Silicon carbide epitaxial equipment detection method and device, electronic equipment and storage medium

Abstract

The invention belongs to the technical field of equipment data processing, and discloses a silicon carbide epitaxy equipment detection method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining historical time series data of a plurality of sensors in silicon carbide epitaxy equipment, carrying out the normalization processing and data conversion of the historical time series data, the method comprises the steps of preprocessing historical time series data of a plurality of sensors to obtain preprocessed historical time series data, constructing an anomaly detection model based on the preprocessed historical time series data through a stochastic gradient descent algorithm of adaptive learning rate adjustment, an early stop method and a model check point mechanism, and inputting the preprocessed real-time data of the plurality of sensors into the anomaly detection model. Determining whether the corresponding real-time data is abnormal or not so as to detect the silicon carbide epitaxial equipment; the silicon carbide epitaxial equipment is detected by inputting the preprocessed real-time data of the plurality of sensors into the anomaly detection model, so that the detection efficiency of the silicon carbide epitaxial equipment is improved.

Description

Technical Field

[0001] The present application relates to the technical field of device data processing, and in particular to a method, device, electronic device and storage medium for detecting silicon carbide epitaxial equipment. Background Art

[0002] During the operation of silicon carbide epitaxial equipment, the historical time series data of key parameters such as temperature, pressure, gas flow, etc. collected by sensors are extremely critical for accurately controlling the epitaxial growth process.

[0003] However, due to the complex operating environment of the equipment, changes in the characteristics of the sensor itself, and interference factors, sensor data is prone to abnormal fluctuations or deviations from normal patterns. Traditional detection methods are difficult to automatically and effectively capture abnormal features when faced with such complex data. Existing related technologies still have many defects when applied to obtaining sensor data of silicon carbide epitaxial equipment.

[0004] Therefore, in order to solve the technical problem that traditional detection methods are difficult to automatically and effectively capture abnormal features in sensor data, resulting in difficulty in detecting abnormal data, there is an urgent need for a silicon carbide epitaxial equipment detection method, device, electronic device and storage medium. Summary of the invention

[0005] The purpose of the present application is to provide a method, device, electronic device and storage medium for detecting silicon carbide epitaxial equipment. By inputting the preprocessed real-time data of multiple sensors into an anomaly detection model, it is determined whether the corresponding real-time data has an anomaly, so as to detect the silicon carbide epitaxial equipment, solve the problem that traditional detection methods are difficult to automatically and effectively capture the abnormal features in the sensor data, resulting in the difficulty in detecting abnormal data. The recognition accuracy of abnormal data is improved through the anomaly detection model, and the false alarm rate and missed alarm rate are effectively reduced. The reliability of equipment operation and the stability of epitaxial wafer production quality can be guaranteed, and the detection efficiency of silicon carbide epitaxial equipment is improved.

[0006] In a first aspect, the present application provides a method for detecting a silicon carbide epitaxial device, comprising: Obtain historical time series data from multiple sensors in a silicon carbide epitaxial device; The historical time series data is normalized by using a Z-score normalization method, and the normalized historical time series data is converted by using a sliding window technology to obtain preprocessed historical time series data; Based on the preprocessed historical time series data, an anomaly detection model is constructed by using a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism; The pre-processed real-time data of the multiple sensors are input into the abnormality detection model to determine whether the corresponding real-time data has abnormalities, so as to detect the silicon carbide epitaxial device.

[0007] The silicon carbide epitaxial equipment detection method provided in the present application can realize abnormality detection of silicon carbide epitaxial equipment. By inputting the preprocessed real-time data of multiple sensors into the abnormality detection model, it is determined whether the corresponding real-time data has an abnormality, so as to detect the silicon carbide epitaxial equipment, thereby solving the problem that traditional detection methods are difficult to automatically and effectively capture abnormal features in sensor data, resulting in difficulty in detecting abnormal data. The recognition accuracy of abnormal data is improved through the abnormal detection model, and the false alarm rate and missed alarm rate are effectively reduced. The reliability of equipment operation and the stability of epitaxial wafer production quality can be guaranteed, and the detection efficiency of silicon carbide epitaxial equipment is improved.

[0008] Optionally, the historical time series data is normalized by using a Z-score normalization method, and the normalized historical time series data is converted by using a sliding window technology to obtain preprocessed historical time series data, including: Eliminating erroneous data in the historical time series data to obtain error-processed historical time series data; Using a Z-score standardization method, the error-processed historical time series data is normalized to obtain normalized historical time series data; The normalized historical time series data is converted into a two-dimensional data matrix through a sliding window technology to obtain the pre-processed historical time series data.

[0009] The silicon carbide epitaxial equipment detection method provided in the present application can realize anomaly detection of silicon carbide epitaxial equipment. Through the Z-score standardization method and sliding window technology, the historical time series data after error processing is processed to obtain pre-processed historical time series data. The anomaly detection model constructed by the pre-processed historical time series data can improve the detection efficiency of silicon carbide epitaxial equipment.

[0010] Optionally, based on the preprocessed historical time series data, an anomaly detection model is constructed by using a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism, including: Introducing residual connections and attention mechanisms into a convolutional autoencoder network corresponding to the preprocessed historical time series data to construct a convolutional autoencoder model; Based on the preprocessed historical time series data, the convolutional autoencoder model is trained and verified through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain an anomaly detection model.

[0011] The silicon carbide epitaxial device detection method provided in the present application can realize anomaly detection of silicon carbide epitaxial devices. Through preprocessed historical time series data, the convolutional autoencoder model is trained and verified by using the stochastic gradient descent algorithm with adaptive learning rate adjustment, the early stopping method and the model checkpoint mechanism to obtain an anomaly detection model. The silicon carbide epitaxial device is detected by the anomaly detection model, which is beneficial to improving the detection efficiency of the silicon carbide epitaxial device.

[0012] Optionally, a residual connection and an attention mechanism are introduced into a convolutional autoencoder network corresponding to the preprocessed historical time series data to construct a convolutional autoencoder model, including: Constructing a convolutional autoencoder network corresponding to the preprocessed historical time series data; Based on residual connection and attention mechanism, the convolutional autoencoder network is optimized to obtain a convolutional autoencoder model.

[0013] Optionally, based on the preprocessed historical time series data, the convolutional autoencoder model is trained and verified by a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain an anomaly detection model, including: Based on the data used for training in the preprocessed historical time series data, the convolutional autoencoder model is trained by a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain a trained convolutional autoencoder model; The trained convolutional autoencoder model is verified based on the data that is not used for training in the preprocessed historical time series data to obtain an anomaly detection model.

[0014] Optionally, based on the data used for training in the preprocessed historical time series data, the convolutional autoencoder model is trained by a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain a trained convolutional autoencoder model, including: By using a stochastic gradient descent algorithm with adaptive learning rate adjustment, sequentially inputting the data used for training in the preprocessed historical time series data into the convolutional autoencoder model to train the convolutional autoencoder model; In the process of training the convolutional autoencoder model, the optimal parameters of the convolutional autoencoder model are calculated by using an early stopping method and a model checkpoint mechanism; The convolutional autoencoder model is optimized using the optimal parameters to obtain a trained convolutional autoencoder model.

[0015] Optionally, before sequentially inputting the data used for training in the preprocessed historical time series data into the convolutional autoencoder model through a stochastic gradient descent algorithm with adaptive learning rate adjustment to train the convolutional autoencoder model, the method further includes: Initialize the parameters of the convolutional autoencoder model.

[0016] In a second aspect, the present application provides a silicon carbide epitaxial equipment detection device, comprising: An acquisition module, used to acquire historical time series data of multiple sensors in a silicon carbide epitaxial device; A conversion module, used to normalize the historical time series data using a Z-score standardization method, and perform data conversion on the normalized historical time series data using a sliding window technology to obtain pre-processed historical time series data; A construction module is used to construct an anomaly detection model based on the preprocessed historical time series data through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism; A detection module is used to input the preprocessed real-time data of the multiple sensors into the abnormality detection model to determine whether the corresponding real-time data has an abnormality, so as to detect the silicon carbide epitaxial device.

[0017] The silicon carbide epitaxial equipment detection device inputs the preprocessed real-time data of multiple sensors into an anomaly detection model to determine whether the corresponding real-time data has an anomaly, so as to detect the silicon carbide epitaxial equipment, thereby solving the problem that the traditional detection method is difficult to automatically and effectively capture the abnormal features in the sensor data, resulting in the difficulty in detecting the abnormal data. The recognition accuracy of the abnormal data is improved through the anomaly detection model, and the false alarm rate and the missed alarm rate are effectively reduced. The reliability of the equipment operation and the stability of the epitaxial wafer production quality can be guaranteed, and the detection efficiency of the silicon carbide epitaxial equipment is improved.

[0018] In a third aspect, the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores a computer program executable by the processor, and when the processor executes the computer program, it runs the steps in the silicon carbide epitaxial equipment detection method described above.

[0019] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in the silicon carbide epitaxial equipment detection method described above are executed.

[0020] Beneficial effects: The silicon carbide epitaxial equipment detection method, device, electronic device and storage medium provided in the present application input the preprocessed real-time data of multiple sensors into the anomaly detection model to determine whether the corresponding real-time data has an abnormality, so as to detect the silicon carbide epitaxial equipment, thereby solving the problem that traditional detection methods are difficult to automatically and effectively capture abnormal features in sensor data, resulting in difficulty in detecting abnormal data. The accuracy of identifying abnormal data is improved through the anomaly detection model, and the false alarm rate and missed alarm rate are effectively reduced, which can ensure the reliability of equipment operation and the stability of epitaxial wafer production quality, and improve the detection efficiency of silicon carbide epitaxial equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A flow chart of a method for detecting silicon carbide epitaxial equipment provided in an embodiment of the present application.

[0022] Figure 2 A schematic diagram of the structure of a silicon carbide epitaxial equipment detection device provided in an embodiment of the present application.

[0023] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0024] Explanation of reference numerals: 1. Acquisition module; 2. Conversion module; 3. Construction module; 4. Detection module; 301. Processor; 302. Memory; 303. Communication bus. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0026] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0027] Please refer to Figure 1 , Figure 1A method for detecting silicon carbide epitaxial equipment in some embodiments of the present application is used to detect abnormalities in silicon carbide epitaxial equipment, comprising the steps of: Step S101, obtaining historical time series data of multiple sensors in a silicon carbide epitaxial device; Step S102, using the Z-score standardization method to normalize the historical time series data, and performing data conversion on the normalized historical time series data using a sliding window technology to obtain pre-processed historical time series data; Step S103, based on the preprocessed historical time series data, an anomaly detection model is constructed by using a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism; Step S104, inputting the preprocessed real-time data of the multiple sensors into the abnormality detection model to determine whether the corresponding real-time data has abnormalities, so as to detect the silicon carbide epitaxial device.

[0028] The silicon carbide epitaxial equipment detection method inputs the preprocessed real-time data of multiple sensors into an anomaly detection model to determine whether the corresponding real-time data has an anomaly, so as to detect the silicon carbide epitaxial equipment, thereby solving the problem that the traditional detection method is difficult to automatically and effectively capture the abnormal features in the sensor data, resulting in the difficulty in detecting the abnormal data. The recognition accuracy of the abnormal data is improved through the anomaly detection model, and the false alarm rate and the missed alarm rate are effectively reduced. The reliability of the equipment operation and the stability of the epitaxial wafer production quality can be guaranteed, and the detection efficiency of the silicon carbide epitaxial equipment is improved.

[0029] Specifically, in step S101, historical time series data of multiple sensors in the silicon carbide epitaxial device are obtained, wherein the multiple sensors include sensors such as temperature sensors and pressure sensors. The historical time series data is data associated with the silicon carbide epitaxial device output by the sensors within a preset period (i.e., historical operation data of the sensors). The historical time series data is arranged in chronological order to ensure the integrity and continuity of the data.

[0030] Specifically, in step S102, the Z-score standardization method is used to normalize the historical time series data, and the normalized historical time series data is converted by the sliding window technology to obtain the pre-processed historical time series data, including: Eliminate the erroneous data in the historical time series data to obtain the historical time series data after error processing; The Z-score standardization method is used to normalize the historical time series data after error processing to obtain the normalized historical time series data; Through the sliding window technology, the normalized historical time series data is converted into a two-dimensional data matrix to obtain the preprocessed historical time series data.

[0031] In step S102, it is necessary to pre-eliminate the erroneous data in the historical time series data to obtain the historical time series data after error processing. Among them, the erroneous data is data with obvious abnormalities, and the value of the erroneous data obviously does not conform to the condition range specified by the silicon carbide epitaxial growth process or the law of physical change. For example, assuming that the temperature range specified by the silicon carbide epitaxial growth process in the silicon carbide epitaxial equipment is 1500℃-1800℃, but the historical temperature obtained by the temperature sensor is 600℃, the historical temperature is erroneous data, or when the difference between the historical temperatures obtained in adjacent sequences is greater than the preset safety temperature difference threshold (the safety temperature difference threshold can be set according to actual needs), the corresponding historical temperature can be considered as erroneous data.

[0032] Calculate the mean and standard deviation of the historical time series data after error processing, and apply the Z-score standardization formula (i.e., subtract the mean from the data point and divide it by the standard deviation) to normalize each data point to obtain the normalized time series data.

[0033] Determine the size and step size of the sliding window (the size and step size of the sliding window can be set according to actual needs), apply the sliding window technology, divide the normalized time series data into multiple windows, and the data points in each window constitute a subsequence. Arrange each subsequence in chronological order to form a two-dimensional data matrix, which is a format more suitable for model processing.

[0034] Specifically, in step S103, based on the preprocessed historical time series data, an anomaly detection model is constructed by using a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism, including: Introducing residual connections and attention mechanisms into the convolutional autoencoder network corresponding to the preprocessed historical time series data to construct a convolutional autoencoder model; Based on the preprocessed historical time series data, the convolutional autoencoder model is trained and verified through the stochastic gradient descent algorithm with adaptive learning rate adjustment, early stopping method and model checkpoint mechanism to obtain the anomaly detection model.

[0035] Specifically, in step S103, residual connection and attention mechanism are introduced into the convolutional autoencoder network corresponding to the preprocessed historical time series data to construct a convolutional autoencoder model, including: Construct a convolutional autoencoder network corresponding to the preprocessed historical time series data; Based on the residual connection and attention mechanism, the convolutional autoencoder network is optimized to obtain the convolutional autoencoder model.

[0036] In step S103, the construction of the convolutional autoencoder network provides a basis for subsequent optimization, while the residual connection and attention mechanism of the feature part play a key role in solving the problem of how to optimize the model to improve the accuracy of anomaly detection. The residual connection can alleviate the problem of gradient disappearance during deep network training, and the attention mechanism can intelligently focus on the key feature areas in the data, significantly improve the sensitivity to abnormal features, and help the model better focus on important time series features.

[0037] The implementation of residual connections can include adding skip connections in each layer of the convolutional autoencoder network, which directly add the output of the previous layer to the input of the next layer, thereby effectively alleviating the gradient vanishing problem. The implementation of the attention mechanism can include introducing a self-attention module in the convolutional layer, which adjusts the importance of the features by calculating the relevance weights of the input features, so that the model can pay more attention to key features.

[0038] By introducing residual connections and attention mechanisms, the convolutional autoencoder model can be significantly optimized and the accuracy of anomaly detection can be improved.

[0039] In the convolutional autoencoder network, there are an encoder part and a decoder part.

[0040] The encoder part uses multiple convolutional layers for feature extraction, and each convolutional layer is followed by a batch normalization layer and a rectified linear unit (ReLU) activation function. Specifically, the size of the first convolution kernel can be 3×3, the number is 16, the size of the second convolution kernel can be 3×3, the number is 32, etc. This structure accelerates network convergence and enhances feature expression capabilities. The combination of residual connection and attention mechanism can further improve the expression ability and training efficiency of the encoder part.

[0041] The decoder part is symmetrical with the encoder part. It uses deconvolution layers to gradually reconstruct the original data. Each layer is also followed by a batch normalization layer and a ReLU activation function, and finally outputs a reconstructed two-dimensional data matrix. During the network training process, the reconstruction error is used as the loss function, and the training effect of the convolutional autoencoder model is determined by minimizing the reconstruction error.

[0042] Specifically, in step S103, based on the preprocessed historical time series data, the convolutional autoencoder model is trained and verified through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain an anomaly detection model, including: Based on the data used for training in the preprocessed historical time series data, the convolutional autoencoder model is trained through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain a trained convolutional autoencoder model; The trained convolutional autoencoder model is verified based on the data that is not used for training in the preprocessed historical time series data to obtain an anomaly detection model.

[0043] Specifically, in step S103, based on the data used for training in the preprocessed historical time series data, the convolutional autoencoder model is trained through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain a trained convolutional autoencoder model, including: The data used for training in the preprocessed historical time series data is sequentially input into the convolutional autoencoder model through a stochastic gradient descent algorithm with adaptive learning rate adjustment, so as to train the convolutional autoencoder model; In the process of training the convolutional autoencoder model, the optimal parameters of the convolutional autoencoder model are calculated through the early stopping method and the model checkpoint mechanism; The convolutional autoencoder model is optimized using the optimal parameters to obtain a trained convolutional autoencoder model.

[0044] In step S103, a stochastic gradient descent algorithm (such as Adagrad algorithm) with adaptive learning rate adjustment can automatically adapt to changes in the learning rate during model training, thereby accelerating convergence and improving training efficiency. Using the early stopping method and the model checkpoint mechanism, during the training process, the reconstruction error after the training data (the data used for training in the preprocessed historical time series data) is calculated in sequence to the convolutional autoencoder model, while monitoring the performance indicators (such as accuracy, recall, etc.) after the training set is input into the convolutional autoencoder model. When the performance indicator no longer improves after several iterations, the training is terminated in advance to prevent the model from overfitting, and the current model parameters are used as the optimal parameters, or when it is detected that the calculated reconstruction error no longer changes (that is, the difference between the reconstruction error calculated for multiple consecutive times and its average value is less than the preset error threshold, wherein the preset error threshold can be set according to actual needs), the current model parameters are saved as the optimal parameters. The parameters of the convolutional autoencoder model are replaced with the optimal parameters to optimize the convolutional autoencoder model to obtain the trained convolutional autoencoder model.

[0045] By combining the stochastic gradient descent algorithm with adaptive learning rate adjustment, the early stopping method and the model checkpoint mechanism, the training process of the convolutional autoencoder model is effectively optimized, and the accuracy and efficiency of anomaly detection are improved.

[0046] Specifically, in step S103, the data used for training in the preprocessed historical time series data is sequentially input into the convolutional autoencoder model through the stochastic gradient descent algorithm with adaptive learning rate adjustment, so as to train the convolutional autoencoder model, and further includes: Initialize the parameters of the convolutional autoencoder model.

[0047] In step S103, before training the model, the parameters of the convolutional autoencoder model may be initialized in advance to ensure the normal operation of the convolutional autoencoder model.

[0048] Specifically, in step S104, after preprocessing the real-time data of multiple sensors (the preprocessing step is specifically as described in the above step S102), the preprocessed real-time data is input into the anomaly detection model to determine whether there is an anomaly in the real-time data, thereby performing anomaly detection on the silicon carbide epitaxial equipment based on the abnormal real-time data to ensure the normal operation of the silicon carbide epitaxial equipment.

[0049] When real-time data is input, the anomaly detection model adopts a dynamic threshold determination method to dynamically adjust the anomaly threshold. Specifically, the kernel density estimation method is used to model the reconstruction error. During the test, the anomaly threshold is determined based on the local statistical information of the input real-time data, such as the mean and variance. When the reconstruction error of the real-time data exceeds the anomaly threshold, it is determined that the real-time data is abnormal and is determined to be abnormal data.

[0050] When abnormal data is detected, the abnormal situation can be deeply diagnosed based on the sensor to which the abnormal data belongs and the time series characteristics of the abnormal data, combined with the operating principle and fault mode library of the silicon carbide epitaxial equipment.

[0051] For example, assuming that abnormal data appears in the temperature sensor data, by comprehensively analyzing the duration of the abnormal data, the amplitude of fluctuations, and the correlation with other relevant sensor data (such as gas flow, pressure, etc.), the possible equipment failure type or process abnormality can be determined, such as heating element failure, temperature out of control caused by gas leakage, and other failures, and early warning information can be issued in time. The early warning information includes the abnormal data, the specific equipment failure type or process abnormality, the possible fault location and severity, and other detailed contents. Based on the early warning information, the staff can quickly take targeted measures to perform equipment maintenance or process adjustments.

[0052] As can be seen from the above, the silicon carbide epitaxial device detection method obtains the historical time series data of multiple sensors in the silicon carbide epitaxial device, adopts the Z-score standardization method to normalize the historical time series data, and converts the normalized historical time series data through the sliding window technology to obtain the pre-processed historical time series data. Based on the pre-processed historical time series data, an anomaly detection model is constructed through the stochastic gradient descent algorithm with adaptive learning rate adjustment, the early stopping method and the model checkpoint mechanism, and the pre-processed real-time data of multiple sensors are input into the anomaly detection model. , determine whether the corresponding real-time data has an abnormality, so as to detect the silicon carbide epitaxial equipment; thus, by inputting the preprocessed real-time data of multiple sensors into the anomaly detection model, it is determined whether the corresponding real-time data has an abnormality, so as to detect the silicon carbide epitaxial equipment, and solve the problem that the traditional detection method is difficult to automatically and effectively capture the abnormal features in the sensor data, resulting in the difficulty in detecting abnormal data. The recognition accuracy of abnormal data is improved through the anomaly detection model, and the false alarm rate and the missed alarm rate are effectively reduced, which can ensure the reliability of equipment operation and the stability of epitaxial wafer production quality, and improve the detection efficiency of silicon carbide epitaxial equipment.

[0053] refer to Figure 2 The present application provides a silicon carbide epitaxial device detection device, which is used to detect abnormalities in silicon carbide epitaxial devices, including: An acquisition module 1 is used to acquire historical time series data of multiple sensors in a silicon carbide epitaxial device; The conversion module 2 is used to normalize the historical time series data by using the Z-score standardization method, and to convert the normalized historical time series data by using the sliding window technology to obtain the pre-processed historical time series data; Building module 3, which is used to build an anomaly detection model based on the preprocessed historical time series data through the stochastic gradient descent algorithm with adaptive learning rate adjustment, early stopping method and model checkpoint mechanism; The detection module 4 is used to input the pre-processed real-time data of multiple sensors into the abnormality detection model to determine whether the corresponding real-time data has abnormalities, so as to detect the silicon carbide epitaxial equipment.

[0054] The silicon carbide epitaxial equipment detection device inputs the preprocessed real-time data of multiple sensors into an anomaly detection model to determine whether the corresponding real-time data has an anomaly, so as to detect the silicon carbide epitaxial equipment, thereby solving the problem that the traditional detection method is difficult to automatically and effectively capture the abnormal features in the sensor data, resulting in the difficulty in detecting the abnormal data. The recognition accuracy of the abnormal data is improved through the anomaly detection model, and the false alarm rate and the missed alarm rate are effectively reduced. The reliability of the equipment operation and the stability of the epitaxial wafer production quality can be guaranteed, and the detection efficiency of the silicon carbide epitaxial equipment is improved.

[0055] Specifically, when the acquisition module 1 is executed, it acquires the historical time series data of multiple sensors in the silicon carbide epitaxial device, wherein the multiple sensors include temperature sensors, pressure sensors and other sensors, and the historical time series data is the data associated with the silicon carbide epitaxial device output by the sensors within a preset period (i.e., the historical operation data of the sensors), and the historical time series data is arranged in chronological order to ensure the integrity and continuity of the data.

[0056] Specifically, when the conversion module 2 uses the Z-score standardization method to normalize the historical time series data, and performs data conversion on the normalized historical time series data through the sliding window technology to obtain the pre-processed historical time series data, it executes: Eliminate the erroneous data in the historical time series data to obtain the historical time series data after error processing; The Z-score standardization method is used to normalize the historical time series data after error processing to obtain the normalized historical time series data; Through the sliding window technology, the normalized historical time series data is converted into a two-dimensional data matrix to obtain the preprocessed historical time series data.

[0057] When the conversion module 2 is executed, it is necessary to pre-eliminate the erroneous data in the historical time series data to obtain the historical time series data after error processing. Among them, the erroneous data is data with obvious abnormalities, and the value of the erroneous data obviously does not conform to the condition range specified by the silicon carbide epitaxial growth process or the law of physical change. For example, assuming that the temperature range specified by the silicon carbide epitaxial growth process in the silicon carbide epitaxial equipment is 1500℃-1800℃, but the historical temperature obtained by the temperature sensor is 600℃, the historical temperature is erroneous data, or when the difference between the historical temperatures obtained in adjacent sequences is greater than the preset safety temperature difference threshold (the safety temperature difference threshold can be set according to actual needs), the corresponding historical temperature can be considered to be erroneous data.

[0058] Calculate the mean and standard deviation of the historical time series data after error processing, and apply the Z-score standardization formula (i.e., subtract the mean from the data point and divide it by the standard deviation) to normalize each data point to obtain the normalized time series data.

[0059] Determine the size and step size of the sliding window (the size and step size of the sliding window can be set according to actual needs), apply the sliding window technology, divide the normalized time series data into multiple windows, and the data points in each window constitute a subsequence. Arrange each subsequence in chronological order to form a two-dimensional data matrix, which is a format more suitable for model processing.

[0060] Specifically, when constructing the anomaly detection model based on the preprocessed historical time series data through the stochastic gradient descent algorithm with adaptive learning rate adjustment, the early stopping method and the model checkpoint mechanism, the construction module 3 executes: Introducing residual connections and attention mechanisms into the convolutional autoencoder network corresponding to the preprocessed historical time series data to construct a convolutional autoencoder model; Based on the preprocessed historical time series data, the convolutional autoencoder model is trained and verified through the stochastic gradient descent algorithm with adaptive learning rate adjustment, early stopping method and model checkpoint mechanism to obtain the anomaly detection model.

[0061] Specifically, when constructing module 3, residual connection and attention mechanism are introduced into the convolutional autoencoder network corresponding to the preprocessed historical time series data to construct the convolutional autoencoder model, the following steps are executed: Construct a convolutional autoencoder network corresponding to the preprocessed historical time series data; Based on the residual connection and attention mechanism, the convolutional autoencoder network is optimized to obtain the convolutional autoencoder model.

[0062] When building module 3, the convolutional autoencoder network is constructed to provide a basis for subsequent optimization, while the residual connection and attention mechanism of the feature part play a key role in solving the problem of how to optimize the model to improve the accuracy of anomaly detection. The residual connection can alleviate the problem of gradient disappearance during deep network training, and the attention mechanism can intelligently focus on the key feature areas in the data, significantly improve the sensitivity to abnormal features, and help the model better focus on important time series features.

[0063] The implementation of residual connections can include adding skip connections in each layer of the convolutional autoencoder network, which directly add the output of the previous layer to the input of the next layer, thereby effectively alleviating the gradient vanishing problem. The implementation of the attention mechanism can include introducing a self-attention module in the convolutional layer, which adjusts the importance of the features by calculating the relevance weights of the input features, so that the model can pay more attention to key features.

[0064] By introducing residual connections and attention mechanisms, the convolutional autoencoder model can be significantly optimized and the accuracy of anomaly detection can be improved.

[0065] In the convolutional autoencoder network, there are an encoder part and a decoder part.

[0066] The encoder part uses multiple convolutional layers for feature extraction, and each convolutional layer is followed by a batch normalization layer and a rectified linear unit (ReLU) activation function. Specifically, the size of the first convolution kernel can be 3×3, the number is 16, the size of the second convolution kernel can be 3×3, the number is 32, etc. This structure accelerates network convergence and enhances feature expression capabilities. The combination of residual connection and attention mechanism can further improve the expression ability and training efficiency of the encoder part.

[0067] The decoder part is symmetrical with the encoder part. It uses deconvolution layers to gradually reconstruct the original data. Each layer is also followed by a batch normalization layer and a ReLU activation function, and finally outputs a reconstructed two-dimensional data matrix. During the network training process, the reconstruction error is used as the loss function, and the training effect of the convolutional autoencoder model is determined by minimizing the reconstruction error.

[0068] Specifically, when building module 3, based on the preprocessed historical time series data, the convolutional autoencoder model is trained and verified through the stochastic gradient descent algorithm with adaptive learning rate adjustment, the early stopping method and the model checkpoint mechanism to obtain the anomaly detection model, the following is executed: Based on the data used for training in the preprocessed historical time series data, the convolutional autoencoder model is trained through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain a trained convolutional autoencoder model; The trained convolutional autoencoder model is verified based on the data that is not used for training in the preprocessed historical time series data to obtain an anomaly detection model.

[0069] Specifically, the construction module 3 trains the convolutional autoencoder model based on the data used for training in the preprocessed historical time series data through the stochastic gradient descent algorithm with adaptive learning rate adjustment, the early stopping method and the model checkpoint mechanism. When the trained convolutional autoencoder model is obtained, the following is executed: The data used for training in the preprocessed historical time series data is sequentially input into the convolutional autoencoder model through a stochastic gradient descent algorithm with adaptive learning rate adjustment, so as to train the convolutional autoencoder model; In the process of training the convolutional autoencoder model, the optimal parameters of the convolutional autoencoder model are calculated through the early stopping method and the model checkpoint mechanism; The convolutional autoencoder model is optimized using the optimal parameters to obtain a trained convolutional autoencoder model.

[0070] When building module 3 is executed, a stochastic gradient descent algorithm (such as the Adagrad algorithm) with adaptive learning rate adjustment is used to automatically adapt to changes in the learning rate during model training, thereby accelerating convergence and improving training efficiency. Using the early stopping method and the model checkpoint mechanism, during the training process, the reconstruction error after the training data (the data used for training in the preprocessed historical time series data) is calculated in sequence to the convolutional autoencoder model, while monitoring the performance indicators (such as accuracy, recall, etc.) after the training set is input into the convolutional autoencoder model. When the performance indicator no longer improves after several iterations, the training is terminated in advance to prevent the model from overfitting, and the current model parameters are used as the optimal parameters, or when it is detected that the calculated reconstruction error no longer changes (that is, the difference between the reconstruction error calculated for multiple consecutive times and its average value is less than the preset error threshold, where the preset error threshold can be set according to actual needs), the current model parameters are saved as the optimal parameters. The parameters of the convolutional autoencoder model are replaced with the optimal parameters to optimize the convolutional autoencoder model and obtain the trained convolutional autoencoder model.

[0071] By combining the stochastic gradient descent algorithm with adaptive learning rate adjustment, the early stopping method and the model checkpoint mechanism, the training process of the convolutional autoencoder model is effectively optimized, and the accuracy and efficiency of anomaly detection are improved.

[0072] Specifically, before the construction module 3 sequentially inputs the data used for training in the preprocessed historical time series data into the convolutional autoencoder model through the stochastic gradient descent algorithm with adaptive learning rate adjustment to train the convolutional autoencoder model, it further executes: Initialize the parameters of the convolutional autoencoder model.

[0073] When building module 3 is executed, before training the model, the parameters of the convolutional autoencoder model can be pre-initialized to ensure the normal operation of the convolutional autoencoder model.

[0074] Specifically, when the detection module 4 is executed, after preprocessing the real-time data of multiple sensors (the preprocessing steps are specifically described in the corresponding steps executed by the above-mentioned conversion module 2), the preprocessed real-time data is input into the anomaly detection model to determine whether there is an abnormality in the real-time data, thereby performing anomaly detection on the silicon carbide epitaxial equipment based on the abnormal real-time data to ensure the normal operation of the silicon carbide epitaxial equipment.

[0075] When real-time data is input, the anomaly detection model adopts a dynamic threshold determination method to dynamically adjust the anomaly threshold. Specifically, the kernel density estimation method is used to model the reconstruction error. During the test, the anomaly threshold is determined based on the local statistical information of the input real-time data, such as the mean and variance. When the reconstruction error of the real-time data exceeds the anomaly threshold, it is determined that the real-time data is abnormal and is determined to be abnormal data.

[0076] When abnormal data is detected, the abnormal situation can be deeply diagnosed based on the sensor to which the abnormal data belongs and the time series characteristics of the abnormal data, combined with the operating principle and fault mode library of the silicon carbide epitaxial equipment.

[0077] For example, assuming that abnormal data appears in the temperature sensor data, by comprehensively analyzing the duration of the abnormal data, the amplitude of fluctuations, and the correlation with other relevant sensor data (such as gas flow, pressure, etc.), the possible equipment failure type or process abnormality can be determined, such as heating element failure, temperature out of control caused by gas leakage, and other failures, and early warning information can be issued in time. The early warning information includes the abnormal data, the specific equipment failure type or process abnormality, the possible fault location and severity, and other detailed contents. Based on the early warning information, the staff can quickly take targeted measures to perform equipment maintenance or process adjustments.

[0078] As can be seen from the above, the silicon carbide epitaxial equipment detection device obtains the historical time series data of multiple sensors in the silicon carbide epitaxial equipment, adopts the Z-score standardization method to normalize the historical time series data, and converts the normalized historical time series data through the sliding window technology to obtain the pre-processed historical time series data. Based on the pre-processed historical time series data, an anomaly detection model is constructed through the stochastic gradient descent algorithm with adaptive learning rate adjustment, the early stopping method and the model checkpoint mechanism, and the pre-processed real-time data of multiple sensors are input into the anomaly detection model. , determine whether the corresponding real-time data has an abnormality, so as to detect the silicon carbide epitaxial equipment; thus, by inputting the preprocessed real-time data of multiple sensors into the anomaly detection model, it is determined whether the corresponding real-time data has an abnormality, so as to detect the silicon carbide epitaxial equipment, and solve the problem that the traditional detection method is difficult to automatically and effectively capture the abnormal features in the sensor data, resulting in the difficulty in detecting abnormal data. The recognition accuracy of abnormal data is improved through the anomaly detection model, and the false alarm rate and the missed alarm rate are effectively reduced, which can ensure the reliability of equipment operation and the stability of epitaxial wafer production quality, and improve the detection efficiency of silicon carbide epitaxial equipment.

[0079] Please refer to Figure 3 , Figure 3 The present invention provides a structural schematic diagram of an electronic device provided in an embodiment of the present application. The present application provides an electronic device, including: a processor 301 and a memory 302. The processor 301 and the memory 302 are interconnected and communicate with each other through a communication bus 303 and / or other forms of connection mechanisms (not shown). The memory 302 stores a computer program executable by the processor 301. When the electronic device is running, the processor 301 executes the computer program to execute the silicon carbide epitaxial device detection method in any optional implementation of the above embodiment to achieve the following functions: obtaining historical time series data of multiple sensors in the silicon carbide epitaxial device, using the Z-score standardization method to normalize the historical time series data, and performing data conversion on the normalized historical time series data through a sliding window technology to obtain pre-processed historical time series data. Based on the pre-processed historical time series data, an anomaly detection model is constructed through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism, and the pre-processed real-time data of the multiple sensors is input into the anomaly detection model to determine whether the corresponding real-time data has an anomaly, so as to detect the silicon carbide epitaxial device.

[0080] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the silicon carbide epitaxial device detection method in any optional implementation of the above-mentioned embodiment is executed to achieve the following functions: historical time series data of multiple sensors in the silicon carbide epitaxial device are obtained, the historical time series data are normalized by a Z-score normalization method, and the normalized historical time series data are converted by a sliding window technology to obtain pre-processed historical time series data; based on the pre-processed historical time series data, an anomaly detection model is constructed by a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism; the pre-processed real-time data of multiple sensors are input into the anomaly detection model to determine whether the corresponding real-time data has anomalies, so as to detect the silicon carbide epitaxial device. Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0081] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0082] In addition, the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0083] Furthermore, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0084] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0085] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting silicon carbide epitaxial equipment, for detecting abnormality of silicon carbide epitaxial equipment, characterized in that: Includes steps: Obtain historical time series data from multiple sensors in a silicon carbide epitaxial device; The historical time series data is normalized by using a Z-score normalization method, and the normalized historical time series data is converted by using a sliding window technology to obtain preprocessed historical time series data; Based on the preprocessed historical time series data, an anomaly detection model is constructed by using a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism; The pre-processed real-time data of the plurality of sensors are input into the abnormality detection model to determine whether the corresponding real-time data has abnormalities, so as to detect the silicon carbide epitaxial device.

2. The method for detecting silicon carbide epitaxial equipment according to claim 1, characterized in that: The historical time series data is normalized by using the Z-score standardization method, and the normalized historical time series data is converted by using the sliding window technology to obtain the preprocessed historical time series data, including: Eliminating erroneous data in the historical time series data to obtain error-processed historical time series data; Using a Z-score standardization method, normalizing the error-processed historical time series data to obtain normalized historical time series data; The normalized historical time series data is converted into a two-dimensional data matrix through a sliding window technology to obtain the pre-processed historical time series data.

3. The method for detecting silicon carbide epitaxial equipment according to claim 1, characterized in that: Based on the preprocessed historical time series data, an anomaly detection model is constructed through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism, including: Introducing residual connections and attention mechanisms into a convolutional autoencoder network corresponding to the preprocessed historical time series data to construct a convolutional autoencoder model; Based on the preprocessed historical time series data, the convolutional autoencoder model is trained and verified through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain an anomaly detection model.

4. The method for detecting silicon carbide epitaxial equipment according to claim 3, characterized in that: Introducing residual connections and attention mechanisms into the convolutional autoencoder network corresponding to the preprocessed historical time series data to construct a convolutional autoencoder model, including: Constructing a convolutional autoencoder network corresponding to the preprocessed historical time series data; Based on residual connection and attention mechanism, the convolutional autoencoder network is optimized to obtain a convolutional autoencoder model.

5. The method for detecting silicon carbide epitaxial equipment according to claim 3, characterized in that: Based on the preprocessed historical time series data, the convolutional autoencoder model is trained and verified by a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain an anomaly detection model, including: Based on the data used for training in the preprocessed historical time series data, the convolutional autoencoder model is trained by a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain a trained convolutional autoencoder model; The trained convolutional autoencoder model is verified based on the data that is not used for training in the preprocessed historical time series data to obtain an anomaly detection model.

6. The method for detecting silicon carbide epitaxial equipment according to claim 5, characterized in that: Based on the data used for training in the preprocessed historical time series data, the convolutional autoencoder model is trained through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism to obtain a trained convolutional autoencoder model, including: By using a stochastic gradient descent algorithm with adaptive learning rate adjustment, sequentially inputting the data used for training in the preprocessed historical time series data into the convolutional autoencoder model to train the convolutional autoencoder model; In the process of training the convolutional autoencoder model, the optimal parameters of the convolutional autoencoder model are calculated by using an early stopping method and a model checkpoint mechanism; The convolutional autoencoder model is optimized using the optimal parameters to obtain a trained convolutional autoencoder model.

7. The method for detecting silicon carbide epitaxial equipment according to claim 6, characterized in that: The method further includes: inputting the data used for training in the preprocessed historical time series data into the convolutional autoencoder model in sequence through a stochastic gradient descent algorithm with adaptive learning rate adjustment to train the convolutional autoencoder model; Initialize the parameters of the convolutional autoencoder model.

8. A silicon carbide epitaxial equipment detection device, used for detecting abnormalities of silicon carbide epitaxial equipment, characterized in that: include: An acquisition module, used to acquire historical time series data of multiple sensors in a silicon carbide epitaxial device; A conversion module, used to normalize the historical time series data using a Z-score standardization method, and perform data conversion on the normalized historical time series data using a sliding window technology to obtain pre-processed historical time series data; A construction module is used to construct an anomaly detection model based on the preprocessed historical time series data through a stochastic gradient descent algorithm with adaptive learning rate adjustment, an early stopping method, and a model checkpoint mechanism; A detection module is used to input the preprocessed real-time data of the multiple sensors into the abnormality detection model to determine whether the corresponding real-time data has an abnormality, so as to detect the silicon carbide epitaxial device.

9. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores a computer program executable by the processor, and when the processor executes the computer program, it runs the steps in the silicon carbide epitaxial equipment detection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for detecting silicon carbide epitaxial equipment according to any one of claims 1 to 7 are executed.

Images