution of the furnace body. Furthermore, they cannot adjust the temperature layout promptly according to changes in production process parameters, resulting in low temperature control accuracy and significant energy waste. With the development of industrial automation and intelligence, there is a need for a method capable of real-time data acquisition, precise analysis, and the generation of a rational temperature layout scheme.

[0003] Therefore, this invention proposes a recommended method for the temperature layout of a vertical furnace body. Summary of the Invention

[0004] This invention provides a method for recommending the temperature layout of a vertical furnace body to solve the aforementioned technical problems.

[0005] This invention provides a method for recommending the temperature layout of a vertical furnace body, comprising the following steps: Step 1: Use sensors distributed at different locations in the furnace to collect the internal temperature of the furnace and the current production process parameters in real time. The production process parameters include production load, heating time and cooling time. The distribution of the sensors is optimized based on the geometry of the furnace, the arrangement of the heating elements and historical temperature distribution data. Step 2: Clean and preprocess the collected data, and divide the preprocessed data into training set and test set according to a preset ratio. The preset ratio is dynamically adjusted according to the scale of the data and the training requirements of the model. Step 3: Train the model using the training set data, combining the backpropagation algorithm and the gradient descent optimization algorithm to allow the model to learn the complex mapping relationship between various factors inside the furnace and the temperature layout. The model is a deep learning model containing multi-layer neural networks. Step 4: Use the trained model to extract features from the test set data. Through multi-layer convolution and pooling operations of the model, obtain the key features of the internal temperature distribution of the furnace body. The key features include the temperature change trend, temperature gradient distribution, and the correlation between temperatures in different regions. Step 5: Based on the key features and the preset temperature layout optimization target, use an intelligent optimization algorithm to optimize the furnace body temperature layout and generate a recommended temperature layout scheme. The temperature layout scheme includes: power adjustment suggestions for each heating element and expected effect evaluation and adjustment strategies. Preferably, before real-time acquisition of the furnace internal temperature, the following steps are included: Establish a three-dimensional geometric model of the furnace body and analyze the heat conduction path and possible hot spots in the temperature distribution inside the furnace body; Based on the arrangement of the heating elements, preliminary candidate locations for the sensors are determined; Based on historical temperature distribution data, spatial interpolation algorithms and error analysis methods are used to calculate the contribution of each candidate location to temperature information collection. The actual installation location of the sensor is chosen based on its high contribution and ability to cover different temperature ranges.

[0006] Preferably, a regularization layer is added to the deep learning model; at the same time, the training parameters are dynamically adjusted according to the characteristics of the training set data and the training effect of the model.

[0007] Preferably, the preset temperature layout optimization objective includes multiple different optimization objectives, wherein each optimization objective corresponds to a different evaluation index and weight coefficient; during the optimization calculation process, a multi-objective optimization algorithm is used to generate a set of optimal solutions; Based on actual production needs and using the weighting coefficient method, the most suitable recommended temperature layout scheme is selected from the optimal solution to achieve comprehensive optimization of the furnace body temperature layout.

[0008] Preferably, before collecting data using sensors, the following steps are included: The sensor is calibrated and standardized. By comparing it with standard temperature sources and process parameter measuring equipment, the measurement error and drift characteristics of the sensor are determined, and the sensor measurement data is corrected in real time. At the same time, a fault detection and diagnosis mechanism for sensors is established. By monitoring the output signal of the sensors in real time, it is determined whether the sensors have malfunctioned. Once a sensor malfunction is detected, redundant sensor replacement or data repair measures are taken for adjustment.

[0009] Preferably, when performing optimization calculations on the furnace body temperature layout, the heat conduction characteristics and thermal inertia inside the furnace body are determined, a heat conduction model of the furnace body is established, and the conduction and diffusion process of temperature inside the furnace body is simulated by combining heat transfer principles and numerical calculation methods, and the changing trend of furnace body temperature layout under different power adjustment schemes is predicted. The prediction results of the heat conduction model are combined with the output results of the trained deep learning model for comprehensive optimization calculation to optimize the temperature layout scheme. Preferably, after generating the recommended temperature layout scheme, the following is also included: The recommended temperature layout scheme is applied to the actual furnace body. Sensors are used to monitor the temperature distribution inside the furnace body and the changes in production process parameters in real time, and the deviation between the actual temperature layout and the recommended scheme is calculated. Based on the deviation, a feedback control algorithm is used to adjust the power of the heating element in real time, so that the furnace temperature layout gradually approaches the recommended scheme. At the same time, an evaluation index system for the effectiveness of the scheme is established to evaluate the recommended scheme from multiple dimensions of indicators, and the model and optimization algorithm are improved and optimized based on the evaluation results.

[0010] Preferably, after cleaning and preprocessing the collected data, the process further includes: For the time series data Ti={tiu,u=1,2,3,...,n} of each sensor node si, a spatiotemporal correlation matrix C is defined, where the matrix element Cij in matrix C represents the reciprocal normalized value of the spatial distance between sensor nodes si and sj; Within the time window Wt, calculate the dynamic threshold for each data point; Multi-sensor time series data is modeled as a three-dimensional tensor, wherein the three-dimensional tensor is related to the number of sensors, time points, and process parameter dimensions; The three-dimensional tensor is decomposed using regularized tensor and combined with a dynamic threshold. The missing data points are then recovered by using alternating least squares method to obtain new data. For each feature dimension f, calculate the local density based on the new data and the distance to the nearest high-density point, and define a normalization function; The optimal transport theory is introduced to achieve cross-dimensional distribution equilibration, and the model input is standardized by combining a normalization function to obtain the standardized sample S.

[0011] Preferably, after obtaining the standardized sample S, the process further includes: Establish a partitioning optimization model based on minimizing conditional entropy, with the objective function as follows:

[0012] Where C represents the process parameter category, For conditional entropy; To divide the proportions; This is a sample of standardized data; The key feature subset Fkey is selected through mutual information analysis. K-means clustering is then performed in the Fkey space to ensure that the sample proportions of the training set Strain and the test set Stest in each process parameter stratification are consistent with the overall population, and satisfy the following: ,

[0013] in, , This is the threshold for balanced distribution. , f represents the mean of features based on the training set, the mean of features based on the sample set, and the standard deviation, respectively; c represents the process parameters.

[0014] Preferably, a normalization function is defined, including:

[0015] in, The target normalization interval is based on the feature dimension f; For local density based on feature dimension f; The distance is the nearest high-density point distance based on the feature dimension f; is the normalization function; min is the minimum value symbol; max is the maximum value symbol.

[0016] Compared with the prior art, the beneficial effects of this application are as follows: By optimizing the layout of sensors to achieve precise data acquisition from the furnace body, and combining data cleaning with deep learning models to uncover the complex correlation between temperature and process parameters, intelligent algorithms are used to generate power regulation schemes that balance temperature uniformity and energy consumption. After implementation, this significantly improves the accuracy of furnace temperature control, reduces product defect rates caused by temperature unevenness, and achieves intelligent temperature layout optimization from experience-driven to data-driven approaches.

[0017] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a method for recommending the temperature layout of a vertical furnace body in an embodiment of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] This invention provides a recommended method for temperature layout of a vertical furnace body, such as... Figure 1 As shown, it includes the following steps: Step 1: Use sensors distributed at different locations in the furnace to collect the internal temperature of the furnace and the current production process parameters in real time. The production process parameters include production load, heating time and cooling time. The distribution of the sensors is optimized based on the geometry of the furnace, the arrangement of the heating elements and historical temperature distribution data. Step 2: Clean and preprocess the collected data, and divide the preprocessed data into training set and test set according to a preset ratio. The preset ratio is dynamically adjusted according to the scale of the data and the training requirements of the model. Step 3: Train the model using the training set data, combining the backpropagation algorithm and the gradient descent optimization algorithm to allow the model to learn the complex mapping relationship between various factors inside the furnace and the temperature layout. The model is a deep learning model containing multi-layer neural networks. Step 4: Use the trained model to extract features from the test set data. Through multi-layer convolution and pooling operations of the model, obtain the key features of the internal temperature distribution of the furnace body. The key features include the temperature change trend, temperature gradient distribution, and the correlation between temperatures in different regions. Step 5: Based on the key features and the preset temperature layout optimization target, use an intelligent optimization algorithm to optimize the furnace body temperature layout and generate a recommended temperature layout scheme. The temperature layout scheme includes: power adjustment suggestions for each heating element and expected effect evaluation and adjustment strategies. Preferably, a regularization layer is added to the deep learning model; at the same time, the training parameters are dynamically adjusted according to the characteristics of the training set data and the training effect of the model.

[0022] In this embodiment, the sensor is a device used to collect physical quantities (temperature, pressure, etc.) inside the furnace in real time. The non-electrical signal is converted into an electrical signal through a sensitive element. For example, K-type thermocouples (temperature sensors) are arranged on the top, side walls and bottom of the furnace. An infrared thermal imager monitors the surface temperature of the furnace chamber. Pressure sensors are installed at the air inlet and outlet. For example, according to the geometry of the furnace (such as a cylindrical furnace divided into 3 layers and 8 columns of grid), the sensors are densely arranged near the heating elements (resistance wire, gas nozzle) and in areas with large historical temperature fluctuations (such as corners). The optimal measuring point is determined by ANSYS Fluent to simulate the thermal flow field.

[0023] In this embodiment, the production process parameters reflect key indicators of the production process status, including production load (processing volume per unit time), heating time (single heating cycle), and cooling time (time to cool down to a specified temperature). For example, the production load is expressed as the number of workpieces in the furnace (e.g., 500 bearings per furnace), the heating time is set according to the workpiece material (e.g., alloy steel requires 3 hours of heating), and the cooling time is controlled by the valve opening of the water cooling system (the target temperature is 40 minutes to drop from 1200℃ to 500℃). Specifically, the production line speed, heating power supply start-up and shutdown time, cooling water pump operating frequency, and other signals are collected in real time by the PLC (Programmable Logic Controller) and converted into standard process parameter format.

[0024] In this embodiment, the optimized sensor distribution is based on the furnace structure, heating element location, and historical data. An algorithm is used to determine the sensor installation positions to maximize data validity. For example, using one year's worth of historical temperature data, the furnace body is divided into a grid (100×100×50 grid). The temperature variance of each grid is calculated, and sensors are placed in the center of the top 20% of the grids with the highest variance to ensure coverage of the area with the largest temperature gradient.

[0025] In this embodiment, data cleaning and preprocessing is the process of removing noisy data, repairing missing values, and standardizing data formats, including outlier detection, missing value imputation, and normalization, for example: Outlier detection: A sensor temperature reading suddenly changes to -100℃ (outside the physical range). The Z-score method (mean ±3σ) is used to identify and mark the outlier.

[0026] Missing value imputation: If the cooldown time data for a certain period is missing, it is repaired by linear interpolation of the data one hour before and after.

[0027] Normalization: Temperature (0-1500℃) and production load (0-1000 units) are scaled uniformly to the [0,1] interval, formula: x′=max(x)-min(x)x-min(x). Implementation: Data cleaning is performed using the Python Pandas library, and normalization is completed using Scikit-learn's StandardScaler.

[0028] In this embodiment, the training set / test set partitioning involves dividing the data into two parts: a training set for model learning and a test set for evaluating the model's generalization ability. The ratio is typically 7:3 or 8:2. For example, 365 days of data are collected, with the first 265 days used as the training set and the last 100 days as the test set to ensure the temporal sequence of the data. Alternatively, stratified sampling can be used, dividing the data equally according to high / medium / low production load. In this embodiment, the deep learning model is a multi-layered neural network, such as a convolutional neural network (CNN) or a recurrent neural network (RNN), used to learn complex mappings of high-dimensional data. For example, a 3-layer CNN model can be built, with the input layer receiving sensor data (100 dimensions), the intermediate layers extracting spatial temperature correlation features through convolutional kernels (5×5 size), and the fully connected layer outputting predicted temperature distribution values ​​(20 dimensions, corresponding to 20 regions of the furnace). The model is built using TensorFlow / Keras, employing the Adam optimizer, and the loss function is mean squared error (MSE).

[0029] In this embodiment, the backpropagation algorithm calculates the gradient of the model's prediction error with respect to the parameters of each layer, and updates the weights using a chain rule to reduce the prediction error. For example, if the predicted temperature of a certain region is 50°C higher than the actual value, the error propagates back along the output layer → hidden layer → input layer, adjusting the weights of the convolutional and fully connected layers to make the next prediction value closer to the true value. In this embodiment, the gradient descent optimization algorithm updates the parameters in the opposite direction of the loss function gradient to find the global minimum. For example, the initial learning rate is set to 0.001, and when the validation set loss does not decrease for 5 consecutive epochs, the learning rate decays to 0.1 times to avoid getting trapped in local optima.

[0030] In this embodiment, feature extraction extracts the most critical information for the target task (temperature layout optimization) from the raw data. For example, the first convolutional kernel of a CNN extracts local temperature gradients (such as temperature differences between adjacent sensors), the second layer extracts temperature change trends (heating / cooling rates), and pooling layers (such as max pooling) retain key features and reduce dimensionality. In this embodiment, the temperature change trend is the rate of temperature change per unit time (°C / min), reflecting the dynamic characteristics of the heating or cooling process. For example, a certain area heats up at 10°C / min during the heating stage and cools down at 15°C / min during the cooling stage. The trend feature is used to determine the current process stage.

[0031] In this embodiment, the temperature gradient distribution is the temperature difference in different areas of the furnace. The larger the gradient, the worse the temperature uniformity. For example, the temperature at the center of the furnace is 1200℃ and the temperature at the edge is 1100℃, with a gradient of 100℃ / m. Hot and cold areas are identified through gradient features.

[0032] In this embodiment, the intelligent optimization algorithm simulates the heuristic algorithm of natural phenomena (such as genetics and particle swarm optimization) to solve multi-objective optimization problems. For example, the genetic algorithm is used to optimize the power of heating elements. Chromosomes are encoded as the power values ​​of each element (0-100%). The fitness function combines temperature uniformity (60%) and energy consumption (40%). The optimal solution is generated through selection, crossover and mutation.

[0033] In this embodiment, the power adjustment suggestion is to provide a specific adjustment scheme for the power output of each heating element (such as resistance wire and gas valve) based on the optimization results. For example, if the temperature on the left side of the furnace is too low, it is recommended to increase the power of the three heating elements in that area from 80% to 90%, and reduce the power of the area on the right side with a higher temperature from 85% to 75%.

[0034] In this embodiment, the expected effect assessment predicts indicators such as temperature uniformity, energy consumption, and production efficiency after the implementation of the optimized scheme. A trained model simulates the temperature distribution after power adjustment, calculates uniformity indicators (such as standard deviation) and predicted energy consumption values, and compares them with the optimization target.

[0035] The beneficial effects of the above technical solution are as follows: Accurate data acquisition from the furnace body is achieved through optimized sensor layout; the complex correlation between temperature and process parameters is uncovered by combining data cleaning and deep learning models; and intelligent algorithms are used to generate a power regulation scheme that balances temperature uniformity and energy consumption. After implementation, this significantly improves the accuracy of furnace temperature control, reduces product defect rates caused by temperature unevenness, and achieves intelligent temperature layout optimization from experience-driven to data-driven approaches.

[0036] This invention provides a method for recommending the temperature layout of a vertical furnace body, which includes, before real-time acquisition of the internal temperature of the furnace body: Establish a three-dimensional geometric model of the furnace body and analyze the heat conduction path and possible hot spots in the temperature distribution inside the furnace body; Based on the arrangement of the heating elements, preliminary candidate locations for the sensors are determined; Based on historical temperature distribution data, spatial interpolation algorithms and error analysis methods are used to calculate the contribution of each candidate location to temperature information collection. The actual installation location of the sensor is chosen based on its high contribution and ability to cover different temperature ranges.

[0037] In this embodiment, the three-dimensional geometric model is constructed using computer-aided design (CAD) technology to create a three-dimensional digital model of the furnace body, accurately reproducing geometric parameters such as the furnace structure, heating element layout, insulation layer thickness, and inlet / outlet positions. For example, the three-dimensional model of a heat treatment furnace includes a cylindrical furnace chamber (2m in diameter, 3m in height), three sets of annular resistance heating strips at the top (0.5m apart), a ceramic insulation layer at the bottom (10cm thick), and four workpiece inlet / outlet ports on the side (0.8m × 0.6m in size). In this embodiment, the heat conduction path is the path through which heat is transferred within the furnace body via conduction (furnace wall material), convection (gas flow within the furnace), and radiation (between heating elements and workpieces). For example, the heat from the resistance heating band is transferred to the workpiece in the middle of the furnace through radiation, and then to the furnace wall through conduction. The hot gas in the furnace circulates to the low-temperature zone at the bottom through convection. For instance, in Fluent, boundary conditions are set (heating band temperature 1200℃, furnace wall thermal conductivity 1.5W / (m•K)) to simulate the steady-state heat flow field, output heat flux density distribution cloud map, and identify the main heat conduction paths (such as the vertical heat flow from the top of the furnace to the bottom of the furnace).

[0038] In this embodiment, hotspot areas are regions within the furnace where the temperature is significantly higher than the average level (typically exceeding the average temperature by more than 10%) due to concentrated heat conduction and poor heat dissipation. Examples include: 10cm directly below the heating zone (temperature 1150℃, average temperature 1000℃), and gaps between stacked workpieces (local temperature 1200℃ due to airflow obstruction). For instance, based on thermal flow field simulation results, regions with temperatures > mean + 1.5σ are extracted as candidate hotspots (σ being the temperature standard deviation) and marked as red grid areas.

[0039] In this embodiment, the arrangement of heating elements—such as resistance wires, gas nozzles, and electromagnetic induction coils—directly determines the heat source distribution within the furnace body. For example, the annular resistance wires are evenly distributed at the top of the furnace (spaced 20cm apart), and the gas nozzles are symmetrically installed on both sides of the furnace (1.2m from the bottom, at a 45° angle). In this embodiment, the initial candidate locations are based on the heating element positions and geometric symmetry, with preset sensor installation points (usually grid nodes or feature points) in areas potentially sensitive to temperature distribution. For instance, one layer is set every 0.5m along the furnace axial direction (0-3m height), and one point is set every 30° in the radial direction (0-1m radius) of each layer, generating a total of 6 layers × 12 points = 72 candidate locations. Emphasis is placed on increasing the density near the heating elements (e.g., the top two layers are increased to one layer every 0.2m).

[0040] In this embodiment, the historical temperature distribution data is time-series temperature data of various regions of the furnace body obtained through temporary sensors or historical production records (typically containing data from at least 100 production cycles). For example, the production data of a furnace body over the past 3 months includes the minute-by-minute temperature values ​​of each temporary measuring point (50 in total) (e.g., at 8:00 AM on January 1, 2023, the temperature at measuring point (1,0,1.5) is 1020℃). In this embodiment, the spatial interpolation algorithm is a mathematical method for estimating the temperature at unknown locations using known measuring point data, such as the inverse distance weighted method (IDW) and Kriging. In this embodiment, the error analysis method evaluates the impact of sensor location on the accuracy of temperature field reconstruction, with commonly used indicators being root mean square error (RMSE) and mean absolute error (MAE).

[0041] In this embodiment, the candidate points with the highest contribution have the greatest impact on the accuracy of temperature field reconstruction (the top 30% of the candidate points). For example, among the 72 candidate points, the top 20 points with a contribution > 0.8 are such as directly below the heating belt (contribution 0.92), the central axis of the furnace body (contribution 0.88), and near the discharge port (contribution 0.85).

[0042] In this embodiment, covering different temperature regions ensures that the sensor distribution covers hot spots (high temperature), cold spots (low temperature), and transitional regions (medium temperature), avoiding concentration in a single temperature range. For example, candidate points are divided into three categories based on the average temperature: 8 points are selected in the high temperature region (>1100℃), 10 points are selected in the medium temperature region (900-1100℃), and 2 points are selected in the low temperature region (<900℃) (such as near the furnace bottom insulation layer).

[0043] The beneficial effects of the above technical solution are: by accurately locating the key path and hot spot area of ​​heat conduction in the furnace body through three-dimensional modeling and heat flow simulation, and by pre-setting sensor candidate points in combination with the layout of heating elements, the contribution of each location to temperature acquisition is quantified by using historical data and spatial interpolation algorithms, and finally, efficient measuring points covering the entire temperature gradient are selected.

[0044] This invention provides a method for recommending the temperature layout of a vertical furnace body. The preset temperature layout optimization objectives include multiple different optimization objectives, each of which corresponds to a different evaluation index and weight coefficient. During the optimization calculation process, a multi-objective optimization algorithm is used to generate a set of optimal solutions. Based on actual production needs and using the weighting coefficient method, the most suitable recommended temperature layout scheme is selected from the optimal solution to achieve comprehensive optimization of the furnace body temperature layout.

[0045] In this embodiment, the preset temperature layout optimization objectives are multiple parallel optimization directions set according to production process requirements. These typically include interrelated and potentially conflicting objectives such as temperature uniformity, energy efficiency, production efficiency, and equipment lifespan. For example, the optimization objectives for an aluminum alloy melting furnace include: Objective A: Temperature uniformity (temperature difference between different areas within the furnace ≤ 50℃, ensuring uniform alloy composition); Objective B: Minimizing energy consumption (power consumption per unit output < 800 kWh / ton); Objective C: Maximizing heating rate (time from room temperature to 1000℃ < 90 minutes, increasing production capacity). In this embodiment, evaluation indicators are specific parameters that quantify the degree of achievement of optimization objectives. Each objective corresponds to 1-2 calculable indicators. In this embodiment, weight coefficients reflect the relative importance of each optimization objective (between 0 and 1, summing to 1), determined by production priority. In this embodiment, the multi-objective optimization algorithm is a heuristic algorithm capable of handling multiple conflicting objectives simultaneously, searching for non-dominated solution sets (Pareto fronts) by simulating natural selection (e.g., genetic algorithms) or swarm intelligence (e.g., particle swarm optimization). In this embodiment, the optimal solution (Pareto optimal solution) is the solution in the multi-objective space that cannot improve any objective without reducing the performance of other objectives, thus forming a Pareto front. In this embodiment, the weighting coefficient method transforms the multi-objective problem into a single-objective problem, calculating the comprehensive score through weighted summation. In this embodiment, actual production needs are considered by dynamically adjusting the weights based on current production tasks (such as product model, delivery date, and energy cost) to select the most suitable solution.

[0046] The beneficial effects of the above technical solution are: by constructing a multi-dimensional optimization objective system, using advanced algorithms such as NSGA-II to generate Pareto optimal solutions covering different priorities, and using the weight coefficient method to achieve objective-oriented solution selection, the problem that traditional single-objective optimization cannot take into account complex production needs is solved.

[0047] This invention provides a method for recommending the temperature layout of a vertical furnace body, which includes, before collecting data using sensors: The sensor is calibrated and standardized. By comparing it with standard temperature sources and process parameter measuring equipment, the measurement error and drift characteristics of the sensor are determined, and the sensor measurement data is corrected in real time. At the same time, a fault detection and diagnosis mechanism for sensors is established. By monitoring the output signal of the sensors in real time, it is determined whether the sensors have malfunctioned. Once a sensor malfunction is detected, redundant sensor replacement or data repair measures are taken for adjustment.

[0048] In this embodiment, sensor calibration is achieved by comparing the sensor output signal with the actual physical quantities (temperature, pressure, time, etc.) by comparing it with standard temperature sources of known accuracy (such as blackbody furnaces and platinum resistance thermometers) and process parameter measuring devices (such as high-precision pressure transmitters and timers), thereby determining the deviation between the sensor output signal and the actual physical quantities. An error correction model is then established.

[0049] In this embodiment, the measurement error is the difference between the sensor measurement value and the true value, including systematic errors (such as fixed deviations caused by uncompensated thermocouple cold junction temperature) and random errors (such as fluctuations caused by electromagnetic interference).

[0050] Drift characteristics refer to the trend of sensor performance changing over time or in the environment (such as long-term high temperature causing thermocouple material aging), with the output signal gradually deviating from the initial calibration value (e.g., monthly temperature drift of +0.5℃).

[0051] In this embodiment, real-time data correction utilizes the error model obtained from calibration to correct the data collected by the sensor in real time online, eliminating the effects of system errors and drift, and outputting a signal that is closer to the true value.

[0052] In this embodiment, the fault detection and diagnosis mechanism monitors abnormal characteristics of the sensor output signal in real time (such as over-range, sudden change, or long-term instability), and combines historical data and the correlation with adjacent sensors to determine whether the sensor has malfunctioned (such as open circuit, short circuit, or performance failure). For example, threshold detection: if the temperature sensor output exceeds the upper limit of the range (such as a K-type thermocouple output >75mV, corresponding to >1372℃, while the furnace body's maximum temperature is designed to be 1300℃), it is determined to be an over-range fault; correlation analysis: if the temperature value of one of the three sensors in the same temperature measurement area deviates from the average value of the other two by more than 10% for 5 minutes, it is determined to be abnormal (possibly due to poor sensor contact).

[0053] In this embodiment, redundant sensor replacement: when the main sensor fails, a backup sensor at the same location (hardware redundancy) or sensor data from an adjacent location (software redundancy) is temporarily used as a replacement. Data repair: for missing or unreliable data during the failure period, repair is performed using interpolation methods (such as linear interpolation, cubic splines) or model prediction (such as LSTM-based time series prediction).

[0054] The beneficial effects of the above technical solution are: by calibrating and extruding sensors, correcting errors in real time, detecting faults and repairing them, a data reliability assurance system is built for the entire process.

[0055] This invention provides a method for recommending the temperature layout of a vertical furnace body. When optimizing the temperature layout of the furnace body, the heat conduction characteristics and thermal inertia inside the furnace body are determined, a heat conduction model of the furnace body is established, and the conduction and diffusion process of temperature inside the furnace body is simulated by combining the principles of heat transfer and numerical calculation methods, and the changing trend of the furnace body temperature layout under different power adjustment schemes is predicted. The prediction results of the heat conduction model are combined with the output results of the trained deep learning model for comprehensive optimization calculation to optimize the temperature layout scheme. In this embodiment, thermal conductivity characteristics refer to the physical laws governing heat transfer within the furnace, including three basic modes: thermal conduction, thermal convection, and thermal radiation.

[0056] In this embodiment, thermal inertia is the ability of the furnace material to impede temperature changes.

[0057] In this embodiment, the heat conduction model is a mathematical model based on the principles of heat transfer.

[0058] In this embodiment, the numerical calculation method is to discretize the continuous heat conduction model into calculable numerical values.

[0059] In this embodiment, temperature conduction simulation and trend prediction are performed by calculating the evolution of the furnace temperature field over time under different power adjustment schemes (such as increasing the heating element power from 80% to 90%) using a heat conduction model, and predicting trends such as hot spot movement and temperature gradient changes.

[0060] In this embodiment, the output of the deep learning model is a neural network model (such as CNN or LSTM) trained with historical data. The input is sensor data and process parameters, and the output is a predicted value of temperature distribution or key features (such as temperature gradient and trend).

[0061] In this embodiment, the comprehensive optimization calculation combines the physical constraints of the heat conduction model (such as heat flow conservation and material heat capacity limitation) with the data-driven prediction of the deep learning model, and generates a more reliable temperature layout scheme through weighted fusion, joint optimization and other methods.

[0062] The beneficial effects of the above technical solution are: by integrating physical mechanism modeling and data-driven prediction, a temperature layout optimization system of mechanism constraint + data enhancement is constructed.

[0063] This invention provides a method for recommending the temperature layout of a vertical furnace body. After generating the recommended temperature layout scheme, the method further includes: The recommended temperature layout scheme is applied to the actual furnace body. Sensors are used to monitor the temperature distribution inside the furnace body and the changes in production process parameters in real time, and the deviation between the actual temperature layout and the recommended scheme is calculated. Based on the deviation, a feedback control algorithm is used to adjust the power of the heating element in real time, so that the furnace temperature layout gradually approaches the recommended scheme. At the same time, an evaluation index system for the effectiveness of the scheme is established to evaluate the recommended scheme from multiple dimensions of indicators, and the model and optimization algorithm are improved and optimized based on the evaluation results.

[0064] In this embodiment, the temperature layout scheme is a specific control strategy generated through prior optimization calculations. It includes the power parameters of each heating element (e.g., "Heating zone A power 90%, heating zone B power 85%"), the expected temperature distribution (e.g., "furnace center temperature 1100±20℃"), and the adjustment strategy (e.g., "uniformity is checked every 10 minutes"). Specifically, temperature and process parameters (production load, heating time, etc.) are collected at high frequency (e.g., once / second) by sensors (e.g., thermocouples, infrared thermometers) distributed in the furnace body, and the current furnace status is fed back in real time. For example, a recommended scheme for an annealing furnace requires the top heating zone power to be 75% and the bottom heating zone power to be 80%, with a target temperature distribution of "temperature difference between the top and bottom of the furnace body ≤30℃, average temperature 850℃". Monitoring implementation: 5 thermocouples are arranged at the top and bottom of the furnace body to collect temperature data in real time (e.g., top measuring point T1=865℃, bottom measuring point T6=830℃), and the real-time power of the heating zone is collected by the PLC (top 74.8%, bottom 80.2%).

[0065] In this embodiment, deviation calculation involves quantitatively comparing the actual temperature distribution (such as temperature values ​​at each measuring point and temperature standard deviation) with the target value of the recommended scheme. Commonly used indicators include: Absolute deviation: Recommended based on actual measurements (single-point temperature difference); Root Mean Square Deviation (RMSE): Measured Recommendation (Overall Temperature Deviation); Uniformity deviation: Recommended based on actual measurements (temperature standard deviation difference).

[0066] Feedback control algorithm: A closed-loop control method that automatically adjusts the power of the heating element based on the deviation.

[0067] In this embodiment, the effect evaluation index system is a set of indicators that quantify the effectiveness of the implementation of the plan from multiple dimensions. Typical dimensions include: Temperature control accuracy: RMSE, maximum single-point deviation; Energy efficiency: energy consumption per unit output (kWh / ton), power regulation frequency (reflecting equipment wear and tear); Production quality: workpiece temperature pass rate (the proportion of measured temperatures within the process window), defect rate (such as the crack rate caused by uneven temperature); Robustness: The stability of the solution under equipment aging (e.g., 5% power reduction of heating elements) and environmental changes (e.g., room temperature fluctuations of ±10℃).

[0068] Model and algorithm improvement involves iteratively optimizing deep learning models and algorithms based on evaluation results. For example, if the temperature prediction RMSE is consistently greater than 20°C, increase the model training data or adjust the network structure; if the optimization algorithm frequently gets stuck in local optima, replace it with a multi-objective algorithm (such as changing from NSGA-II to MOEA / D).

[0069] The beneficial effects of the above technical solution are: through the closed-loop control of implementation-monitoring-adjustment-evaluation-improvement, a continuous iterative system for optimizing temperature layout is constructed.

[0070] This invention provides a method for recommending the temperature layout of a vertical furnace body, which, after cleaning and preprocessing the collected data, further includes: For the time series data Ti={tiu,u=1,2,3,...,n} of each sensor node si, a spatiotemporal correlation matrix C is defined, where the matrix element Cij in matrix C represents the inverse normalized value of the spatial distance between sensor nodes si and sj; Within the time window Wt, calculate the dynamic threshold for each data point; Multi-sensor time series data is modeled as a three-dimensional tensor, wherein the three-dimensional tensor is related to the number of sensors, time points, and process parameter dimensions; The three-dimensional tensor is decomposed using regularized tensor and combined with a dynamic threshold. The missing data points are then recovered by using alternating least squares method to obtain new data. For each feature dimension f, calculate the local density based on the new data and the distance to the nearest high-density point, and define a normalization function; The optimal transport theory is introduced to achieve cross-dimensional distribution equilibration, and the model input is standardized by combining a normalization function to obtain the standardized sample S.

[0071] Preferably, after obtaining the standardized sample S, the process further includes: Establish a partitioning optimization model based on minimizing conditional entropy, with the objective function as follows:

[0072] Where C represents the process parameter category, For conditional entropy; To divide the proportions; This is a sample of standardized data; The key feature subset Fkey is selected through mutual information analysis. K-means clustering is then performed in the Fkey space to ensure that the sample proportions of the training set Strain and the test set Stest in each process parameter stratification are consistent with the overall population, and satisfy the following: ,

[0073] in, , This is the threshold for balanced distribution. , f represents the mean of features based on the training set, the mean of features based on the sample set, and the standard deviation, respectively; c represents the process parameters.

[0074] Preferably, a normalization function is defined, including:

[0075] in, The target normalization interval is based on the feature dimension f; For local density based on feature dimension f; The distance is the nearest high-density point distance based on the feature dimension f; is the normalization function; min is the minimum value symbol; max is the maximum value symbol.

[0076] In this embodiment, sensor node si is the physical location identifier of monitoring sensors such as temperature, pressure, and time deployed on the furnace body. Each node continuously collects time series data to record the temperature change of node si over time. For example, the hourly temperature data collected by thermocouple s1 at the top of the furnace body is: T1=[850,852,848,...,860] (unit: °C).

[0077] In this embodiment, the matrix element Cij in the spatiotemporal correlation matrix C is the normalized value of the reciprocal of the spatial distance between sensor nodes si and sj (the closer the distance, the higher the correlation weight), which characterizes the spatial correlation between sensors. The furnace temperature distribution has spatial continuity, and the correlation matrix provides spatial constraints for subsequent tensor modeling, avoiding isolated analysis of individual sensors.

[0078] In this embodiment, the time window Wt is a sliding time interval (e.g., 10 minutes) used to locally analyze the dynamic characteristics of the data (e.g., calculating the temperature fluctuation threshold within the window). For example, within Wt=[t,t+600] (seconds), the temperature standard deviation σt of sensor si is calculated as the basis for the dynamic threshold.

[0079] In this embodiment, the dimension of the three-dimensional tensor is a data structure of number of sensors × time point × process parameters (e.g., 100 sensors × 1000 time points × 5 process parameters), which integrates the correlation information of spatiotemporal and process dimensions. Because the traditional two-dimensional matrix loses the correlation between dimensions (e.g., the coupling relationship of "time-temperature-pressure"), the tensor more completely retains the multi-dimensional interaction features and improves the data modeling capability.

[0080] In this embodiment, regularized tensor decomposition involves adding a regularization term (such as the L2 norm) to tensor decomposition (such as Tucker decomposition) to constrain the decomposition process. Combined with a dynamic threshold (the fluctuation range within a time window), the missing data points are recovered by iteratively solving the problem using alternating least squares.

[0081] In this embodiment, local density refers to the density of the neighbors around a data point under feature dimension f (such as the number of points within Euclidean distance), reflecting the crowding of the local distribution. Distance to the nearest high-density point refers to the distance from the current point to a point with higher density under the same dimension, characterizing the isolation of the data.

[0082] In this embodiment, mutual information measures the degree of correlation between features and process parameters (e.g., the higher the mutual information between temperature and the "quenching" process, the more critical the temperature is to the process), and Fkey is used to filter out the features with the top-N mutual information (e.g., removing noisy features that are irrelevant to the process).

[0083] The beneficial effects of the above technical solution are: based on the entire process of multi-source industrial data preprocessing → feature optimization → balanced partitioning, it solves the problems of data missing, dimensional redundancy and uneven distribution in industrial scenarios.

[0084] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for recommending the temperature layout of a vertical furnace body, characterized in that, Includes the following steps: Step 1: Use sensors distributed at different locations in the furnace to collect the internal temperature of the furnace and the current production process parameters in real time. The production process parameters include production load, heating time and cooling time. The distribution of the sensors is optimized based on the geometry of the furnace, the arrangement of the heating elements and historical temperature distribution data. Step 2: Clean and preprocess the collected data, and divide the preprocessed data into training set and test set according to a preset ratio. The preset ratio is dynamically adjusted according to the scale of the data and the training requirements of the model. Step 3: Train the model using the training set data, combining the backpropagation algorithm and the gradient descent optimization algorithm to allow the model to learn the complex mapping relationship between various factors inside the furnace and the temperature layout. The model is a deep learning model containing multi-layer neural networks. Step 4: Use the trained model to extract features from the test set data. Through multi-layer convolution and pooling operations of the model, obtain the key features of the internal temperature distribution of the furnace body. The key features include the temperature change trend, temperature gradient distribution, and the correlation between temperatures in different regions. Step 5: Based on the key features and the preset temperature layout optimization target, use an intelligent optimization algorithm to optimize the furnace body temperature layout and generate a recommended temperature layout scheme. The temperature layout scheme includes: power adjustment suggestions for each heating element and expected effect evaluation and adjustment strategy. When optimizing the furnace body temperature layout, the heat conduction characteristics and thermal inertia inside the furnace body are determined, a heat conduction model of the furnace body is established, and the conduction and diffusion process of temperature inside the furnace body is simulated by combining the principles of heat transfer and numerical calculation methods, and the changing trend of furnace body temperature layout under different power adjustment schemes is predicted. The prediction results of the heat conduction model are combined with the output results of the trained deep learning model for comprehensive optimization calculation to optimize the temperature layout scheme. This process, including cleaning and preprocessing the collected data, also includes: For the time series data Ti={tiu,u=1,2,3,...,n} of each sensor node si, a spatiotemporal correlation matrix A is defined, where the matrix element Aij in matrix A represents the reciprocal normalized value of the spatial distance between sensor nodes si and sj; Within the time window Wt, calculate the dynamic threshold for each data point; Multi-sensor time series data is modeled as a three-dimensional tensor, wherein the three-dimensional tensor is related to the number of sensors, time points, and process parameter dimensions; The three-dimensional tensor is decomposed using regularized tensor and combined with a dynamic threshold. The missing data points are then recovered by using alternating least squares method to obtain new data. For each feature dimension f, calculate the local density based on the new data and the distance to the nearest high-density point, and define a normalization function; The optimal transport theory is introduced to achieve cross-dimensional distribution equilibration, and the model input is standardized by combining a normalization function to obtain the standardized sample S.

2. The method for recommending the temperature layout of a vertical furnace body according to claim 1, characterized in that, Before real-time acquisition of the furnace's internal temperature, the following steps are included: Establish a three-dimensional geometric model of the furnace body and analyze the heat conduction path and possible hot spots in the temperature distribution inside the furnace body; Based on the arrangement of the heating elements, preliminary candidate locations for the sensors are determined; Based on historical temperature distribution data, spatial interpolation algorithms and error analysis methods are used to calculate the contribution of each candidate location to temperature information collection. The actual installation location of the sensor is chosen based on its high contribution and ability to cover different temperature ranges.

3. The method for recommending the temperature layout of a vertical furnace body according to claim 1, characterized in that, Regularization layers are added to the deep learning model; at the same time, training parameters are dynamically adjusted based on the characteristics of the training set data and the training effect of the model.

4. The method for recommending the temperature layout of a vertical furnace body according to claim 1, characterized in that, The preset temperature layout optimization objectives include multiple different optimization objectives, each of which corresponds to a different evaluation index and weight coefficient; during the optimization calculation process, a multi-objective optimization algorithm is used to generate a set of optimal solutions; Based on actual production needs and using the weighting coefficient method, the most suitable recommended temperature layout scheme is selected from the optimal solution to achieve comprehensive optimization of the furnace body temperature layout.

5. The method for recommending the temperature layout of a vertical furnace body according to claim 1, characterized in that, Before using sensors to collect data, including: The sensor is calibrated and standardized. By comparing it with standard temperature sources and process parameter measuring equipment, the measurement error and drift characteristics of the sensor are determined, and the sensor measurement data is corrected in real time. At the same time, a fault detection and diagnosis mechanism for sensors is established. By monitoring the output signal of the sensors in real time, it is determined whether the sensors have malfunctioned. Once a sensor malfunction is detected, redundant sensor replacement or data repair measures are taken for adjustment.

6. The method for recommending the temperature layout of a vertical furnace body according to claim 1, characterized in that, After generating the recommended temperature layout scheme, the following is also included: The recommended temperature layout scheme is applied to the actual furnace body. Sensors are used to monitor the temperature distribution inside the furnace body and the changes in production process parameters in real time, and the deviation between the actual temperature layout and the recommended scheme is calculated. Based on the deviation, a feedback control algorithm is used to adjust the power of the heating element in real time, so that the furnace temperature layout gradually approaches the recommended scheme. At the same time, an evaluation index system for the effectiveness of the scheme is established to evaluate the recommended scheme from multiple dimensions of indicators, and the model and optimization algorithm are improved and optimized based on the evaluation results.

7. The method for recommending the temperature layout of a vertical furnace body according to claim 1, characterized in that, After obtaining the standardized sample S, the following is also included: Establish a partitioning optimization model based on minimizing conditional entropy, with the objective function as follows: Where C represents the process parameter category, For conditional entropy; To divide the proportions; This is a sample of standardized data; The key feature subset Fkey is selected through mutual information analysis. K-means clustering is then performed in the Fkey space to ensure that the sample proportions of the training set Strain and the test set Stest in each process parameter stratification are consistent with the overall population, and satisfy the following: , in, , This is the threshold for balanced distribution. , f represents the mean of features based on the training set, the mean of features based on the sample set, and the standard deviation, respectively; c represents the process parameters.

8. The method for recommending the temperature layout of a vertical furnace body according to claim 1, characterized in that, Define a normalization function, including: in, The target normalization interval is based on the feature dimension f; For local density based on feature dimension f; The distance is the nearest high-density point distance based on the feature dimension f; is the normalization function; min is the minimum value symbol; max is the maximum value symbol.

Images