Method and device for detecting generic semiconductor industry operation and maintenance data abnormal equipment
By constructing an abnormal equipment detection model based on self-attention mechanism, variational autoencoder and long short-term memory network, the problem of low detection accuracy of existing methods in pan-semiconductor industrial operation and maintenance data is solved, and efficient equipment anomaly detection and prediction is achieved, which improves production efficiency and reduces maintenance costs.
Patent Information
- Application Number
- CN202510385173.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-28
- Publication Date
- 2025-10-17
AI Technical Summary
Existing statistical analysis, machine learning, and deep learning methods are difficult to adapt to complex and changeable equipment data in pan-semiconductor industrial operation and maintenance data, resulting in low accuracy in abnormal equipment detection and insufficient production efficiency.
An abnormal equipment detection model is constructed using self-attention mechanism, variational autoencoder and long short-term memory network. By preprocessing and training historical operation and maintenance data, a model training set is generated, and the model is trained using self-supervised learning method to detect equipment anomalies.
It improves the accuracy and production efficiency of equipment data anomaly detection, reduces maintenance costs, and can respond to equipment anomalies in a timely manner in complex and changing industrial environments to avoid production interruptions.
Smart Images
Figure CN120804724A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of abnormal equipment detection, and particularly relates to a detection method and device for abnormal equipment in pan-semiconductor industrial operation and maintenance data. BACKGROUND
[0002] When detecting abnormal equipment in pan-semiconductor industrial operation and maintenance data, statistical analysis, machine learning and deep learning methods are usually used. However, these methods have respective limitations. For example, a statistical analysis-based method usually relies on rules, thresholds or statistical models to identify abnormalities, and lacks adaptability to complex data patterns. In an industrial environment, equipment data usually has high nonlinearity and time sequence correlation, and it is difficult for a statistical method to capture these characteristics. For another example, a machine learning-based method usually needs a large amount of labeled data for training, and has limited processing capability for complex nonlinear relationships. For yet another example, a deep learning-based method has problems of large data demand, high training complexity and limited generalization capability. Therefore, the above methods are difficult to adapt to complex and variable industrial data environments, thereby affecting the improvement of pan-semiconductor industrial production efficiency. SUMMARY
[0003] The present application aims to at least partly solve one of the technical problems in the related art. To this end, a first object of the present application is to provide a detection method for abnormal equipment in pan-semiconductor industrial operation and maintenance data, applied to a model construction stage, the method comprising: obtaining initial historical operation and maintenance data of pan-semiconductor industrial equipment; preprocessing the initial historical operation and maintenance data to obtain historical operation and maintenance data satisfying a preset condition; generating a model training set according to the historical operation and maintenance data, and training an abnormal equipment detection model using the model training set, to construct an abnormal equipment detection model for detecting whether the pan-semiconductor industrial operation and maintenance data equipment is abnormal. The detection method of the present application trains the abnormal equipment detection model according to the historical operation and maintenance data of the pan-semiconductor industrial equipment, and uses the abnormal equipment detection model to detect whether the pan-semiconductor industrial operation and maintenance data equipment is abnormal. In this way, the abnormal situation in the equipment data can be accurately captured and identified, and the production efficiency is improved and the maintenance cost is reduced.
[0004] A second object of the present application is to provide a detection method for abnormal equipment in pan-semiconductor industrial operation and maintenance data.
[0005] A third object of the present application is to provide a detection device for abnormal equipment in pan-semiconductor industrial operation and maintenance data.
[0006] A fourth object of the present application is to provide a detection device for abnormal equipment in pan-semiconductor industrial operation and maintenance data.
[0007] A fifth object of the present application is to provide an electronic device.
[0008] To achieve the above object, the first aspect of the present application proposes a method for detecting abnormal equipment of generic semiconductor industry operation and maintenance data, applied to a model construction stage, wherein the method comprises the following steps: obtaining initial historical operation and maintenance data of generic semiconductor industry equipment; preprocessing the initial historical operation and maintenance data to obtain historical operation and maintenance data satisfying a preset condition; generating a model training set according to the historical operation and maintenance data, and training an abnormal equipment detection model using the model training set to construct an abnormal equipment detection model for detecting whether the generic semiconductor industry operation and maintenance data equipment is abnormal.
[0009] According to one embodiment of the present application, the model training set is generated according to the historical operation and maintenance data, comprising: determining a plurality of positive sample pairs based on the historical operation and maintenance data; determining a plurality of negative sample pairs based on the historical operation and maintenance data; and taking the plurality of positive sample pairs and the plurality of negative sample pairs as the model training set.
[0010] According to one embodiment of the present application, the model training set is used to train the abnormal equipment detection model, comprising: taking one or more positive sample pairs and one or more negative sample pairs in the model training set as inputs of a first attention layer to output first attention weighted feature data; taking the first attention weighted feature data as an input of a preset encoder to output first feature data; taking the first feature data as an input of a second attention layer to output second attention weighted feature data; taking the second attention weighted feature data as an input of a preset recurrent neural network to output second feature data; taking the second feature data as an input of a preset decoder to output third feature data; determining a similarity value between each positive sample pair and a similarity value between each negative sample pair according to the third feature data; calculating a discriminative loss based on the similarity value between the positive sample pairs and the similarity value between the negative sample pairs to obtain a calculation result; updating model parameters of the abnormal equipment detection model according to the calculation result until the updated abnormal equipment detection model satisfies a preset convergence condition to obtain the abnormal equipment detection model.
[0011] According to one embodiment of the present application, the variational autoencoder comprises a preset encoder and a preset decoder.
[0012] According to one embodiment of the present application, the preset recurrent neural network comprises a long short-term memory network.
[0013] To achieve the above object, the second aspect of the present application proposes a method for detecting abnormal equipment of generic semiconductor industry operation and maintenance data, applied to a model application stage, wherein the method comprises the following steps: obtaining current operation and maintenance data of generic semiconductor industry equipment; inputting the current operation and maintenance data into an abnormal equipment detection model to output an abnormal detection result of the generic semiconductor industry equipment, wherein the abnormal equipment detection model is constructed from historical operation and maintenance data of the generic semiconductor industry equipment.
[0014] According to one embodiment of the present application, the current operation and maintenance data is input into the abnormal equipment detection model to output the abnormal detection result of the general semiconductor industrial equipment, including: inputting the current operation and maintenance data into the abnormal equipment detection model to output target detection data; calculating the residual of the current operation and maintenance data and the target detection data; in the case that the residual is greater than a preset residual threshold, determining that the general semiconductor industrial equipment is abnormal.
[0015] To achieve the above-mentioned purpose, the third aspect of the present application proposes a device for detecting abnormal equipment of general semiconductor industrial operation and maintenance data, which is applied to the model construction stage. The device comprises: an acquisition module configured to acquire initial historical operation and maintenance data of a general semiconductor industrial equipment; a preprocessing module configured to preprocess the initial historical operation and maintenance data to obtain historical operation and maintenance data satisfying a preset condition; and a training module configured to generate a model training set according to the historical operation and maintenance data, and train an abnormal equipment detection model using the model training set, so as to construct an abnormal equipment detection model for detecting whether the general semiconductor industrial operation and maintenance data equipment is abnormal.
[0016] To achieve the above-mentioned purpose, the fourth aspect of the present application proposes a device for detecting abnormal equipment of general semiconductor industrial operation and maintenance data, which is applied to the model application stage. The device comprises: an acquisition module configured to acquire current operation and maintenance data of a general semiconductor industrial equipment; and a detection module configured to input the current operation and maintenance data into the abnormal equipment detection model to output an abnormal detection result of the general semiconductor industrial equipment, wherein the abnormal equipment detection model is constructed by historical operation and maintenance data of the general semiconductor industrial equipment.
[0017] To achieve the above-mentioned purpose, the fifth aspect of the present application proposes an electronic device, which comprises a memory, a processor, and a detection program of abnormal equipment of general semiconductor industrial operation and maintenance data stored in the memory and executable on the processor. When the processor executes the detection program of abnormal equipment of general semiconductor industrial operation and maintenance data, the detection method of abnormal equipment of general semiconductor industrial operation and maintenance data described above is realized.
[0018] According to the detection method and device of abnormal equipment of general semiconductor industrial operation and maintenance data, the initial historical operation and maintenance data of the general semiconductor industrial equipment is acquired, and the initial historical operation and maintenance data is preprocessed to obtain historical operation and maintenance data satisfying a preset condition. A model training set is generated according to the historical operation and maintenance data, and an abnormal equipment detection model is trained using the model training set, so as to construct an abnormal equipment detection model for detecting whether the general semiconductor industrial operation and maintenance data equipment is abnormal. The detection method of the present application trains the abnormal equipment detection model according to the historical operation and maintenance data of the general semiconductor industrial equipment, and uses the abnormal equipment detection model to detect whether the general semiconductor industrial operation and maintenance data equipment is abnormal. In this way, the abnormal situation in the equipment data can be accurately captured and identified, and the production efficiency is improved and the maintenance cost is reduced. BRIEF DESCRIPTION OF DRAWINGS
[0019] Figure 1 Flowchart of a method for detecting abnormal equipment of operation and maintenance data in a pan-semiconductor industry according to some embodiments of the present application;
[0020] Figure 2 Schematic diagram of initial historical operation and maintenance data of equipment in a pan-semiconductor industry according to some embodiments of the present application;
[0021] Figure 3 Flowchart of a method for training an abnormal equipment detection model according to some embodiments of the present application;
[0022] Figure 4 Principle diagram of a self-attention mechanism according to some embodiments of the present application;
[0023] Figure 5 Principle diagram of a preset recurrent neural network layer according to some embodiments of the present application;
[0024] Figure 6 Flowchart of a method for detecting abnormal equipment of operation and maintenance data in a pan-semiconductor industry according to some other embodiments of the present application;
[0025] Figure 7 Schematic diagram of a framework of an abnormal equipment detection model according to some embodiments of the present application;
[0026] Figure 8 Block schematic diagram of a device for detecting abnormal equipment of operation and maintenance data in a pan-semiconductor industry according to some embodiments of the present application;
[0027] Figure 9 Block schematic diagram of a device for detecting abnormal equipment of operation and maintenance data in a pan-semiconductor industry according to some other embodiments of the present application;
[0028] Figure 10 Block schematic diagram of an electronic device according to some embodiments of the present application. DETAILED DESCRIPTION
[0029] Embodiments of the present application are described in detail below with reference to the accompanying drawings, in which the same or similar components have the same or similar reference numbers throughout. The embodiments described below by reference to the drawings are exemplary and are intended to explain the present application, and should not be understood as limiting the present application.
[0030] The method and device for detecting abnormal equipment of operation and maintenance data in a pan-semiconductor industry according to embodiments of the present application are described in detail below with reference to the accompanying drawings.
[0031] Figure 1A flowchart of a method for detecting abnormal equipment in semiconductor industrial operation and maintenance data according to some embodiments of the present application. Referring to Figure 1 The method for detecting abnormal equipment in semiconductor industrial operation and maintenance data according to some embodiments of the present application can include the following steps:
[0032] S110, obtaining initial historical operation and maintenance data of the semiconductor industrial equipment.
[0033] Specifically, the initial historical operation and maintenance data of the semiconductor industrial equipment includes a production temperature time series, a humidity time series, a rotation speed time series, etc. The production temperature time series can be collected by a temperature sensor arranged on the semiconductor industrial equipment, the humidity time series can be collected by a humidity sensor arranged on the semiconductor industrial equipment, and the rotation speed time series can be collected by a rotation speed sensor arranged on the semiconductor industrial equipment. The time-effect value change curve of the initial historical operation and maintenance data collected from each sensor is as shown in Figure 2
[0034] S120, preprocessing the initial historical operation and maintenance data to obtain historical operation and maintenance data satisfying a preset condition.
[0035] Specifically, since various data have different measurement units, in order to eliminate the influence of different feature scales of data on the performance of the model, it is necessary to preprocess the initial historical operation and maintenance data, for example, to clean the initial historical operation and maintenance data, remove noise and missing values, and ensure data quality, and then use the z-score (standard score) normalization method to standardize the data to obtain historical operation and maintenance data satisfying a preset condition.
[0036] For example, the preprocessing of the initial temperature time series is described, but this is not a limitation of the present application. An initial temperature data feature matrix is determined according to the initial temperature time series of each equipment:
[0037]
[0038] wherein T represents the initial temperature data feature matrix; represents the initial temperature time series of equipment 1; represents the initial temperature time series of equipment 2; represents the initial temperature time series of equipment m; and n represents the number of temperature values in the initial temperature time series.
[0039] The initial temperature time series of each equipment in the initial temperature data feature matrix T is preprocessed by the following formula:
[0040]
[0041] wherein X z-score represents the temperature value after preprocessing; X represents the initial temperature value; μ represents the mean value of the initial temperature time series of the device; and σ represents the standard deviation of the initial temperature time series of the device.
[0042] After preprocessing the initial temperature time series of each device respectively, the temperature data feature matrix is as follows:
[0043]
[0044] wherein T' represents the temperature data feature matrix after preprocessing; represents the temperature time series after preprocessing; represents the temperature time series of device 2 after preprocessing; represents the temperature time series of device m after preprocessing.
[0045] In this way, the other initial historical operation and maintenance data is preprocessed using the above preprocessing method to determine the historical operation and maintenance data.
[0046] S130, generating a model training set according to the historical operation and maintenance data, and training an abnormal device detection model using the model training set to construct an abnormal device detection model for detecting whether the devices of the general semiconductor industry operation and maintenance data are abnormal.
[0047] Specifically, in order to detect the abnormal devices of the general semiconductor industry operation and maintenance data, an abnormal device detection model can be constructed by combining models such as self-attention mechanism, VAE (Variational Autoencoder) encoder, LSTM (Long Short-Term Memory) and the like, and the model training set is used to train the same. For example, the self-supervised learning method can be used to train the abnormal device detection model to construct an abnormal device detection model for detecting whether the devices of the general semiconductor industry operation and maintenance data are abnormal. For example, the model training set is input into the abnormal device detection model, and the feature representation is extracted through each layer in turn, the loss function is calculated according to the feature representation to obtain a calculation result, and the model parameters of the abnormal device detection model are updated according to the calculation result to construct an abnormal device detection model for detecting whether the devices of the general semiconductor industry operation and maintenance data are abnormal.
[0048] The detection method of the present application trains the abnormal device detection model according to the historical operation and maintenance data of the general semiconductor industry device, and uses the abnormal device detection model to detect whether the devices of the general semiconductor industry operation and maintenance data are abnormal. In this way, the abnormal situation in the device data can be accurately captured and identified, and the production efficiency is improved and the maintenance cost is reduced.
[0049] In some embodiments, the model training set is generated according to historical operation and maintenance data, including: determining a plurality of positive sample pairs based on the historical operation and maintenance data; determining a plurality of negative sample pairs based on the historical operation and maintenance data; and taking the plurality of positive sample pairs and the plurality of negative sample pairs as the model training set.
[0050] Specifically, a positive sample pair is usually composed of two similar samples, and taking positive samples as the model training set can help the model learn the feature representation of normal data. A negative sample pair is usually composed of a normal sample and an abnormal sample, or an abnormal sample generated through data augmentation, and taking negative samples as the model training set can help the model learn the feature representation of distinguishing normal and abnormal data. Therefore, by data labeling on the historical operation and maintenance data, normal operation and maintenance data and abnormal operation and maintenance data in the historical operation and maintenance data can be determined, and a plurality of positive sample pairs and a plurality of negative sample pairs can be generated based on the normal operation and maintenance data and the abnormal operation and maintenance data, and the plurality of positive sample pairs and the plurality of negative sample pairs can be taken as the model training set.
[0051] For example, time points t and t+1 can be selected from the normal operation and maintenance data, and the data at the two time points can be taken as a pair of positive sample pairs; or, time windows [t, t+W] and [t+W, t+2W] can also be selected from the normal operation and maintenance data, and the data in the two time windows can be taken as a pair of positive sample pairs.
[0052] If the historical operation and maintenance data includes known abnormal operation and maintenance data, the abnormal operation and maintenance data can be directly paired with the normal operation and maintenance data to form negative sample pairs. If there is not enough abnormal operation and maintenance data in the historical operation and maintenance data, abnormal samples can be generated by data augmentation on the normal operation and maintenance data, for example, noise can be added to the normal operation and maintenance data or random disturbance can be performed, and the normal samples can be paired with the abnormal samples to form negative sample pairs.
[0053] It should be noted that the determination method of the positive sample pair and the negative sample pair is not specifically limited here.
[0054] In some embodiments, training the abnormal device detection model with the model training set comprises: inputting one or more positive sample pairs and one or more negative sample pairs in the model training set as inputs of a first attention layer to output first attention weighted feature data; inputting the first attention weighted feature data as inputs of a preset encoder to output first feature data; inputting the first feature data as inputs of a second attention layer to output second attention weighted feature data; inputting the second attention weighted feature data as inputs of a preset recurrent neural network to output second feature data; inputting the second feature data as inputs of a preset decoder to output third feature data; determining similarity values between each positive sample pair and between each negative sample pair according to the third feature data; performing calculation of a discriminative loss based on the similarity values between the positive sample pairs and the similarity values between the negative sample pairs to obtain a calculation result; and updating model parameters of the abnormal device detection model according to the calculation result until the updated abnormal device detection model satisfies a preset convergence condition to obtain the abnormal device detection model.
[0055] Specifically, referring to Figure 3 , the steps of training the abnormal device detection model with the model training set are as follows:
[0056] S301, a first attention layer (input attention layer) captures similar features.
[0057] One or more positive sample pairs and one or more negative sample pairs in the model training set are organized into an input sequence, and the input sequence is input into the first attention layer, such as a self-attention mechanism layer, to calculate first attention weights, and finally obtain a weighted sum sequence representation, i.e., first attention weighted feature data. For example, referring to Figure 4 , first, the input sequence X is linearly changed through three learnable matrices Wq, Wk, and Wv to obtain corresponding query matrix Q, key matrix K, and value matrix V. Wherein, Q=XWq, K=XWk, V=Wv, the dimensions of Q, K, and V are all d k ; then, the similarity between Q and K is calculated through the formula to obtain first attention weights A, wherein is used to normalize the weights; then, the first attention weights A are processed through an activation function to obtain Softmax(A); finally, the first attention weights A are multiplied with the value matrix to obtain the first attention weighted feature data. In addition, a group of parameter matrices Wo can also be added to adjust the final output first attention weighted feature data. In this way, the attention mechanism can focus on important data and ignore or reduce irrelevant data.
[0058] S302, a preset encoder reduces dimensions.
[0059] The preset encoder is usually a neural network (for example, a convolutional neural network or other structure suitable for data feature extraction), which is used to convert high-dimensional input data into distribution parameters of a low-dimensional latent variable space, which is usually a mean vector and a variance vector (or other forms of parameterized distribution), describing the law of latent variation of the input data. Therefore, after receiving the first attention weighted feature data, the preset encoder maps it to a low-dimensional latent space to extract the latent feature representation of the data, that is, outputs the first feature data.
[0060] In S303, a second attention layer (Feature attention) captures the time sequence features.
[0061] The second attention layer is used to weight the latent feature representation (first feature data) and select key features to better capture the long-term correlation of the time sequence. The subsequent preset recurrent neural network captures the long-term dependence of the time sequence, and the accuracy is improved through the screening of the attention mechanism. The process of inputting the first feature data into the second attention layer to calculate the second attention weighted feature data has been described in detail in the above embodiment, and will not be repeated here.
[0062] In S304, a preset recurrent neural network layer captures.
[0063] The preset recurrent neural network layer can be one of LSTM, GRU (Gated Recurrent Unit) or RNN (Recurrent Neural Networks). For details, please refer to Figure 5 Suppose at time t, the d-dimensional data x (t) ∈R d×1 When input into the preset recurrent neural network, new information is memorized through memory cell state update. The hidden state and storage state of the memory cell at time t-1 are represented as y (t-1) ∈R k-1 and c (t-1) ∈R k-1 , k represents the number of preset recurrent neural network hidden nodes. In the preset recurrent neural network unit, the forget gate determines which information should be deleted from the previous memory cell c (t-1) , and its calculation expression is as follows:
[0064] f (t) =σ(W f x (t) )+R f y (t-1) +b f )
[0065] σ(x)=1 / (1+e 1-x)
[0066] where i (t) denotes the output of the input gate; σ denotes the first non-linear activation function; W f and R f ∈ R k×(d+k) denote the associated weights for the input gate, respectively; b f denotes the second bias weight.
[0067] The function of the input gate is to update the current input data and determine which data is input into the memory cell, which is expressed as follows:
[0068] i (t) = σ(W i x (t) ) + R i y (t-1) + b i )
[0069] σ(x) = 1 / (1 + e 1-x )
[0070] where i (t) denotes the output of the input gate; σ denotes the first non-linear activation function; W i and R i ∈ R k×(d+k) denote the associated weights for the input gate, respectively; b i denotes the second bias weight.
[0071] The next step is to calculate the memory cell value, which determines the current memory and the information flowing to the next layer, which is expressed as follows:
[0072] z (t) = g(W z x (t) ) + R z y (t-1) + b z )
[0073] g(x) = h(x) = tan(x)
[0074] c (t) = f (t) e c (t-1) + z(t) e i (t)
[0075] where z (t) denotes the output of the memory cell; g denotes the second non-linear activation function; h denotes the third non-linear activation function; W z and R z ∈ R k×(d+k) denote the associated weights for the memory cell, respectively; b z denotes the third bias weight; c(t) represents the storage state at time t.
[0076] The output gate is to control the data output of the model and flow from the current memory cell to the next hidden layer, and its expression is as follows:
[0077] o (t) = σ(W o x (t) + R o y (t-1) + b o
[0078] y (t) = o (t) e g(c (t) )
[0079] wherein o (t) represents the output of the output unit; W o and R o ∈ R k×(d+k) respectively represent the associated weights of the output unit; b o represents the fourth bias weight; y (t) represents the hidden state of the memory cell at time t.
[0080] The second attention weighted feature data is input into the preset recurrent neural network, and the information flow is controlled by using the gating mechanism, so as to capture the long-term dependence and space-time correlation between the data, and output the second feature data.
[0081] S305, the preset decoder is dimensioned.
[0082] The preset decoder is also a neural network, which functions to remap these latent variables back to the original high-dimensional data space, reconstruct the input data as accurately as possible, and help the model understand and generate data. Therefore, after receiving the second feature data, the preset decoder maps it back to the original high-dimensional data space to output the third feature data.
[0083] S306, the similarity value is calculated.
[0084] Illustratively, the third feature data includes feature representations of positive sample pairs and feature representations of negative sample pairs, and the cosine similarity can be used to calculate the similarity value between each positive sample pair and the similarity value between each negative sample pair.
[0085] S307, the discriminative loss is calculated.
[0086] Exemplarily, the loss function can be InfoNCE (Noise-Contrastive Estimation), which aims to make the similarity of positive sample pairs higher and the similarity of negative sample pairs lower. The calculation formula of InfoNCE is as follows:
[0087]
[0088] where Γ represents the calculation result of the loss function; sim(z i ,z j ) represents the similarity value between positive sample pairs; sim(z i ,z k ) represents the similarity value between the positive sample and all samples; 1 k≠i represents an indicator function, which ensures that the denominator does not contain itself; τ represents a temperature parameter, which controls the scaling of the similarity; and N represents the number of samples.
[0089] S308, model parameter updating.
[0090] Exemplarily, the parameters of the model are updated according to the calculation result in a backward propagation manner to minimize the loss function, until the updated abnormal equipment detection model meets a preset convergence condition, for example, the value of the loss function no longer decreases significantly, or a preset number of training rounds is reached, and finally an abnormal equipment detection model is obtained.
[0091] In this way, two layers of attention mechanisms are introduced, which can dynamically extract key features from device operation and maintenance data. The first attention layer improves the accuracy of the model by selecting the most relevant features in the spatial features of the input, and the second attention layer weights the latent variables and selects key features to better capture the long-term correlation of time series and adapt to complex industrial equipment data features.
[0092] In some embodiments, the variational autoencoder includes a preset encoder and a preset decoder.
[0093] Specifically, the VAE can learn the complex latent structure of the data to generate new data samples, and combine the new data samples with the original model training set to form a new model training set. The abnormal equipment detection model is trained using the self-supervised learning method and the new model training set, so that the loss function can use unmarked input data as the standard answer, and the model can learn an accurate and meaningful representation of the input data without labeling or annotation. To a certain extent, a large amount of labeled data can be avoided, and the difficulty of model training is reduced.
[0094] In addition, by using the variational autoencoder VAE, the latent distribution and variables behind the data are learned, effective dimension reduction and compression of high-dimensional equipment data are realized, the problem of falling into dimension disaster or needing a large amount of manual feature engineering when processing high-dimensional equipment data is solved to a certain extent, the calculation efficiency of the model is improved, and the generalization ability of the model to unknown data is enhanced.
[0095] In some embodiments, the preset recurrent neural network comprises a long short-term memory network.
[0096] In this way, the LSTM can effectively capture the time sequence pattern and complex time dependence in the equipment data. In addition, the LSTM has a memory unit that can store and update information for a long time, so it is more accurate and reliable when analyzing and predicting equipment behavior.
[0097] With reference to Figure 6 The method for detecting abnormal equipment in the pan-semiconductor industrial operation and maintenance data according to the embodiments of the present application can further include the following steps:
[0098] S210, obtaining current operation and maintenance data of the pan-semiconductor industrial equipment.
[0099] S210, inputting the current operation and maintenance data into the abnormal equipment detection model to output an abnormal detection result of the pan-semiconductor industrial equipment, wherein the abnormal equipment detection model is constructed by historical operation and maintenance data of the pan-semiconductor industrial equipment.
[0100] In some embodiments, inputting the current operation and maintenance data into the abnormal equipment detection model to output an abnormal detection result of the pan-semiconductor industrial equipment comprises: inputting the current operation and maintenance data into the abnormal equipment detection model to output target detection data; calculating the residual error of the current operation and maintenance data and the target detection data; and determining that the pan-semiconductor industrial equipment is abnormal when the residual error is greater than a preset residual error threshold. The preset residual error threshold can be determined according to actual conditions, which is not limited here.
[0101] Specifically, the current operation and maintenance data of the pan-semiconductor industrial equipment is preprocessed and input into the abnormal equipment detection model to output target detection data. The difference between the preprocessed current operation and maintenance data and the target detection data is the residual error (reconstruction score) of the current operation and maintenance data and the target detection data. By comparing the residual error with the preset residual error threshold, it is determined whether the pan-semiconductor industrial equipment is abnormal.
[0102] For example, if the residual error is greater than the preset residual error threshold, it is determined that the pan-semiconductor industrial equipment is abnormal; if the residual error is less than or equal to the preset residual error threshold, it is determined that the pan-semiconductor industrial equipment is normal.
[0103] Therefore, in the intelligent manufacturing system in the general semiconductor industry, the method can be used to optimize the production process, improve the utilization rate of equipment, and timely respond to equipment abnormalities, to a certain extent, avoid production interruption or quality problems; at the same time, the method can be associated with Internet of Things equipment monitoring, real-time monitoring and abnormal detection of connected equipment (such as sensors, machines, etc.), which helps to improve the reliability and safety of the equipment.
[0104] As a specific example, the framework of the abnormal equipment detection model is as shown in Figure 7 The data successively passes through the first layer of self-attention mechanism, VAE encoder, second layer of attention mechanism, LSTM and VAE decoder, and finally the reconstruction score is calculated to locate the abnormal equipment of the data. Among them, the VAE encoder, LSTM and VAE decoder are 3 layers respectively, but are not limited to 3 layers, which depends on the size of the operation and maintenance data. When the general semiconductor operation and maintenance data is large, the number of layers can be appropriately increased and the detailed parameters of each layer can be configured.
[0105] In summary, the detection method of the present application combines two layers of self-attention mechanism, VAE and LSTM, which can realize accurate and timely detection and positioning of industrial equipment data anomalies, and successfully applies the self-supervised learning method to the industrial operation and maintenance environment; and the method has stronger self-adaptability and higher accuracy, which can effectively capture abnormal situations in equipment data, automatically determine whether abnormal data has occurred using unlabeled raw data, and can provide more reliable anomaly detection and prediction services in complex industrial environments, thereby realizing equipment failure prevention, improving production efficiency and reducing maintenance costs.
[0106] Corresponding to the above embodiment, the present application also provides a detection device for abnormal equipment of general semiconductor industrial operation and maintenance data.
[0107] Referring to Figure 8 , the detection device for abnormal equipment of general semiconductor industrial operation and maintenance data is applied to the model construction stage, wherein the detection device for abnormal equipment of general semiconductor industrial operation and maintenance data 400 comprises: an acquisition module 410 configured to acquire initial historical operation and maintenance data of general semiconductor industrial equipment. A preprocessing module 420 is configured to preprocess the initial historical operation and maintenance data to obtain historical operation and maintenance data satisfying a preset condition. A training module 430 is configured to generate a model training set according to the historical operation and maintenance data, and train an abnormal equipment detection model using the model training set, to construct an abnormal equipment detection model for detecting whether the general semiconductor industrial operation and maintenance data equipment is abnormal.
[0108] According to an embodiment of the present application, a plurality of positive sample pairs are determined based on the historical operation and maintenance data; a plurality of negative sample pairs are determined based on the historical operation and maintenance data; and the plurality of positive sample pairs and the plurality of negative sample pairs are used as a model training set.
[0109] According to an embodiment of the present application, the training module 430 is specifically configured to take one or more positive sample pairs and one or more negative sample pairs in the model training set as inputs of the first attention layer to output first attention weighted feature data; take the first attention weighted feature data as inputs of the preset encoder to output first feature data; take the first feature data as inputs of the second attention layer to output second attention weighted feature data; take the second attention weighted feature data as inputs of the preset recurrent neural network to output second feature data; take the second feature data as inputs of the preset decoder to output third feature data; determine similarity values between each positive sample pair and similarity values between each negative sample pair according to the third feature data; perform calculation of discriminative loss based on the similarity values between the positive sample pairs and the similarity values between the negative sample pairs to obtain a calculation result; update model parameters of the abnormal device detection model according to the calculation result until the updated abnormal device detection model meets a preset convergence condition to obtain the abnormal device detection model.
[0110] According to an embodiment of the present application, the variational autoencoder comprises the preset encoder and the preset decoder.
[0111] According to an embodiment of the present application, the preset recurrent neural network comprises a long short-term memory network.
[0112] Corresponding to the above-mentioned embodiments, the present application further provides a device for detecting abnormal equipment of pan-semiconductor industrial operation and maintenance data.
[0113] Referring to Figure 9 , the device for detecting abnormal equipment of pan-semiconductor industrial operation and maintenance data is applied to a model application stage, wherein the device for detecting abnormal equipment of pan-semiconductor industrial operation and maintenance data comprises: an acquisition module 510 configured to acquire current operation and maintenance data of a pan-semiconductor industrial device. A detection module 520 is configured to input the current operation and maintenance data into an abnormal device detection model to output an abnormal detection result of the pan-semiconductor industrial device, wherein the abnormal device detection model is constructed by historical operation and maintenance data of the pan-semiconductor industrial device.
[0114] According to an embodiment of the present application, the detection module 520 is specifically configured to input the current operation and maintenance data into the abnormal device detection model to output target detection data; calculate a residual error between the current operation and maintenance data and the target detection data; and in the case that the residual error is greater than a preset residual error threshold, determine that the pan-semiconductor industrial device is abnormal.
[0115] It should be noted that the above-mentioned embodiments and beneficial effects of the method for detecting abnormal equipment of pan-semiconductor industrial operation and maintenance data are also applicable to the device for detecting abnormal equipment of pan-semiconductor industrial operation and maintenance data of the present application, and to avoid redundancy, they will not be described in detail here.
[0116] Corresponding to the above-mentioned embodiments, the present application also provides an electronic device.
[0117] Referring to Figure 10 As shown in the figure, the electronic device 600 of the present application comprises a memory 610, a processor 620, and a semiconductor industry operation and maintenance data anomaly device detection program stored in the memory 610 and executable on the processor 620. When the processor executes the semiconductor industry operation and maintenance data anomaly device detection program, the above-mentioned semiconductor industry operation and maintenance data anomaly device detection method is realized.
[0118] It should be noted that the above-mentioned embodiments and beneficial effects of the semiconductor industry operation and maintenance data anomaly device detection method are also applicable to the electronic device of the present application. To avoid redundancy, they will not be described in detail here.
[0119] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute the instructions, or in conjunction with these instruction execution systems, apparatus or devices. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by or in connection with an instruction execution system, apparatus or device, or in conjunction with these instruction execution systems, apparatus or devices. More specific examples (non-exhaustive list) of computer-readable medium include the following: electrical connections having one or more wires (electronic devices), portable computer diskettes (magnetic devices), random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memories), fiber optic devices, and portable compact disk read-only memories (CDROMs). In addition, the computer-readable medium can even be paper or other suitable medium on which the program can be printed, as the program can be electronically obtained, for example, by optical scanning of the paper or other medium, followed by electronic editing, interpretation or processing, if necessary, in other suitable ways, and then stored in a computer memory.
[0120] It should be understood that portions of the present application can be realized with a hardware, software, firmware or a combination thereof. In the above-described embodiments, a plurality of steps or methods can be realized with software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if realized with hardware, and as in another embodiment, it can be realized with any one or a combination of the following technologies known in the art: discrete logic circuit having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), etc.
[0121] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Also, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.
[0122] In addition, the terms "first", "second", "third", etc. are used only for descriptive purposes and should not be construed as indicating or implying relative importance or implying the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise explicitly specified and limited.
[0123] In the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connection", "fixing" and the like should be understood in a broad sense, for example, it can be fixed connection, or detachable connection, or integral; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, it can be the internal communication of two elements or the interaction relationship between two elements, unless otherwise explicitly limited. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.
[0124] Although the embodiments of the present application have been shown and described above, it can be understood that the above-described embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-described embodiments within the scope of the present application.
Claims
1. A method for detecting abnormal equipment in the operation and maintenance data of a pan-semiconductor industry, characterized in that: Applied to the model building stage, wherein the method comprises the following steps: Obtain initial historical operation and maintenance data of pan-semiconductor industrial equipment; Preprocessing the initial historical operation and maintenance data to obtain historical operation and maintenance data that meets preset conditions; A model training set is generated based on the historical operation and maintenance data, and an abnormal device detection model is trained using the model training set to construct an abnormal device detection model for detecting whether a pan-semiconductor industry operation and maintenance data device is abnormal.
2. The method for detecting abnormal equipment in the pan-semiconductor industry operation and maintenance data according to claim 1 is characterized in that: Generating a model training set based on the historical operation and maintenance data includes: Determining a plurality of positive sample pairs based on the historical operation and maintenance data; Determining a plurality of negative sample pairs based on the historical operation and maintenance data; The plurality of positive sample pairs and the plurality of negative sample pairs are used as model training sets.
3. The method for detecting abnormal equipment in the pan-semiconductor industry operation and maintenance data according to claim 1 is characterized in that: The abnormal device detection model is trained using the model training set, including: Taking one or more positive sample pairs and one or more negative sample pairs in the model training set as input of a first attention layer to output first attention weighted feature data; Using the first attention weighted feature data as input to a preset encoder to output first feature data; Using the first feature data as input to a second attention layer to output second attention-weighted feature data; Using the second attention-weighted feature data as input to a preset recurrent neural network to output second feature data; using the second feature data as input to a preset decoder to output third feature data; Determining a similarity value between each of the positive sample pairs and a similarity value between each of the negative sample pairs according to the third feature data; Calculating the discrimination loss based on the similarity values between the positive sample pairs and the similarity values between the negative sample pairs to obtain a calculation result; The model parameters of the abnormal device detection model are updated according to the calculation results until the updated abnormal device detection model meets a preset convergence condition, so as to obtain the abnormal device detection model.
4. The method for detecting abnormal equipment in the pan-semiconductor industry operation and maintenance data according to claim 3 is characterized in that: The variational autoencoder includes the preset encoder and the preset decoder.
5. The method for detecting abnormal equipment in the pan-semiconductor industry operation and maintenance data according to claim 3 is characterized in that: The preset recurrent neural network includes a long short-term memory network.
6. A method for detecting abnormal equipment in the operation and maintenance data of a pan-semiconductor industry, characterized in that: Applied to the model application stage, wherein the method comprises the following steps: Obtain current operation and maintenance data of pan-semiconductor industrial equipment; The current operation and maintenance data is input into an abnormal equipment detection model to output an abnormality detection result of the pan-semiconductor industrial equipment, wherein the abnormal equipment detection model is constructed by historical operation and maintenance data of the pan-semiconductor industrial equipment.
7. The method for detecting abnormal equipment in the pan-semiconductor industry operation and maintenance data according to claim 6 is characterized in that: Inputting the current operation and maintenance data into an abnormal equipment detection model to output an abnormality detection result of the pan-semiconductor industrial equipment includes: Inputting the current operation and maintenance data into an abnormal equipment detection model to output target detection data; Calculating the residual between the current operation and maintenance data and the target detection data; When the residual is greater than a preset residual threshold, it is determined that the pan-semiconductor industrial equipment is abnormal.
8. A device for detecting abnormal equipment in the operation and maintenance data of a pan-semiconductor industry, characterized in that: The device comprises: Acquisition module, used to obtain the initial historical operation and maintenance data of pan-semiconductor industrial equipment; A preprocessing module, configured to preprocess the initial historical operation and maintenance data to obtain historical operation and maintenance data that meets preset conditions; A training module is used to generate a model training set based on the historical operation and maintenance data, and use the model training set to train an abnormal device detection model to build an abnormal device detection model for detecting whether the pan-semiconductor industry operation and maintenance data equipment is abnormal.
9. A device for detecting abnormal equipment in the operation and maintenance data of a pan-semiconductor industry, characterized in that: The device comprises: Acquisition module, used to obtain the current operation and maintenance data of pan-semiconductor industrial equipment; A detection module is used to input the current operation and maintenance data into an abnormal equipment detection model to output an abnormality detection result of the pan-semiconductor industrial equipment, wherein the abnormal equipment detection model is constructed based on the historical operation and maintenance data of the pan-semiconductor industrial equipment.
10. An electronic device, characterized in that: It includes a memory, a processor, and a detection program for pan-semiconductor industrial operation and maintenance data abnormality equipment stored in the memory and runnable on the processor. When the processor executes the detection program for pan-semiconductor industrial operation and maintenance data abnormality equipment, it implements the detection method for pan-semiconductor industrial operation and maintenance data abnormality equipment according to any one of claims 1-5 or 6-7.