CN114417248B - DCGAN-based linear contour process quality anomaly monitoring method and system - Google Patents
DCGAN-based linear contour process quality anomaly monitoring method and system Download PDFInfo
- Publication number
- CN114417248B CN114417248B CN202210062233.9A CN202210062233A CN114417248B CN 114417248 B CN114417248 B CN 114417248B CN 202210062233 A CN202210062233 A CN 202210062233A CN 114417248 B CN114417248 B CN 114417248B
- Authority
- CN
- China
- Prior art keywords
- data
- linear contour
- quality map
- neural network
- convolutional neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 136
- 230000008569 process Effects 0.000 title claims abstract description 103
- 238000012544 monitoring process Methods 0.000 title claims abstract description 47
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 37
- 230000002159 abnormal effect Effects 0.000 claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 29
- 238000001514 detection method Methods 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 15
- 230000003321 amplification Effects 0.000 claims description 13
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 13
- 238000013507 mapping Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 11
- 230000005856 abnormality Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 4
- 238000001819 mass spectrum Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 abstract description 3
- 238000011176 pooling Methods 0.000 description 18
- 238000004519 manufacturing process Methods 0.000 description 14
- 238000010586 diagram Methods 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000009826 distribution Methods 0.000 description 6
- 230000004913 activation Effects 0.000 description 5
- 210000002569 neuron Anatomy 0.000 description 5
- 230000004044 response Effects 0.000 description 5
- 239000013598 vector Substances 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Pure & Applied Mathematics (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Probability & Statistics with Applications (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention relates to a linear contour process quality anomaly monitoring method and system based on DCGAN, comprising an offline training stage and an online monitoring stage, wherein in the offline training stage, linear contour process data under a historical normal operation state and an abnormal state are respectively acquired, samples are amplified, a convolutional neural network is trained based on the amplified samples, in the online monitoring stage, process anomaly detection is carried out according to the convolutional neural network after training, the problem of limited manual extraction characteristics can be solved by applying deep convolution generation to monitoring of the linear contour process quality anomaly, monitoring efficiency of the linear contour process quality anomaly can be improved, historical data can be trained under the condition of an unbalanced data set, abnormal process identification accuracy with small sample data volume is guaranteed, and the efficiency of linear contour process quality anomaly monitoring can be effectively improved by using the convolutional neural network.
Description
Technical Field
The invention relates to a linear contour process quality anomaly monitoring method and system based on DCGAN.
Background
Sensors for measuring equipment parameters and quality characteristics such as displacement, pressure, temperature and humidity, and dimensions are installed in large numbers in modern production processes, and process data collected by the sensors gradually progress from single values in the past to complex time series or space series data. When these data characterizing the process operating state can be approximated by a linear model, they are referred to as linear profiles. How to find the internal operation rule of the system from a large amount of linear contour data and construct a real-time intelligent quality monitoring model suitable for the modern production process becomes a key problem of ensuring the safe production of enterprises and reducing the quality cost.
In the linear contour monitoring related research, a monitoring method based on a control chart and an intelligent model is most common. The control diagram-based monitoring method generally uses a single-element or multi-element control diagram to monitor parameters such as slope, intercept, root mean square error and the like of the linear contour, the used control diagram mainly comprises T 2, EWMA, CUSUM, K, MEWMA control diagrams and the like, although the process deviation can be effectively monitored, the position of a variable point is difficult to estimate when the small fluctuation is alarmed, and in order to compensate the deficiency, the variable point model is widely applied to the linear contour monitoring. Because the equipment parameters and quality characteristics required to be monitored in the automatic production process are numerous, timely response is required, the data types are various and have strong uncertainty, the monitoring method based on the control chart is required to establish an accurate mathematical model, and in general, the inter-mapping relation between the state types and the abnormal reasons cannot be reflected by supposing that the measured values in the group are mutually independent, so that the requirements of real-time quality monitoring and diagnosis in the high-speed manufacturing process are difficult to meet.
With the development of computer and artificial intelligence technology, intelligent quality monitoring methods such as support vector machines and artificial neural networks have been widely studied because of the capability of more accurately expressing process states. Compared with statistical methods such as control charts, the intelligent method has the characteristics of no need of modeling, self-learning, no need of determining controlled parameters and the like in contour monitoring, and has stronger practicability. However, the related research of the contour monitoring based on the intelligent method at present uses a shallow model which does not exceed two layers of nonlinear feature transformation, and the problems of limited manual extraction features, insufficient classification precision, insufficient generalization capability and the like still exist.
Disclosure of Invention
In view of the above, the present invention provides a method and a system for monitoring quality anomalies in linear profile process based on DCGAN in order to solve the above-mentioned technical problems.
The invention adopts the following technical scheme:
A linear contour process quality anomaly monitoring method based on DCGAN comprises the following steps:
Acquiring linear contour process data in a historical normal running state and an abnormal state respectively, and marking corresponding label information for the linear contour process data;
carrying out gray level mapping on the linear contour process data to obtain an initial quality map;
generating an countermeasure network based on the depth convolution, and performing sample data amplification on the linear contour process data in an abnormal state to obtain an amplification quality map;
obtaining a target quality map according to the initial quality map and the amplification quality map;
Training a convolutional neural network according to the target quality map;
And performing process anomaly detection on the actually acquired data based on the convolutional neural network after training.
Further, the performing gray level mapping on the linear contour process data to obtain an initial quality map includes:
normalizing the linear contour process data to obtain a normalized matrix;
And carrying out gray level mapping on the normalized matrix to obtain an initial quality map.
Further, the calculating process of the initial quality map includes:
the linear profile process data is represented as a data matrix of:
The calculation formula of the initial mass spectrum is as follows:
Wherein M represents the number of sensors, N represents the number of measurements acquired by each sensor, j=1, 2, … …, N, k=1, 2, … …, M; round () represents a round function.
Further, the deep convolution generating countermeasure network comprises a generator and a discriminator, wherein the generator is used for generating a false data set according to the linear contour process data in an abnormal state, the false data set and the real data set are used as inputs of the discriminator, and the discriminator outputs the probability that the obtained data belongs to the real data set so as to judge whether the sample is a real sample or a false sample.
Further, the data actually collected are obtained through a sliding window value taking mode.
Further, the training-completed convolutional neural network-based process anomaly detection is performed on the actually collected data, and the method comprises the following steps:
Processing the actually acquired data to obtain a process quality map;
and inputting the process quality map into the trained convolutional neural network, and judging whether the process quality map is abnormal.
Further, the process of judging whether the abnormality exists includes:
when the process quality map is normal, the monitoring window continues to slide forwards according to a set moving step length, and process quality map generation and convolutional neural network anomaly identification are carried out on data in the window once sliding; when the process quality map is abnormal, the convolutional neural network outputs an abnormal category.
A DCGAN-based linear profile process quality anomaly monitoring system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the DCGAN-based linear profile process quality anomaly monitoring method described above when the computer program is executed by the processor.
The beneficial effects of the invention are as follows: the method comprises the steps of obtaining an initial quality map according to collected linear contour process data under a history normal operation state and an abnormal state, generating an countermeasure network based on deep convolution, carrying out sample amplification to obtain a sample for training a convolution neural network, and finally carrying out process abnormality monitoring according to the convolution neural network after training, so that the problem of limited manual extraction characteristics can be solved. In addition, the depth convolution generation countermeasure network is applied to monitoring of the quality abnormality of the linear contour process, so that minority sample data information generated in the production process can be effectively utilized, and beneficial and more data information is provided for the construction of a follow-up monitoring model; the deep convolution is used for generating two deep learning models of the countermeasure network and the convolution neural network, so that the learning mode of the original data information is simplified, and various abnormal states and normal states generated in the linear contour production process can be efficiently identified and distinguished; according to the historical data of the linear contour production process, the running process in an abnormal state can be effectively identified, a specific abnormal mode is determined, and the monitoring efficiency of the quality abnormality of the linear contour process is improved; the method can train historical data under the condition of unbalanced data sets, ensure the recognition precision of the abnormal process with smaller sample data quantity, and effectively improve the quality abnormality monitoring efficiency of the linear contour process by using the convolutional neural network.
Drawings
In order to more clearly illustrate the technical solution of the embodiments of the present invention, the following briefly describes the drawings that are required to be used in the embodiments:
FIG. 1 is a schematic overall flow diagram of a linear contour process quality anomaly monitoring method based on DCGAN according to an example embodiment of the present application;
fig. 2 is a schematic diagram of a specific implementation of a linear contour process quality anomaly monitoring method based on DCGAN according to an embodiment of the present application.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.
It should be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It should also be understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.
As used in the present description and the appended claims, the term "if" may be interpreted as "when..once" or "in response to a determination" or "in response to detection" depending on the context. Similarly, the phrase "if a determination" or "if a [ described condition or event ] is detected" may be interpreted in the context of meaning "upon determination" or "in response to determination" or "upon detection of a [ described condition or event ]" or "in response to detection of a [ described condition or event ]".
Furthermore, the terms "first," "second," "third," and the like in the description of the present specification and in the appended claims, are used for distinguishing between descriptions and not necessarily for indicating or implying a relative importance.
Reference in the specification to "one embodiment" or "some embodiments" or the like means that a particular feature, structure, or characteristic described in connection with the embodiment is included in one or more embodiments of the application. Thus, appearances of the phrases "in one embodiment," "in some embodiments," "in other embodiments," and the like in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments" unless expressly specified otherwise. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless expressly specified otherwise.
In order to explain the technical scheme of the application, the following description is given by a specific embodiment.
Referring to fig. 1, a flowchart of a linear contour process quality anomaly monitoring method based on DCGAN according to an embodiment of the present application is provided, and for convenience of explanation, only a portion related to the embodiment of the present application is shown.
The linear contour process quality anomaly monitoring method based on DCGAN provided by the embodiment of the application comprises two stages, namely an offline training stage and an online monitoring stage, wherein the steps S1-S5 belong to the offline training stage, and the step S6 belongs to the online monitoring stage.
Step S1: linear contour process data under a historical normal operation state and an abnormal state are respectively collected, and corresponding label information is marked on the linear contour process data:
And respectively acquiring linear contour process data in the historical normal running state and the abnormal state, and marking corresponding label information for the linear contour process data, wherein the label information can be normal information and abnormal information.
Setting M to represent the number of sensors and N to represent the number of measurements collected by each sensor, then M x N measurements collected by M sensors.
The linear profile process data is represented as a data matrix of:
Step S2: gray level mapping is carried out on the linear contour process data, and an initial quality map is obtained:
And (3) carrying out gray level mapping on the linear contour process data obtained in the step (S1) to obtain an initial quality map. As a specific embodiment, the linear contour process data is normalized, that is, elements in the data matrix X are normalized to obtain a normalized matrix, and then gray level mapping is performed on the normalized matrix according to the following formula (2) to obtain an initial quality map.
Wherein j=1, 2, … …, N, k=1, 2, … …, M; round () represents a round function.
Step S3: generating an countermeasure network based on the depth convolution, and performing sample data amplification on the linear contour process data in an abnormal state to obtain an amplification quality map:
Since the following conditions exist in industrial production, abnormal production states rarely occur in the actual production process as compared with normal production states, and process data collected in the abnormal production states is far less than that in the normal states. Classifying and numbering the collected historical data according to the reasons of abnormality existing in the production process, constructing a quality map according to the gray mapping method in the step S2 for a few types of sample data, and taking the quality map as the input of a deep convolution generation countermeasure network, so as to realize the generation of the countermeasure network according to the deep convolution, and carrying out sample data amplification on the linear contour process data in an abnormal state, wherein the obtained quality map is defined as an amplified quality map. Sample amplification based on deep convolution generation against a network is a prior art, and this embodiment provides a specific implementation procedure as follows.
The deep convolution generating countermeasure network (deep convolutional GENERATIVE ADVERSARIAL network, DCGAN) is a generation model, and includes a generator and a discriminator. The purpose of the generator is to generate a dummy sample (dummy sample) that approximately obeys the real data distribution, i.e. to generate a dummy data set from the linear contour process data in the abnormal state, the dummy data set and the real data set being inputs to the arbiter. The task of the discriminator is to correctly distinguish the real data from the false data, namely, the discriminator outputs the probability that the obtained data belongs to the real data set so as to judge whether the sample is a real sample or a false sample. The aim of expanding the sample set is finally achieved by performing countermeasure training inside the model. DCGAN is trained to approximate the distribution of the generator generated data in the network to the original data. Random vectors are generated using the uniform distribution function as input to DCGAN, and the generated quality map is similar to the original quality map. And respectively taking quality maps with smaller data quantity in the categories of the data set as DCGAN input, and enabling the distribution of the data generated by the generator in the network to be similar to the input data through continuous iterative learning.
The training process may be represented by the following formula:
During the training of the model, a random variable z is taken as input to the generator. z are randomly sampled at a probability distribution. And (3) utilizing a complex network structure of the generator to carry out nonlinear mapping on the random variables and obtaining a data set G (z) of the required dimension. The false data set G (z) and the real data set x are used as inputs of a discriminator, and the probability that the obtained data belongs to the real data set is output so as to judge whether the sample is a real sample or a false sample.
The generator and arbiter parameter update process is as follows:
M samples { z 1,z2,z3,...,zm } are sampled from the selected probability distribution. The same amount of data { x 1,x2,x3,...,xm } is sampled from the true sample set.
The arbiter D loss function:
Let the intra-discriminant parameter be θ D, the parameter update direction be the gradient v L D of the corresponding parameter of the increasing formula (4).
Generator G loss function:
Let the generator internal parameter be θ G, and the parameter update direction be the gradient v L G of the parameter corresponding to the reduction formula (5).
The parameters in the discriminator can be updated multiple times or once in the steps, and the parameters of the generator are updated again, and after training is completed, the parameters are randomly sampled from P z (z) and input into the generator, so that a sample similar to P data (x) is generated.
The learning process of DCGAN is completed by comprehensively considering the convergence of the loss values in the generator and the discriminator and the quality of the generated image. And finally, using a random vector generated by random distribution as an input of DCGAN to obtain a generated quality map.
Step S4: obtaining a target quality map according to the initial quality map and the amplification quality map:
The initial quality map is an initial obtained sample, the amplified quality map is an amplified sample, the generated quality map is added into the original data set to obtain a new data set, the amplification of the minority class sample data is completed, namely, the initial quality map and the amplified quality map are integrated together to obtain a quality map used for subsequent training, and the quality map is defined as a target quality map.
Step S5: training a convolutional neural network according to the target quality map:
And training the convolutional neural network according to the target quality map and the corresponding label information. The convolutional neural network (Convolutional neural network, CNN) is a feed-forward neural network, and its basic structure is composed of an Input layer (Input layer), a convolutional layer (Convolutional layer), a pooling layer (Pooling layer), a full-connection layer (Fully connected layer) and an Output layer (Output layer). The updating process of the model parameters in the convolutional neural network is as follows:
1) Convolution layer:
The convolution layer (Convolutional layer) is formed of a plurality of Feature planes (Feature maps), each pixel value in the Feature plane representing a neuron, and each neuron being connected to a square region of mxm in the Feature plane of the previous layer. And carrying out two-dimensional convolution operation on the local square region of the previous layer and the convolution kernel, and transmitting the numerical value obtained by the weighted sum to an activation function to obtain an output feature matrix. The calculation formula is as follows:
Wherein l is the current layer number; w a,b is the weight in the convolution kernel; b is offset; f is an activation function, and a ReLU (RECTIFIED LINEAR Units) activation function is often selected in CNN.
2) Pooling layer:
The pooling layer (Pooling layer), i.e. the downsampling layer (Down-SAMPLING LAYER), divides the input feature planes into non-overlapping 2×2 rectangular regions, performs pooling operation on each region, and each feature plane output by the pooling layer corresponds to the feature plane output by the convolution layer of the previous layer without changing the number thereof. The feature surface is further reduced in dimension through the pooling layer, parameters of a network can be reduced, and robustness of the model is improved. Common pooling methods are mean pooling (Average pooling), maximum pooling (Max-pooling) and random pooling (Stochastic pooling). After pooling, the size of the characteristic surface is reduced to 1/2 of that before pooling, and training parameters of a network are not increased, so that the phenomenon of overfitting can be effectively avoided.
3) Full tie layer:
After repeated convolution and downsampling, the full-connection layer sequentially expands and arranges the pixel values of all the feature images into a row to form feature vectors. Each neuron in the fully connected layer is fully connected with the neurons of the subsequent layer, and features extracted by the convolution layer and the downsampling layer can be integrated to extract features with more differentiation. The last layer of the full-connection layer is an output layer, and the output layer is usually classified by adopting softmax logistic regression, so that probability values belonging to various categories are output.
Formulas (7) and (8) are the middle layer and softmax output layer of the fully-connected layer, respectively. In the formula (7), x * is a feature vector output by the last pooling layer, k ij is a connection weight of the j-th input and the i-th intermediate layer neuron, b * is bias, and f is an activation function. In equation (8), P j is a probability value that the input data belongs to the j-th class, and W j is a weight vector that the j-th output layer node is connected to the previous layer.
4) Error back propagation:
Convolutional neural networks employ a back-propagation algorithm (Back propagation algorithm) similar to the conventional neural network, then the loss function of the p-th training sample can be defined as:
Where y pj is the actual probability that the p-th training sample belongs to the j-th class, and S pj is the predicted probability that the p-th training sample belongs to the j-th class. Equation (9) consists of cross entropy and regularization term. The back propagation algorithm is to train the model by optimizing the weights and parameters to minimize the loss function. The key to the BP algorithm is to find the partial derivatives of the loss function for weight and bias. For a single training sample, the bias derivative calculation process of the weight is as follows:
wherein, Can be calculated by the chain rule:
similarly, errors are propagated to the upper layer in the same manner:
the bias guide calculation process of the bias is the same as the weight, and the weight and the bias adjustment direction are respectively as follows:
where η is the learning rate.
And taking the data set corresponding to the target quality map as input of the convolutional neural network, and completing the learning process of the model on the data set according to the recognition accuracy of the convolutional neural network model on the data set and the convergence of the loss function value. And comparing the convolutional neural network model identification effects under different hyper-parameter combinations by using a grid search method, and further determining a last used model structure. The adjustment process is as follows: different levels are set and combined for the structure (the number of layers and the number of hidden points) of the convolutional neural network model and the parameters (the learning rate, the optimizer, the activation function and the like) of the convolutional neural network model; training convolutional neural network models under different combinations respectively, and obtaining corresponding recognition accuracy; and selecting the convolutional neural network model with highest recognition accuracy as a final model, and storing.
Step S6: based on the convolutional neural network after training, carrying out process anomaly detection on the actually acquired data:
In this embodiment, data actually collected by the sensor is obtained by a sliding window value mode. The actually collected data are processed to obtain a process quality map, and the conversion mode of the quality map is specifically described above and will not be described again.
Inputting the process quality map into the convolutional neural network after training, and carrying out online identification on the process quality map so as to judge whether the process is abnormal. The convolutional neural network judges the state type of the process quality map and outputs the mode type with highest probability, and the mode type corresponds to the normal mode and the abnormal mode in the training data set, so that the real-time running state of the production process is obtained. When the process is normal, the monitoring window continues to slide forwards according to the set moving step length, and quality map generation and convolutional neural network identification are carried out on the data in the window once sliding; when the process is abnormal, the convolutional neural network model outputs an abnormal category, so that the monitoring of the linear contour process abnormality is realized. Fig. 2 is a schematic diagram of a specific implementation of a linear contour process quality anomaly monitoring method based on DCGAN according to an embodiment of the present application.
The embodiment also provides a linear contour process quality anomaly monitoring system based on DCGAN, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes the steps of the linear contour process quality anomaly monitoring method based on DCGAN when executing the computer program.
The above embodiments are only for illustrating the technical solution of the present application, and not for limiting the same; although the application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present application, and are intended to be included in the scope of the present application.
Claims (5)
1. A method for monitoring quality anomalies in a linear contour process based on DCGAN, comprising:
Acquiring linear contour process data in a historical normal running state and an abnormal state respectively, and marking corresponding label information for the linear contour process data;
carrying out gray level mapping on the linear contour process data to obtain an initial quality map;
generating an countermeasure network based on the depth convolution, and performing sample data amplification on the linear contour process data in an abnormal state to obtain an amplification quality map;
obtaining a target quality map according to the initial quality map and the amplification quality map;
Training a convolutional neural network according to the target quality map;
performing process anomaly detection on the actually acquired data based on the convolutional neural network after training;
The calculation process of the initial quality map comprises the following steps:
the linear profile process data is represented as a data matrix of:
The calculation formula of the initial mass spectrum is as follows:
wherein M represents the number of sensors, N represents the number of measurements acquired by each sensor, j=1, 2, … …, N, k=1, 2, … …, M; round () represents a round function;
the convolutional neural network based on training completion carries out process anomaly detection on the actually collected data, and the convolutional neural network based on training completion comprises the following steps:
Processing the actually acquired data to obtain a process quality map;
inputting the process quality map to the trained convolutional neural network, and judging whether the process quality map is abnormal;
the process for judging whether the abnormality exists comprises the following steps:
when the process quality map is normal, the monitoring window continues to slide forwards according to a set moving step length, and process quality map generation and convolutional neural network anomaly identification are carried out on data in the window once sliding; when the process quality map is abnormal, the convolutional neural network outputs an abnormal category.
2. The method for monitoring quality anomalies of a linear contour process based on DCGAN as set forth in claim 1, wherein said performing gray level mapping on the linear contour process data to obtain an initial quality map includes:
normalizing the linear contour process data to obtain a normalized matrix;
And carrying out gray level mapping on the normalized matrix to obtain an initial quality map.
3. The DCGAN-based linear contour process quality anomaly monitoring method of claim 1, wherein said deep convolution generation countermeasure network comprises a generator and a discriminator, said generator is configured to generate a false data set from the linear contour process data in the anomaly state, said false data set and the true data set are used as inputs of the discriminator, said discriminator outputs a probability that the obtained data belongs to the true data set, and further determines whether the sample is a true sample or a false sample.
4. The method for monitoring quality anomalies of a linear contour process based on DCGAN as claimed in claim 1, wherein said actual collected data is obtained by means of a sliding window.
5. A DCGAN-based linear profile process quality anomaly monitoring system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the DCGAN-based linear profile process quality anomaly monitoring method of any one of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210062233.9A CN114417248B (en) | 2022-01-19 | 2022-01-19 | DCGAN-based linear contour process quality anomaly monitoring method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210062233.9A CN114417248B (en) | 2022-01-19 | 2022-01-19 | DCGAN-based linear contour process quality anomaly monitoring method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114417248A CN114417248A (en) | 2022-04-29 |
| CN114417248B true CN114417248B (en) | 2024-09-10 |
Family
ID=81274557
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210062233.9A Active CN114417248B (en) | 2022-01-19 | 2022-01-19 | DCGAN-based linear contour process quality anomaly monitoring method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114417248B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109998500A (en) * | 2019-04-30 | 2019-07-12 | 陕西师范大学 | A kind of pulse signal generation method and system based on production confrontation network |
| CN113298230A (en) * | 2021-05-14 | 2021-08-24 | 西安理工大学 | Prediction method based on unbalanced data set for generating countermeasure network |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019232435A1 (en) * | 2018-06-01 | 2019-12-05 | Grail, Inc. | Convolutional neural network systems and methods for data classification |
| CN113743592A (en) * | 2021-09-16 | 2021-12-03 | 北京航空航天大学 | Telemetry data anomaly detection method based on GAN |
-
2022
- 2022-01-19 CN CN202210062233.9A patent/CN114417248B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109998500A (en) * | 2019-04-30 | 2019-07-12 | 陕西师范大学 | A kind of pulse signal generation method and system based on production confrontation network |
| CN113298230A (en) * | 2021-05-14 | 2021-08-24 | 西安理工大学 | Prediction method based on unbalanced data set for generating countermeasure network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114417248A (en) | 2022-04-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhao et al. | Deep multi-scale convolutional transfer learning network: A novel method for intelligent fault diagnosis of rolling bearings under variable working conditions and domains | |
| CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
| Ayodeji et al. | Causal augmented ConvNet: A temporal memory dilated convolution model for long-sequence time series prediction | |
| CN111967486A (en) | Complex equipment fault diagnosis method based on multi-sensor fusion | |
| CN108596327B (en) | Seismic velocity spectrum artificial intelligence picking method based on deep learning | |
| Thoppil et al. | Deep learning algorithms for machinery health prognostics using time-series data: A review | |
| CN109992921A (en) | A kind of online soft sensor method and system of the coal-fired plant boiler thermal efficiency | |
| CN111768000A (en) | Industrial process data modeling method for online adaptive fine-tuning deep learning | |
| CN116448419A (en) | Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization | |
| CN112731890B (en) | Method and device for detecting power plant equipment faults | |
| CN114548591A (en) | Time sequence data prediction method and system based on hybrid deep learning model and Stacking | |
| CN112949836A (en) | Method for carrying out regression prediction on-line migration learning on time-varying distribution data | |
| CN116415505A (en) | System fault diagnosis and state prediction method based on SBR-DBN model | |
| CN112949009A (en) | Active power distribution network distributed state estimation method and system based on transfer learning | |
| CN117010263A (en) | Residual life prediction method based on convolutional neural network and long-term and short-term memory network | |
| CN114417248B (en) | DCGAN-based linear contour process quality anomaly monitoring method and system | |
| CN115062764B (en) | Intelligent illuminance adjustment and environmental parameter Internet of things big data system | |
| CN116702839A (en) | Model training method and application system based on convolutional neural network | |
| CN116522065A (en) | Coal mill health degree assessment method based on deep learning | |
| CN113779724B (en) | Intelligent fault prediction method and system for filling packaging machine | |
| CN114254828B (en) | Power load prediction method based on mixed convolution feature extractor and GRU | |
| CN116306292A (en) | Hydropower station physical field digital twin model construction method based on convolutional neural network | |
| CN115600498A (en) | Wind speed forecast correction method based on artificial neural network | |
| CN114298130A (en) | Short-term wind power prediction method and device | |
| CN114692729A (en) | New energy station bad data identification and correction method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |