Sintering furnace temperature measuring device fault identification method and system based on process formula
By constructing a neural network model based on the process formula, the faults of the temperature measuring device in the silicon nitride ceramic sintering equipment are identified, which solves the problem of difficulty in identifying subtle temperature differences and occasional data distortion in the existing technology, and improves the accuracy and efficiency of sintering temperature control.
Patent Information
- Application Number
- CN202511857789.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-10
- Publication Date
- 2026-03-31
AI Technical Summary
Existing technologies make it difficult to identify faults in temperature measuring devices in silicon nitride ceramic sintering equipment in a timely manner, especially subtle temperature differences and occasional data distortions, resulting in insufficient accuracy in sintering temperature control.
A neural network model based on the process formula is constructed. By collecting sintering temperature and heating element power data, a two-dimensional feature vector of time series is generated. The BP neural network algorithm is used to identify abnormalities in the temperature measuring device, and the training dataset is updated regularly to optimize the control weights.
It enables timely identification of temperature measuring device malfunctions, improves the accuracy and efficiency of sintering temperature control, breaks through the limitations of traditional single-parameter prediction, and enhances the temperature prediction accuracy of industrial sintering furnaces.
Smart Images
Figure CN121761650A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of sintering furnace control technology, and specifically relates to a fault identification method and system for sintering furnace temperature measuring devices based on process formula. Background Technology
[0002] High-performance silicon nitride ceramic materials are the cornerstone of the power semiconductor industry. Technological and industrial innovation in their manufacturing equipment is a core element in forming new productive forces and a key support for the development of new productive forces in the power semiconductor industry. Currently, domestic silicon nitride ceramic sintering equipment lags significantly behind leading global manufacturers in terms of intelligent sintering control. With the rapid development of technologies such as computer vision and machine learning, artificial intelligence has enabled various intelligent control applications in fields such as new energy and semiconductors. The sintering process control of silicon nitride ceramic sintering equipment is also an application of control systems; therefore, introducing relevant artificial intelligence technologies could potentially lead to a qualitative leap in the sintering control of silicon nitride ceramic sintering equipment.
[0003] The silicon nitride ceramic sintering equipment needs to collect information such as temperature signals, sintering pressure, heating element voltage, and heating element resistance during the sintering process. After calculation, it outputs information such as heating element current, PID control parameters, temperature compensation, and temperature compensation (e.g., ...). Figure 1 (As shown). These parameters in the sintering process are often interconnected, and a change in one can have a ripple effect.
[0004] Meanwhile, since the sintering space of silicon nitride ceramic sintering equipment is relatively large and a large number of signals are collected internally, if a parameter is incorrect during the sintering process, it is very likely that the output parameters of the heating element will be affected. Therefore, it is very necessary to perform pattern recognition on the collected data to determine whether there are any abnormalities in the collected parameters. In particular, the temperature measuring device is very important because it directly reflects the temperature inside the furnace. However, the existing technology has the following disadvantages in detecting abnormalities in the temperature measuring device: (1) Large abnormalities in the detected temperature value can be distinguished based on experience, but it is difficult to judge subtle temperature differences; (2) It cannot identify temperature measuring device faults in a timely manner, especially for occasional temperature data distortions that cannot be detected online.
[0005] Therefore, it is necessary to develop a method for identifying faults in the temperature measuring device of the sintering furnace, which can identify the temperature measuring status of the device in a timely manner online, so as to take timely sintering strategies to deal with abnormalities; and, can identify subtle temperature distortions in the temperature measuring data of the device, thereby improving the control accuracy of the sintering temperature. Summary of the Invention
[0006] To address the above problems, one objective of this invention is to provide a fault identification method for temperature measuring devices in sintering furnaces based on process formulations, capable of promptly identifying faults in the temperature measuring devices of high-temperature sintering equipment. The specific technical solution is as follows: The fault identification method for sintering furnace temperature measuring devices based on process formulation includes the following steps: S1. Construct a data model for the sintering furnace, wherein the data model is used to output a two-dimensional feature vector with a time series based on sintering temperature data and heating element power data; S2. The process formula curve is processed through a data model to obtain the training dataset and test dataset of the neural network; the process formula curve includes a sintering temperature curve and a heating element power curve designed based on historical data; the neural network is trained and tested based on the training dataset and the test dataset, and the abnormal or normal state of the temperature measuring device is output; during the neural network training process, the temperature deviation error and the power deviation error are calculated. If one of the deviation errors exceeds the set error threshold, the training continues and the control weights are updated; otherwise, the training is completed. S3. Collect real-time temperature data of the sintering furnace and real-time power data of the heating element, process the data through the data model, and input the data into the trained neural network to identify faults in the temperature measuring device.
[0007] Furthermore, the neural network is a feedforward neural network structure.
[0008] Furthermore, the error thresholds include a temperature error threshold and a power error threshold; the temperature error threshold is a deviation of r1% from the design standard curve value, and the power error threshold is a deviation of r2% from the design standard curve value. 1= 5~8, r2=10~15.
[0009] Furthermore, the control weights are obtained by iteratively employing a BP neural network algorithm.
[0010] Furthermore, based on the sintering frequency of the sintering furnace, the training dataset is updated periodically to optimize the control weight data.
[0011] Another object of the present invention is to provide a fault identification system for a sintering furnace temperature measuring device based on a process formula, which applies the above-mentioned fault identification method. The system includes: The sensor acquisition module collects real-time temperature data of the sintering furnace and real-time power data of the heating element. The data processing module is used to construct a data model of the sintering furnace. The data model is used to output a two-dimensional feature vector with a time series based on sintering temperature data and heating element power data. The formula sintering process modeling module trains and tests the neural network based on training and testing datasets, and outputs the abnormal or normal state of the temperature measuring device. The training and testing datasets are obtained by processing the process formula curves through a data model. During neural network training, temperature deviation error and power deviation error are calculated. If either deviation error exceeds a set error threshold, training continues and the control weights are updated; otherwise, training is complete. The module receives sensor input data, inputs it into the trained neural network, and outputs the fault identification result of the temperature measuring device. The sensor input data is obtained by processing real-time temperature data and real-time power data of the heating element obtained by the sensor acquisition module through a data model.
[0012] Furthermore, the system also includes a reinforcement learning module, which periodically updates the training dataset of the formula sintering process modeling module based on the actual sintering frequency of the sintering furnace, and optimizes the control weight data.
[0013] Furthermore, the calculation process of each module in the system is carried out in a computer, and the weight data obtained from the training is transmitted to the data storage unit of the sintering furnace controller.
[0014] Compared with the prior art, one or more of the above technical solutions can achieve at least one of the following beneficial effects: 1. The sintering temperature curve and the corresponding heating element power curve are divided into two-dimensional feature vector neural network inputs with time series, and an industrial big data model of silicon nitride ceramic sintering equipment is built. 2. A method for identifying heating element failures in silicon nitride ceramic sintering equipment by collecting parameters (including process formula data) without relying on all sensors is proposed, thus overcoming the limitation of traditional neural networks that predict sintering temperature based solely on single parameter data. 3. By separating model calculations from control parameters, a breakthrough in the precision and efficiency of temperature control in industrial high-temperature sintering furnaces has been achieved. 4. Based on the sintering frequency of the industrial sintering furnace, the training data of the formula sintering process module is changed regularly and systematically, so that the accuracy of the sintering temperature prediction of the industrial sintering furnace is continuously improved over time. Attached Figure Description
[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a schematic diagram of the control parameter interface of the silicon nitride ceramic sintering equipment in Example 1.
[0017] Figure 2 This is a schematic diagram illustrating the principle of the fault identification method for the sintering furnace temperature measuring device based on the process formula in Example 1. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1
[0019] This embodiment provides a fault identification method for a sintering furnace temperature measuring device based on process formulation, including the following steps: S1. Construct a data model for the sintering furnace, wherein the data model is used to output a two-dimensional feature vector with a time series based on sintering temperature data and heating element power data; S2. The process formula curve is processed through a data model to obtain the training dataset and test dataset of the neural network; the process formula curve includes a sintering temperature curve and a heating element power curve designed based on historical data; the neural network is trained and tested based on the training dataset and the test dataset, and the abnormal or normal state of the temperature measuring device is output; during the neural network training process, the temperature deviation error and the power deviation error are calculated. If one of the deviation errors exceeds the set error threshold, the training continues and the control weights are updated; otherwise, the training is completed. S3. Collect real-time temperature data of the sintering furnace and real-time power data of the heating element, process the data through the data model, and input the data into the trained neural network to identify faults in the temperature measuring device.
[0020] In this paper, the sintering temperature curve is established based on time and the corresponding sintering furnace temperature data; the heating element power curve is established based on time and the corresponding heating element power data.
[0021] This embodiment also provides a sintering furnace heating element fault identification system based on process formulation. This embodiment takes silicon nitride ceramic sintering equipment as an example. The sintering furnace heating element fault identification system includes a sensor acquisition module, a data processing module, and a formulation sintering process modeling module.
[0022] The sensor acquisition module is used to collect data, including sintering furnace temperature data and heating element power data. In this embodiment, data is collected once every 1 second.
[0023] The data processing module is used to construct a data model of the sintering furnace. This data model outputs a two-dimensional feature vector with a time series based on sintering temperature data and heating element power data. For example... Figure 2 As shown, the sintering temperature curve and the heating element power curve are divided into 3000 time-series neural network inputs; specifically, every 30 seconds, a point T is taken from each curve. ji 、 P ji The input consists of two-dimensional feature data (T) ji P ji ),in j Number the thermocouples and their corresponding heating elements. i This is the time sequence number.
[0024] The formula sintering process modeling module trains and tests the neural network based on training and testing datasets, and outputs the abnormal or normal state of the temperature measuring device. The training and testing datasets are obtained by processing the process formula curves through a data model. During neural network training, temperature deviation error and power deviation error are calculated. If either deviation error exceeds a set error threshold, training continues and the control weights are updated; otherwise, training is complete. The module receives sensor input data, inputs it into the trained neural network, and outputs the fault identification result of the temperature measuring device. The sensor input data is obtained by processing real-time temperature data and real-time power data of the heating element obtained by the sensor acquisition module through a data model.
[0025] In one specific implementation, the initial standard for abnormal temperature measurement device settings in the training data is: T ji Deviation from the design temperature curve value r1%, r1 = 5~8, P ji The deviation from the design power curve value is r2%, where r2 = 10~15. Half of the values are marked as normal and the other half as abnormal. The output results include the normal and abnormal states of the temperature measuring device.
[0026] The training data contains 2000 process recipe curves, of which 1000 represent normal conditions and 1000 represent abnormal conditions. The test data contains 500 process recipe curves, of which 250 represent normal conditions and 250 represent abnormal conditions. The weights are iterated 5000 times. In one implementation, the neural network is a feedforward neural network. During the training and testing processes, a backpropagation (BP) neural network algorithm is used iteratively.
[0027] In practice, the module calculation process is placed on a computer with more powerful computing capabilities, and the weight data obtained from training is transmitted to the data storage unit of the sintering furnace controller.
[0028] In one implementation, the system further includes a reinforcement learning module. The reinforcement learning module is used to periodically update the training dataset of the formula sintering process modeling module based on the actual sintering frequency of the sintering furnace, and to optimize the control weight data, such as periodically updating a training data point.
[0029] Obviously, the above embodiments are merely examples to clearly illustrate the technical solutions of the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.
Claims
1. A fault identification method for sintering furnace temperature measuring devices based on process formulation, characterized in that, The method includes the following steps: S1. Construct a data model for the sintering furnace, wherein the data model is used to output a two-dimensional feature vector with a time series based on sintering temperature data and heating element power data; S2. The process formula curve is processed through a data model to obtain the training dataset and test dataset of the neural network; the process formula curve includes a sintering temperature curve and a heating element power curve designed based on historical data; the neural network is trained and tested based on the training dataset and the test dataset, and the abnormal or normal state of the temperature measuring device is output; during the neural network training process, the temperature deviation error and the power deviation error are calculated. If one of the deviation errors exceeds the set error threshold, the training continues and the control weights are updated; otherwise, the training is completed. S3. Collect real-time temperature data of the sintering furnace and real-time power data of the heating element, process the data through the data model, and input the data into the trained neural network to identify faults in the temperature measuring device.
2. The fault identification method for sintering furnace temperature measuring device based on process formula according to claim 1, characterized in that, The neural network is a feedforward neural network structure.
3. The fault identification method for sintering furnace temperature measuring device based on process formula according to claim 1, characterized in that, The error thresholds include a temperature error threshold and a power error threshold; the temperature error threshold is a deviation from the design standard curve value of r1%, and the power error threshold is a deviation from the design standard curve value of r2%, r... 1= 5~8, r2=10~15.
4. The fault identification method for sintering furnace temperature measuring device based on process formula according to claim 1, characterized in that, The control weights are obtained by iteratively using a BP neural network algorithm.
5. The fault identification method for sintering furnace temperature measuring device based on process formula according to claim 4, characterized in that, Based on the sintering frequency of the sintering furnace, the training dataset is updated regularly to optimize the control weight data.
6. A fault identification system for sintering furnace temperature measuring devices based on process formulation, employing the fault identification method for sintering furnace temperature measuring devices based on process formulation as described in any one of claims 1 to 5, characterized in that, The system includes: The sensor acquisition module collects real-time temperature data of the sintering furnace and real-time power data of the heating element. The data processing module is used to construct a data model of the sintering furnace. The data model is used to output a two-dimensional feature vector with a time series based on sintering temperature data and heating element power data. The formula sintering process modeling module trains and tests the neural network based on training and testing datasets, and outputs the abnormal or normal state of the temperature measuring device. The training and testing datasets are obtained by processing the process formula curves through a data model. During neural network training, temperature deviation error and power deviation error are calculated. If either deviation error exceeds a set error threshold, training continues and the control weights are updated; otherwise, training is complete. The module receives sensor input data, inputs it into the trained neural network, and outputs the fault identification result of the temperature measuring device. The sensor input data is obtained by processing real-time temperature data and real-time power data of the heating element obtained by the sensor acquisition module through a data model.
7. The fault identification system for sintering furnace temperature measuring device based on process formula according to claim 6, characterized in that, The system also includes a reinforcement learning module, which periodically updates the training dataset of the formula sintering process modeling module based on the actual sintering frequency of the sintering furnace, and optimizes the control weight data.
8. The fault identification system for sintering furnace temperature measuring device based on process formula according to claim 6 or 7, characterized in that, The calculation process of each module in the system is carried out in a computer, and the weight data obtained from training is transmitted to the data storage unit of the sintering furnace controller.