CN111665819A - Deep learning multi-model fusion-based complex chemical process fault diagnosis method - Google Patents
Deep learning multi-model fusion-based complex chemical process fault diagnosis method Download PDFInfo
- Publication number
- CN111665819A CN111665819A CN202010511370.7A CN202010511370A CN111665819A CN 111665819 A CN111665819 A CN 111665819A CN 202010511370 A CN202010511370 A CN 202010511370A CN 111665819 A CN111665819 A CN 111665819A
- Authority
- CN
- China
- Prior art keywords
- layer
- data
- chemical process
- diagnosis
- deep learning
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0262—Confirmation of fault detection, e.g. extra checks to confirm that a failure has indeed occurred
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/20—Pc systems
- G05B2219/24—Pc safety
- G05B2219/24065—Real time diagnostics
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a fault diagnosis method for a complex chemical process based on deep learning multi-model fusion. The two neural networks automatically extract fault characteristics from two aspects respectively, then fuse the characteristics and input the characteristics into a multilayer perceptron (MLP) for further characteristic compression and extraction, and finally output a diagnosis result. The invention extracts features through a Convolutional Neural Network (CNN) and a long-short term memory network (LSTM) respectively, so that the finally extracted features of the network have space and time characteristics at the same time, and final diagnosis is carried out by integrating the characteristics of two aspects, thereby not only overcoming the problem of large calculated amount of the traditional diagnosis technology, but also overcoming the technical problem that fault diagnosis can not be accurately carried out in a complex chemical process due to too many facets of the features extracted by a single network.
Description
Technical Field
The invention relates to a chemical process fault diagnosis method, in particular to a complex chemical process fault diagnosis method based on deep learning multi-model fusion.
Background
Chemical industry is the basic industry of national development, along with the development of modern industrial technology and process control technology, the chemical technology is developed towards the direction of complexity, the process signal data has the characteristics of high dimension, time variation, non-Gaussian distribution, nonlinearity, strong coupling and the like, the data volume is extremely large, the fault characteristics are difficult to select, the fault diagnosis accuracy of the chemical process is low, once a fault occurs, a series of problems can be caused, inestimable loss can be caused, and therefore the accurate and efficient fault diagnosis has great significance for the safety production of the chemical process.
The conventional fault identification method has large calculation amount and needs certain industrial knowledge, and the fault diagnosis requirement of the conventional complex chemical process is difficult to meet no matter the diagnosis precision or the diagnosis time. With the rapid development of machine learning technology, deep learning technology comes along with the emergence of the machine learning technology, the deep learning technology is widely applied to a plurality of fields such as images, texts and voices, the deep learning technology obtains new research progress in fault diagnosis by virtue of strong learning capacity of the deep learning technology, and fault diagnosis researchers at home and abroad build deep learning network models aiming at diagnostic objects for fault diagnosis, so that the experimental effect is good, and the performance is excellent.
Disclosure of Invention
Aiming at the technical defects in the prior art, the invention provides a complex chemical process fault diagnosis method based on deep learning multi-model fusion.
The invention comprises the following steps:
step one, preprocessing an experimental data set
1-1, randomly disorganizing the data set for the experiment after labeling, and removing data points which are not related to the fault through dimensionality reduction.
And 1-2, carrying out normalization processing on the data, namely scaling the data in proportion and scaling the value ranges of all the characteristic data points to be between 0 and 1.
1-3, dividing the normalized data into a training set and a testing set according to a certain proportion.
Step two, constructing a neural network model with multi-model fusion
2-1, after the training set is subjected to standardization processing by a batch standardization layer, constructing a parallel network of CNN and LSTM, and respectively carrying out automatic feature extraction on the sample data of the training set.
2-2, after the features extracted by the parallel network pass through a random discarding layer, fusing and inputting MLP to perform further feature compression and extraction.
2-3, unfolding the fused features into x1,x2,x3.., as vector X [1 ]]The weight from MLP input layer to next layer is w1,w2,w3.., as vector W [1 ]]The output of the first layer is then A1]＝σ(W[1]X[1]+b[1]) Where σ is the activation function, b [1 ]]Biased for the first layer, output A1]Is also the input value of the next layer, namely X2]＝A[1]And the class pushes to the next layer in turn.
Step three, obtaining a diagnosis result and training a neural network model through back propagation
And the last full-connection layer is output by a softmax classifier to obtain a diagnosis result, the cross entropy between the diagnosis result output by the neural network model and the real label is used as a loss function, the parameters of the whole neural network model are updated through back propagation of an optimization function, convergence optimization is carried out on the loss function, the diagnosis precision is improved, the final effect of the neural network model is displayed through a test set, and the hyper-parameters are correspondingly adjusted according to the diagnosis result of the test set.
The invention has the beneficial effects that: the invention extracts features through a Convolutional Neural Network (CNN) and a long-short term memory network (LSTM) respectively, so that the finally extracted features of the network have space and time characteristics at the same time, and final diagnosis is carried out by integrating the characteristics of two aspects, thereby not only overcoming the problem of large calculated amount of the traditional diagnosis technology, but also overcoming the technical problem that fault diagnosis can not be accurately carried out in a complex chemical process due to too many facets of the features extracted by a single network.
Drawings
FIG. 1 is a constructed multi-model fusion neural network model.
Detailed Description
The invention adopts the following technical scheme:
step one, preprocessing an experimental data set
1-1, labeling the data set for experiment, randomly disorganizing, and removing data points which are not related to faults through a dimensionality reduction technology.
1-2, carrying out standardization processing on the obtained data, namely scaling the data in proportion, scaling the value ranges of all characteristic data points to be between 0 and 1, facilitating later-stage network training and learning characteristic information, and processing formulas thereof:
where x represents the raw data for a characteristic data point in the data set, x*Represents the data point obtained after processing, min represents the minimum value of the data point in the data set, and max represents the maximum value of the data point in the data set.
1-3, dividing the normalized data into a training set and a testing set according to a certain proportion.
Step two, constructing a neural network model with multi-model fusion
2-1, after the training set is subjected to standardization processing by a batch standardization layer, constructing a parallel network of CNN and LSTM, and respectively carrying out automatic feature extraction.
The CNN part in the invention is formed by alternately stacking a plurality of convolutional layers and pooling layers, each characteristic graph in the convolutional layers corresponds to a convolution kernel, the convolution kernels deconvolve input information through weight and form a group of characteristic outputs which are used as input data information of the next layer, and a formula is calculated:
A[L]＝σ(W[L]*A[L-1]+b[L])
wherein L represents the L-th layer of convolution, "+" represents the convolution operation, W is the convolution kernel, a is the input, b is the offset, σ represents the selected activation function.
A pooling layer is added after the convolutional layer for downsampling, and maximum pooling is used to calculate the most important part (maximum) of the local unit in the input feature map.
The LSTM units of the invention comprise memory cells CtAnd three thresholds: input door itForgetting door ftAnd an output gate otThe specific calculation formula is as follows:
ft＝σ(Whfht-1+Wxfxt+bf)
it＝σ(Whiht-1+Wxixt+bi)
ot＝σ(Whoht-1+Wxoxt+bo)
wherein the input sequence is xtWhen σ is Sigmoid activation function, Whf、Whi、Who、WhCRespectively representing the weight matrixes between the forgetting gate, the input gate, the output gate and the memory cell and the previous hidden layer, ht-1Is the output of the cell at the previous time, Wxf、Wxi、Wxo、WxCRespectively representing the weight matrix between the input and forgetting gates, the input gate, the output gate, the memory cell, bf、bi、bo、bCRespectively, the offsets of the forgetting gate, the input gate, the output gate, and the memory cell.
In order to avoid overfitting and increase the model generalization capability, the invention adds L2 regularization and random discarding mechanism (Dropout) in each layer of LSTM, and adds L2 regularization in each layer of CNN.
2-2, after the features extracted by the CNN and the LSTM are randomly discarded, the features are fused and input into an MLP (multi-layer linear programming) for further feature compression and extraction, wherein the MLP is formed by stacking a plurality of fully connected layers, and a Dropout layer is added between every two fully connected layers, so that the over-fitting problem is avoided.
Unfolding the fused features into x1,x2,x3.., as vector X [1 ]]The weight from MLP input layer to next layer is w1,w2,w3.., as vector W [1 ]]Where 1 denotes the weight of the first layer of the MLP, offset b [1 ]]The same process is carried out; the calculation of the first layer is A1]＝σ(W[1]X[1]+b[1]) Where σ is the activation function and the output is A [1 ]]I.e. the input value of the next layer, i.e. X[2]＝A[1]And the class pushes to the next layer in turn.
Step three, obtaining a diagnosis result and training a network model through back propagation
And the last full-connection layer is output by a softmax classifier to obtain a diagnosis result, the cross entropy between the diagnosis result output by the neural network model and the real label is used as a loss function, the parameters of the whole neural network model are updated through back propagation of an optimization function, convergence optimization is carried out on the loss function, the diagnosis precision is improved, the final effect of the neural network model is displayed through a test set, and the hyper-parameters are correspondingly adjusted according to the diagnosis result of the test set.
The method has obvious time sequence for the complex chemical process data set, the whole complex chemical process is a continuous process, and a fault often has a series of influences on a subsequent system, so that the method selects the LSTM to perform feature extraction on the time domain of the data set, and the LSTM learns the long-period correlation of the time sequence data by using a nonlinear gating function and is particularly effective. The CNN has a remarkable effect on the extraction of spatial features, in process data, local value groups are often highly correlated, and the same fault type can form the same local mode, so that the fault identification is easier; unique local patterns can appear anywhere in the input feature map, and different position units sharing the same weight in the convolutional network can help to detect the same pattern, so that the CNN is selected to perform spatial feature extraction of chemical process data.
Most of the multi-model fusion methods used today are to stack different networks in sequence to form a serial structure, and as a result, although the finally obtained features have multi-aspect features to some extent, it has a problem of feature loss to some extent. The invention constructs a parallel network structure to extract the characteristics, uses CNN and LSTM to extract the characteristics of the original data, then simply splices and fuses the characteristics extracted by the two networks, and inputs the spliced and fused characteristics into a multilayer sensor to further perform characteristic fusion and characteristic extraction, thus the finally obtained characteristics have the characteristics of two aspects of time domain and space domain, simultaneously the problem of characteristic loss is avoided, the two aspects of characteristics can be well synthesized to perform fault diagnosis, the network performance is improved, and more accurate and efficient diagnosis effect is achieved.
Example (b):
the TE process is a complex chemical simulation process model proposed by Eastman chemical company, usa, which contains 41 measured variables and 12 controlled variables, but the 12 th controlled variable is not constant in stirring speed, all process measurements contain gaussian noise, 21 faults are preset, faults 1 to 7 are related to step changes in the process variables, faults 8 to 12 are related to variability of some process variables, fault 13 is a slow drift in the reflection dynamics, and faults 14, 15, 21 are related to valve sticking.
The application of the method of the present invention to the TE process simulation object operates as follows.
Step one, preprocessing an experimental data set
1-1, each fault type and normal state comprises 1280 samples, the experimental data set is labeled and then randomly disturbed, the relevance between the sample characteristic points and the fault types is observed through drawing, and the six characteristic points which are not relevant to the fault and are 11 th, 13 th, 14 th, 36 th, 47 th and 48 th are removed from 52 variables.
1-2, normalizing the sample, and scaling the value ranges of all the feature points to be between 0 and 1, so as to facilitate later-stage network training and feature information learning.
1-3, dividing the normalized data into a training set and a testing set according to 80% and 20%, and inputting the training set into a network for feature extraction.
Step two, constructing a neural network model with multi-model fusion
2-1, after the training set is subjected to standardization processing by a batch standardization layer, constructing a parallel network of CNN and LSTM, and respectively carrying out automatic feature extraction. The CNN is formed by stacking six one-dimensional convolution layers and pooling layers alternately, the convolution kernels are all 3 in size, and elu is selected as the activation function. The LSTM is formed by stacking three long-short term memory layers. To avoid overfitting and increase the model generalization capability, L2 regularization and Dropout were added in each layer of the LSTM and L2 regularization was added in each layer of the CNN.
2-2, after the features extracted by CNN and LSTM are discarded randomly, the features are fused and input into MLP for further feature compression and extraction, wherein the MLP is composed of three full-connection layers, the first two layers of output nodes are 256 and 64 respectively, the last layer is an output layer, elu is selected as an activation function, adam is an optimization function, a Dropout layer is added between every two full-connection layers, L2 regularization is added to each full-connection layer to reduce overfitting and improve the generalization capability of the model, and the specific structure is shown in FIG. 1.
Step three, obtaining a diagnosis result and training a network model through back propagation
And the last full-connection layer is output by a softmax classifier to obtain a diagnosis result, the cross entropy between the diagnosis result output by the model and the real label is used as a loss function, the parameters of the whole model are updated through back propagation of an Adam optimizer, and convergence optimization is carried out on the loss function. Each training batch contains 128 samples, 150 rounds of training are carried out, in order to achieve faster and better convergence, the learning rate of 45 rounds of training is reduced to one tenth of the original learning rate, and the accuracy rate of final fault diagnosis is superior to that of a traditional algorithm and a deep learning model of a single network.
In order to verify the effectiveness of the method, a network model with sequentially stacked common CNN-LSTM is set, original data is firstly input into three layers of CNN networks for feature extraction, then the obtained features are input into two layers of LSTM for subsequent feature extraction, and finally three layers of fully-connected networks are input, wherein 22 system state diagnosis accuracy rate experimental results are shown in Table 1:
TABLE 1 two sets of experimental results
It is obvious from the results that the invention is advantageous, and the micro-average accuracy is taken as an example, which is improved by 2.3% compared with the common CNN-LSTM, and in order to further observe the experimental results in detail, Table 2 shows the diagnostic rate of each type of state of two groups of experiments.
TABLE 2 diagnostic results for each condition
As can be seen from Table 2, the diagnostic rate of most system states is improved to a different extent compared with that of the ordinary CNN-LSTM, which further confirms the effectiveness of the present invention.
Claims (4)
1. A fault diagnosis method for a complex chemical process based on deep learning multi-model fusion is characterized by comprising the following steps:
step one, preprocessing an experimental data set
1-1, labeling a data set for an experiment, randomly disordering, and removing data points irrelevant to faults through dimensionality reduction;
1-2, carrying out standardization processing on the data, namely scaling the data in proportion and scaling the value ranges of all characteristic data points to be between 0 and 1;
1-3, dividing the normalized data into a training set and a test set according to a certain proportion;
step two, constructing a neural network model with multi-model fusion
2-1, after the training set is subjected to standardization processing by a batch standardization layer, constructing a parallel network of CNN and LSTM, and respectively carrying out automatic feature extraction on the sample data of the training set;
2-2, after the features extracted by the parallel network pass through a random discarding layer, fusing and inputting an MLP (Multi-layer processor) to perform further feature compression and extraction;
2-3, unfolding the fused features into x1,x2,x3.., as vector X [1 ]]The weight from MLP input layer to next layer is w1,w2,w3.., as vector W [1 ]]The output of the first layer is then A1]＝σ(W[1]X[1]+b[1]) Where σ is the activation function, b [1 ]]Biased for the first layer, output A1]Is also the input value of the next layer, namely X2]＝A[1]Pushing to the next layer in sequence;
step three, obtaining a diagnosis result and training a network model through back propagation
And the last full-connection layer is output by a softmax classifier to obtain a diagnosis result, the cross entropy between the diagnosis result output by the neural network model and the real label is used as a loss function, the parameters of the whole neural network model are updated through back propagation of an optimization function, convergence optimization is carried out on the loss function, the diagnosis precision is improved, the final effect of the neural network model is displayed through a test set, and the hyper-parameters are correspondingly adjusted according to the diagnosis result of the test set.
2. The method for diagnosing the faults of the complex chemical process based on the deep learning multi-model fusion as claimed in claim 1, wherein:
the CNN is formed by alternately stacking a plurality of convolutional layers and pooling layers, each feature graph in each convolutional layer corresponds to a convolution kernel, the convolution kernels perform convolution operation on input data through corresponding weights, and the obtained output results form a group of feature outputs to be used as input data information of the next layer;
and adding a pooling layer after the convolutional layer for down-sampling, wherein the maximal pooling is used for calculating the most important part of the local unit in the input feature map.
3. The method for diagnosing the faults of the complex chemical process based on the deep learning multi-model fusion as claimed in claim 1, wherein: an L2 regularization and random dropping mechanism is added in each layer of the LSTM, and an L2 regularization is added in each layer of the CNN.
4. The method for diagnosing the faults of the complex chemical process based on the deep learning multi-model fusion as claimed in claim 1, wherein: the MLP is formed by stacking a plurality of fully connected layers, and a Dropout layer is added between every two fully connected layers.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010511370.7A CN111665819A (en) | 2020-06-08 | 2020-06-08 | Deep learning multi-model fusion-based complex chemical process fault diagnosis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010511370.7A CN111665819A (en) | 2020-06-08 | 2020-06-08 | Deep learning multi-model fusion-based complex chemical process fault diagnosis method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111665819A true CN111665819A (en) | 2020-09-15 |
Family
ID=72385751
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010511370.7A Pending CN111665819A (en) | 2020-06-08 | 2020-06-08 | Deep learning multi-model fusion-based complex chemical process fault diagnosis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111665819A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112486148A (en) * | 2020-12-29 | 2021-03-12 | 浙江大学 | Automobile fault diagnosis method and system based on convolutional memory self-coding network |
| CN112561035A (en) * | 2020-12-08 | 2021-03-26 | 上海海事大学 | Fault diagnosis method based on CNN and LSTM depth feature fusion |
| CN112598186A (en) * | 2020-12-28 | 2021-04-02 | 重庆邮电大学 | Improved LSTM-MLP-based small generator fault prediction method |
| CN112668459A (en) * | 2020-12-25 | 2021-04-16 | 合肥工业大学 | Rolling bearing fault diagnosis method based on supervised contrast learning and convolutional neural network |
| CN112988714A (en) * | 2021-04-06 | 2021-06-18 | 中车青岛四方机车车辆股份有限公司 | Bearing fault database construction method, analysis method and system |
| CN113158722A (en) * | 2020-12-24 | 2021-07-23 | 哈尔滨理工大学 | Rotary machine fault diagnosis method based on multi-scale deep neural network |
| CN113984946A (en) * | 2021-10-21 | 2022-01-28 | 江南大学 | Crayfish freshness detection method based on gas phase electronic nose and machine learning |
| CN114326638A (en) * | 2021-12-30 | 2022-04-12 | 杭州电子科技大学 | Multi-scale convolution and depth bidirectional GRU network fusion performance evaluation method |
| CN115238671A (en) * | 2022-09-22 | 2022-10-25 | 四川大学 | Anti-interference detection method based on grammatical features and integration strategy |
| CN115708028A (en) * | 2021-08-19 | 2023-02-21 | 中国石油化工股份有限公司 | Chemical process fault diagnosis method and system, electronic device and storage medium |
| CN117349712A (en) * | 2023-12-04 | 2024-01-05 | 贵州大学 | Unmanned aerial vehicle fault diagnosis method based on space-time characteristic self-adaptive shared network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109213119A (en) * | 2018-07-11 | 2019-01-15 | 佛山科学技术学院 | Complex industrial critical component failure prediction method and system based on on-line study |
| CN110398369A (en) * | 2019-08-15 | 2019-11-01 | 贵州大学 | A kind of Fault Diagnosis of Roller Bearings merged based on 1-DCNN and LSTM |
| CN110674280A (en) * | 2019-06-21 | 2020-01-10 | 四川大学 | Answer selection algorithm based on enhanced question importance expression |
| CN110717330A (en) * | 2019-09-23 | 2020-01-21 | 哈尔滨工程大学 | Word-sentence level short text classification method based on deep learning |
| US20200041984A1 (en) * | 2018-07-31 | 2020-02-06 | Canon Kabushiki Kaisha | Production system, and method of producing article |
| CN111079748A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting oil throwing fault of rolling bearing of railway wagon |
| CN111123894A (en) * | 2019-12-30 | 2020-05-08 | 杭州电子科技大学 | Chemical process fault diagnosis method based on combination of LSTM and MLP |
-
2020
- 2020-06-08 CN CN202010511370.7A patent/CN111665819A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109213119A (en) * | 2018-07-11 | 2019-01-15 | 佛山科学技术学院 | Complex industrial critical component failure prediction method and system based on on-line study |
| US20200041984A1 (en) * | 2018-07-31 | 2020-02-06 | Canon Kabushiki Kaisha | Production system, and method of producing article |
| CN110674280A (en) * | 2019-06-21 | 2020-01-10 | 四川大学 | Answer selection algorithm based on enhanced question importance expression |
| CN110398369A (en) * | 2019-08-15 | 2019-11-01 | 贵州大学 | A kind of Fault Diagnosis of Roller Bearings merged based on 1-DCNN and LSTM |
| CN110717330A (en) * | 2019-09-23 | 2020-01-21 | 哈尔滨工程大学 | Word-sentence level short text classification method based on deep learning |
| CN111079748A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting oil throwing fault of rolling bearing of railway wagon |
| CN111123894A (en) * | 2019-12-30 | 2020-05-08 | 杭州电子科技大学 | Chemical process fault diagnosis method based on combination of LSTM and MLP |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112561035A (en) * | 2020-12-08 | 2021-03-26 | 上海海事大学 | Fault diagnosis method based on CNN and LSTM depth feature fusion |
| CN112561035B (en) * | 2020-12-08 | 2023-07-25 | 上海海事大学 | Fault diagnosis method based on CNN and LSTM depth feature fusion |
| CN113158722A (en) * | 2020-12-24 | 2021-07-23 | 哈尔滨理工大学 | Rotary machine fault diagnosis method based on multi-scale deep neural network |
| CN112668459B (en) * | 2020-12-25 | 2022-09-13 | 合肥工业大学 | Rolling bearing fault diagnosis method based on supervised contrast learning and convolutional neural network |
| CN112668459A (en) * | 2020-12-25 | 2021-04-16 | 合肥工业大学 | Rolling bearing fault diagnosis method based on supervised contrast learning and convolutional neural network |
| CN112598186A (en) * | 2020-12-28 | 2021-04-02 | 重庆邮电大学 | Improved LSTM-MLP-based small generator fault prediction method |
| CN112598186B (en) * | 2020-12-28 | 2022-06-24 | 重庆邮电大学 | Improved LSTM-MLP-based small generator fault prediction method |
| CN112486148B (en) * | 2020-12-29 | 2022-02-18 | 浙江大学 | Automobile fault diagnosis method and system based on convolutional memory self-coding network |
| CN112486148A (en) * | 2020-12-29 | 2021-03-12 | 浙江大学 | Automobile fault diagnosis method and system based on convolutional memory self-coding network |
| CN112988714A (en) * | 2021-04-06 | 2021-06-18 | 中车青岛四方机车车辆股份有限公司 | Bearing fault database construction method, analysis method and system |
| CN115708028A (en) * | 2021-08-19 | 2023-02-21 | 中国石油化工股份有限公司 | Chemical process fault diagnosis method and system, electronic device and storage medium |
| CN113984946A (en) * | 2021-10-21 | 2022-01-28 | 江南大学 | Crayfish freshness detection method based on gas phase electronic nose and machine learning |
| CN114326638A (en) * | 2021-12-30 | 2022-04-12 | 杭州电子科技大学 | Multi-scale convolution and depth bidirectional GRU network fusion performance evaluation method |
| CN115238671A (en) * | 2022-09-22 | 2022-10-25 | 四川大学 | Anti-interference detection method based on grammatical features and integration strategy |
| CN117349712A (en) * | 2023-12-04 | 2024-01-05 | 贵州大学 | Unmanned aerial vehicle fault diagnosis method based on space-time characteristic self-adaptive shared network |
| CN117349712B (en) * | 2023-12-04 | 2024-03-12 | 贵州大学 | Unmanned aerial vehicle fault diagnosis method based on space-time characteristic self-adaptive shared network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111665819A (en) | Deep learning multi-model fusion-based complex chemical process fault diagnosis method | |
| 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 | |
| Zhao et al. | A visual long-short-term memory based integrated CNN model for fabric defect image classification | |
| CN109116834B (en) | Intermittent process fault detection method based on deep learning | |
| CN109242223B (en) | Quantum support vector machine evaluation and prediction method for urban public building fire risk | |
| CN112926641B (en) | Three-stage feature fusion rotating machine fault diagnosis method based on multi-mode data | |
| CN108445752B (en) | Random weight neural network integrated modeling method for self-adaptively selecting depth features | |
| CN111768000A (en) | Industrial process data modeling method for online adaptive fine-tuning deep learning | |
| CN108879732B (en) | Transient stability evaluation method and device for power system | |
| CN115018021A (en) | Machine room abnormity detection method and device based on graph structure and abnormity attention mechanism | |
| CN111123894B (en) | Chemical process fault diagnosis method based on combination of LSTM and MLP | |
| CN110472695B (en) | Abnormal working condition detection and classification method in industrial production process | |
| CN113486578A (en) | Method for predicting residual life of equipment in industrial process | |
| Peng et al. | Towards robust and understandable fault detection and diagnosis using denoising sparse autoencoder and smooth integrated gradients | |
| de Campos Souza et al. | Uninorm based regularized fuzzy neural networks | |
| CN113110398B (en) | Industrial process fault diagnosis method based on dynamic time consolidation and graph convolution network | |
| Zhang et al. | Sparsity and manifold regularized convolutional auto-encoders-based feature learning for fault detection of multivariate processes | |
| Wang et al. | Application research on long short-term memory network in fault diagnosis | |
| Alom et al. | Object recognition using cellular simultaneous recurrent networks and convolutional neural network | |
| CN114548591A (en) | Time sequence data prediction method and system based on hybrid deep learning model and Stacking | |
| Liu et al. | Deep ensemble forests for industrial fault classification | |
| Xu et al. | Global attention mechanism based deep learning for remaining useful life prediction of aero-engine | |
| Zha et al. | Recognizing plans by learning embeddings from observed action distributions | |
| CN114943016A (en) | Cross-granularity joint training-based graph comparison representation learning method and system | |
| Yang et al. | iCausalOSR: invertible Causal Disentanglement for Open-set Recognition |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200915 |
|
| RJ01 | Rejection of invention patent application after publication |