CN116821811A - Coal mill unit fault diagnosis method and system based on multi-layer graph convolution neural network - Google Patents
Coal mill unit fault diagnosis method and system based on multi-layer graph convolution neural network Download PDFInfo
- Publication number
- CN116821811A CN116821811A CN202310606439.8A CN202310606439A CN116821811A CN 116821811 A CN116821811 A CN 116821811A CN 202310606439 A CN202310606439 A CN 202310606439A CN 116821811 A CN116821811 A CN 116821811A
- Authority
- CN
- China
- Prior art keywords
- fault
- layer
- samples
- neural network
- nodes
- 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
- 239000003245 coal Substances 0.000 title claims abstract description 104
- 238000003745 diagnosis Methods 0.000 title claims abstract description 62
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 22
- 239000011159 matrix material Substances 0.000 claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000013507 mapping Methods 0.000 claims abstract description 29
- 238000012360 testing method Methods 0.000 claims abstract description 22
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 19
- 239000010410 layer Substances 0.000 claims description 133
- 239000011229 interlayer Substances 0.000 claims description 39
- 230000006870 function Effects 0.000 claims description 22
- 238000003062 neural network model Methods 0.000 claims description 13
- 230000007246 mechanism Effects 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 5
- 239000012634 fragment Substances 0.000 claims description 5
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 claims description 4
- 230000002776 aggregation Effects 0.000 claims description 4
- 238000004220 aggregation Methods 0.000 claims description 4
- 239000003546 flue gas Substances 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 239000000203 mixture Substances 0.000 claims description 4
- 239000013598 vector Substances 0.000 claims description 4
- 230000004931 aggregating effect Effects 0.000 claims description 2
- 210000002268 wool Anatomy 0.000 claims description 2
- 238000000227 grinding Methods 0.000 description 15
- 238000010248 power generation Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 7
- 239000000843 powder Substances 0.000 description 4
- 238000005096 rolling process Methods 0.000 description 4
- 239000002356 single layer Substances 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000013508 migration Methods 0.000 description 3
- 230000005012 migration Effects 0.000 description 3
- 230000000644 propagated effect Effects 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 239000002817 coal dust Substances 0.000 description 2
- 238000002485 combustion reaction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- 238000005299 abrasion Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000008485 antagonism Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000005111 flow chemistry technique Methods 0.000 description 1
- 239000007789 gas Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011068 loading method Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007790 scraping Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a coal mill unit fault diagnosis method and system based on a multi-layer graph convolutional neural network, wherein the method comprises the following steps: collecting fault data of different working conditions of the coal mill unit by using a sensor; normalizing sensor data, uniformly intercepting and obtaining N samples by utilizing a sliding window strategy, and then mapping fault samples under different working conditions to the same feature space through a domain countermeasure model; constructing a feature matrix X by the features of the N mapped fault samples, constructing a multi-layer network by three different indexes, and obtaining a corresponding adjacent matrix A, wherein the fault labels are labels of nodes in the multi-layer network; the fault samples are divided into a training set and a testing set, a feature matrix X and different adjacent matrixes A are input into a graph convolution neural network, a prediction label is output through convolution, and a cross entropy loss function is utilized to iterate a training model. The method solves the technical problems of poor diagnosis performance and low diagnosis precision under the variable working condition without considering the influence of various complex correlations among samples on fault diagnosis in the prior art.
Description
Technical Field
The invention relates to the field of maintenance of coal-fired processing equipment, in particular to a coal mill unit fault diagnosis method and system based on a multi-layer graph convolution neural network.
Background
Although the development speed of new energy power generation technologies such as wind power generation, water power generation, nuclear power generation and the like is high in recent years, thermal power generation is used as a current main power generation mode in China and still occupies more than 70% of the total power generation amount in China. In a thermal power generation system, a coal mill unit is an important component in the system, and is a key point for improving the utilization rate of raw materials and the productivity of the thermal power plant. Therefore, the detection and diagnosis of the running state of the coal mill unit in the thermal power plant are preconditions for ensuring the safe and stable running of the thermal power generation system, and are also important directions for the development of the intelligent power grid.
The coal mill unit is core equipment of a thermal power plant and is used for crushing and grinding coal blocks into coal dust, and then the coal dust is conveyed to a hearth by a primary fan to supply combustion power generation. In the actual working process, the coal mill unit has the characteristics of long-time high-load operation, complex unit structure and the like, and the probability of failure of the coal mill unit is greatly increased. Once the coal mill set fails, the combustion of the boiler can be affected, the production efficiency of the thermal power plant is reduced, and the safety of other equipment and production staff is threatened, so that the intelligent diagnosis of the failure of the coal mill set becomes more necessary and urgent. If the intelligent fault diagnosis can be carried out on the coal mill unit when the fault occurs, the position and the type of the fault are accurately positioned, and the intelligent fault diagnosis method has important theoretical value and practical significance for safe operation of the unit and economic benefit of a thermal power plant.
Early fault diagnosis methods are mainly artificial experience methods, and the methods have the characteristics of easy understanding and wide application range, but when a system is complex, the diagnosis accuracy is low and the method is easily influenced by subjective factors. With advances in sensor technology and improvements in data storage capacity, data-driven fault diagnosis methods have become a research hotspot. As a typical class of data-driven approaches, deep learning-based fault diagnosis studies have achieved a great deal of theoretical and practical results. For example, the method disclosed in the prior patent document of publication No. CN106153179A is characterized in that firstly, fault characteristic analysis is carried out on the vibration of a coal mill to obtain the current of the coal mill which is an important monitoring parameter for the vibration fault analysis of the coal mill, then, four auxiliary variables including the wind pressure of a grinding outlet, the coal feeding amount of the coal feeder, the primary air quantity of the grinding inlet and the temperature of the grinding outlet are selected from analysis and screening through correlation of historical data, the current is predicted by the four auxiliary variables, the actual measurement current and the predicted current are used for differencing, the method of constructing the grinding vibration quantity by taking a residual sequence of the current of the coal mill is adopted, three-layer wavelet packet decomposition is carried out on the residual sequence of the grinding current, the energy ratio of 8 frequency bands is obtained, the energy ratio of the two faults is statistically analyzed, and the characteristic quantities of the two faults of serious abrasion of a grinding roller and foreign matter entering in the grinding are obtained. As is known from the specific implementation content of the prior art, in the fault diagnosis scheme of the coal mill disclosed in the document, the differential pressure of an inlet and an outlet of the coal mill, the coal feeding amount of the coal feeder, the primary air quantity of an inlet and the temperature of an outlet of the coal mill which are subjected to data pretreatment are used as the input of a coal mill current prediction model, a current prediction signal is obtained through prediction of a BP neural network, the actual value of the coal mill current is differenced from the current prediction value of the coal mill, and the obtained current residual sequence signal of the coal mill characterizes the vibration of the coal mill. For example, the prior patent application document with publication number of CN115496188A, namely a coal mill fault early warning method based on a deep learning convolutional neural network, is used for establishing a CNN-LSTM-Attention fault early warning model by acquiring measuring point parameter data of a coal mill in a historical normal working state, selecting measuring point parameter data related to coal blocking fault of the coal mill as an input variable and performing online training; obtaining a residual sequence between a predicted value ym and an actual value y obtained by a CNN-LSTM-Attention fault early warning model, and obtaining an average deviation sequence through a sliding window method; and calculating a probability density function of the average deviation sequence by using a kernel density estimation method. However, when faced with large-scale, non-euclidean structured fault data under different conditions, the neural networks employed in the foregoing prior art, as well as the conventional CNN, DBN, etc. models, have poor diagnostic performance and even fail. Meanwhile, in the existing fault diagnosis model based on deep learning, only one index is generally adopted to construct a network, which is not beneficial to describing various complex correlations existing between samples, and the characteristic information of different types of neighbors is difficult to aggregate, so that the fault diagnosis precision is low. Furthermore, although GCNs are capable of handling non-european graph data, they cannot be used directly in multi-layer networks.
In summary, the influence of multiple complex correlations among samples on fault diagnosis is not considered in the prior art, and the technical problems of poor diagnosis performance and low diagnosis precision under the variable working condition exist.
Disclosure of Invention
The invention aims to solve the technical problems that the influence of various complex correlations among samples on fault diagnosis is not considered in the prior art, and the diagnosis performance is poor and the diagnosis precision is low under the variable working condition is solved.
The invention adopts the following technical scheme to solve the technical problems: the fault diagnosis method of the coal mill unit based on the multi-layer graph convolution neural network comprises the following steps:
s1, collecting fault data of different working conditions of a coal mill set by using a sensor;
s2, normalizing and processing difference working condition fault data, intercepting average length data fragments by utilizing a sliding window strategy to obtain N difference working condition fault samples, and mapping the difference working condition fault samples to the same feature space by utilizing a domain countermeasure model to obtain mapped fault samples;
s3, constructing a feature matrix X by utilizing the characteristics of N mapping fault samples, and constructing a multi-layer network through at least 2 preset indexes to obtain an adjacent matrix A, wherein corresponding nodes in the multi-layer network are the same mapping fault sample, namely the fault labels of the samples are common node labels of the same nodes in the multi-layer network, so that the model parameters can be updated by minimizing a loss function;
s4, dividing N mapping fault samples into a training set and a testing set, inputting a feature matrix X and a difference adjacent matrix A into a graph convolution neural network to perform interlayer convolution and intra-layer convolution, outputting a prediction label of the training set, and performing iterative training by using a cross entropy loss function to obtain a multi-layer graph convolution neural network model;
s5, inputting the test set into the applicable multi-layer graph convolution neural network model to perform fault diagnosis, and measuring the classification accuracy of the applicable multi-layer graph convolution neural network model by using the Acc accuracy index and the Macro-F1 index.
The invention solves the problem of fault diagnosis of the coal mill unit under different working conditions based on domain countermeasure and multi-layer graph convolutional neural network MGCN, and the domain countermeasure can map fault data under different working conditions to the same feature space, thereby being more close to the actual working scene of the coal mill unit and improving the robustness and practicability of the model. The multi-layer graph convolutional neural network MGCN can better represent various potential correlations among the nodes, and an interlayer and layer convolutional mechanism can aggregate information of different neighbor nodes for the nodes, so that fault diagnosis performance is improved.
In a more specific embodiment, the measured variables of the sensor in step S1 include: coal mill current, coal amount of a coal feeder, inlet air temperature of the coal mill, bearing temperature of an induced draft fan, primary air fan current, blower current and air preheater outlet flue gas temperature.
The variable working condition fault diagnosis framework of the coal mill unit based on the multi-layer graph convolution neural network MGCN provided by the invention can process and analyze large-scale non-Euclidean structured data under different working conditions, overcomes the defects of insufficient data characteristics and difficult realization of working condition migration in the existing single-layer graph network method, and has better fault diagnosis performance and stronger method practicability.
In a more specific technical solution, step S2 includes:
s21, processing the difference working condition fault data by using a maximum and minimum normalization method to obtain normalized fault data, and intercepting and obtaining the characteristics of N difference working condition fault samples from the normalized fault data through a sliding window w;
s22, taking one working condition data in the difference working condition fault sample as a source domain and the rest working condition data as a target domain;
s23, mapping the fault wool of the different working conditions to the same feature space by using a domain countermeasure model according to the source domain and the target domain, and obtaining a mapping fault sample.
According to the invention, data samples under different working conditions are mapped to the same feature space, and meanwhile, the intra-layer convolution and inter-layer convolution in the multi-layer graph convolution neural network MGCN can effectively extract and aggregate information of different neighbor nodes to improve training performance of the model, so that accuracy of fault diagnosis and practicability of the method are improved. According to the invention, the domain countermeasure strategy is utilized to map the fault data under different working conditions into the same feature space, so that the conversion between different working conditions can be realized, the method is more close to an actual working scene, and the robustness of the model can be improved.
In a more specific technical scheme, in step S21, the following maximum and minimum normalization logic is used to process the differential working condition fault data:
wherein x is the original data, x max And x min Respectively, the maximum and minimum values in the original data.
In a more specific technical solution, step S3 includes:
s31, utilizing N mapping failure samplesThe feature matrix X epsilon R of the feature composition diagram N×w ；
S32, calculating to obtain the distance between the current node and other nodes and the similarity between the nodes according to the feature matrix;
s33, sorting the nodes according to the distances among the nodes and the similarity among the nodes;
s34, taking K minimum distance nodes and K maximum similarity nodes as neighbor nodes of the current node, and obtaining a first graph corresponding adjacent matrix A according to the K minimum distance nodes and the K maximum similarity nodes 1 ∈R N×N Adjacent matrix A corresponding to the second graph 2 ∈R N×N ；
S35, taking the previous time node and the next time node of the current node as neighbor nodes to obtain a third graph corresponding adjacent matrix A 3 ∈R N×N Thereby extracting the time sequence relation among the nodes.
In a more specific embodiment, in step S32, the inter-node distance S is obtained by the following logic calculation 1 Similarity S between nodes 2 ：
Where a and b are feature vectors of any two nodes.
In a more specific technical solution, step S4 includes:
s41, combining the feature matrix X and the differential adjacent matrix A l (l=1, 2, 3) input to the multi-layer graph convolutional network model;
s42, aggregating neighbor node information of different structures for the nodes by using an intra-layer convolution and an inter-layer convolution mechanism of the multi-layer graph convolution network model, and learning two groups of GCN parameters to decouple intra-layer and inter-layer propagation;
s42, summing the graph network feature matrixes in not less than 2 to obtain aggregate node features;
s43, performing iterative training by using the softmax function and the cross entropy loss function to obtain the applicable multi-layer graph convolution neural network model.
The intra-layer and inter-layer convolution mechanism adopted by the invention decouples intra-layer and inter-layer propagation by learning two groups of GCN parameters, so that the model can learn different importance in two propagation directions. The multi-layer graph rolling network MGCN model adopted by the invention can better represent various correlations among samples by constructing a multi-layer network. And then, node characteristics are independently propagated by utilizing different GCN layers, and node representations of different layers are aggregated through intra-layer and inter-layer convolution, so that the diagnostic performance is improved.
In a more specific technical solution, in step S42, the following logic is used to represent the neighbor node information of different structures:
where iα represents an i-th node in the alpha-th layer network, intra-k represents a k-th intra-layer convolution, inter-k represents a k-th inter-layer convolution,representing the characteristics of node i after k+1 convolution aggregations, x iα Is an input feature of the node i of the alpha layer graph.
In a more specific embodiment, in step S5,
the Acc index is represented by the following logic:
wherein TP is the number of positive samples predicted to be positive, FP is the number of negative samples predicted to be positive, TN is the number of negative samples predicted to be negative, FN is the number of positive samples predicted to be negative;
the Acc index is represented by the following logic:
where TP is the number of positive samples predicted to be positive, FP is the number of negative samples predicted to be positive, TN is the number of negative samples predicted to be negative, and FN is the number of positive samples predicted to be negative.
In a more specific technical scheme, the fault diagnosis system of the coal mill unit based on the multi-layer graph convolution neural network comprises:
the acquisition module is used for collecting fault data of different working conditions of the coal mill unit by using the sensor;
the domain countermeasure module is used for normalizing and processing the difference working condition fault data, intercepting the average length data fragments by utilizing a sliding window strategy to obtain N difference working condition fault samples, mapping the difference working condition fault samples to the same feature space by utilizing a domain countermeasure model to obtain mapped fault samples, and connecting the domain countermeasure module with the acquisition module;
the adjacent matrix acquisition module is used for constructing a feature matrix X by utilizing the characteristics of N mapping fault samples, and constructing a multi-layer network through at least 2 preset indexes to obtain an adjacent matrix A, so as to acquire a multi-layer network node label of the mapping fault samples, update model parameters by a minimized loss function, and connect with the domain countermeasure module;
the model iterative training module is used for dividing N mapping fault samples into a training set and a testing set, inputting a feature matrix X and a difference adjacent matrix A into the graph convolution neural network to perform interlayer convolution and intra-layer convolution, outputting a prediction label of the training set, performing iterative training by using a cross entropy loss function to obtain a model applicable to the multi-layer graph convolution neural network, and connecting the model iterative training module with the adjacent matrix acquisition module;
the fault diagnosis and precision measurement module is used for inputting the test set into the applicable multi-layer graph convolution neural network model to perform fault diagnosis, and measuring the classification precision of the applicable multi-layer graph convolution neural network model by using the Acc accuracy index and the Macro-F1 index, wherein the fault diagnosis and precision measurement module is connected with the model iteration training module.
Compared with the prior art, the invention has the following advantages: the invention solves the problem of fault diagnosis of the coal mill unit under different working conditions based on domain countermeasure and multi-layer graph convolutional neural network MGCN, and the domain countermeasure can map fault data under different working conditions to the same feature space, thereby being more close to the actual working scene of the coal mill unit and improving the robustness and practicability of the model. The multi-layer graph convolutional neural network MGCN can better represent various potential correlations among the nodes, and an interlayer and layer convolutional mechanism can aggregate information of different neighbor nodes for the nodes, so that fault diagnosis performance is improved.
The variable working condition fault diagnosis framework of the coal mill unit based on the multi-layer graph convolution neural network MGCN provided by the invention can process and analyze large-scale non-Euclidean structured data under different working conditions, overcomes the defects of insufficient data characteristics and difficult realization of working condition migration in the existing single-layer graph network method, and has better fault diagnosis performance and stronger method practicability.
The invention maps the data samples of different working conditions to the same feature space, and meanwhile, the intra-layer and inter-layer convolution in the MGCN can effectively extract and aggregate the information of different neighbor nodes to improve the training performance of the model, thereby improving the accuracy of fault diagnosis and the practicability of the method. According to the invention, the domain countermeasure strategy is utilized to map the fault data under different working conditions into the same feature space, so that the conversion between different working conditions can be realized, the method is more close to an actual working scene, and the robustness of the model can be improved.
The intra-layer and inter-layer convolution mechanism adopted by the invention decouples intra-layer and inter-layer propagation by learning two groups of GCN parameters, so that the model can learn different importance in two propagation directions. The multi-layer graph rolling network MGCN model adopted by the invention can better represent various correlations among samples by constructing a multi-layer network. And then, node characteristics are independently propagated by utilizing different GCN layers, and node representations of different layers are aggregated through intra-layer and inter-layer convolution, so that the diagnostic performance is improved. The invention solves the technical problems of poor diagnosis performance and low diagnosis precision under the variable working condition without considering the influence of various complex correlations among samples on fault diagnosis in the prior art.
Drawings
FIG. 1 is a schematic diagram of data flow processing of a coal mill unit fault diagnosis method based on a multi-layer graph convolutional neural network according to embodiment 1 of the present invention;
FIG. 2 is a schematic diagram of the basic steps of a coal mill unit fault diagnosis method based on a multi-layer convolutional neural network according to embodiment 1 of the present invention;
FIG. 3 is a schematic view of a medium speed coal pulverizer in accordance with embodiment 2 of the present invention;
FIG. 4a is a graph showing variation of test set Acc values with iteration number according to example 2 of the present invention;
FIG. 4b is a graph showing the variation of the Macro-F1 value with the number of iterations in example 2 of the present invention;
FIG. 5 is a confusion matrix diagram of test set failure samples of embodiment 2 of the present invention;
fig. 6 is a graph of the diagnostic results of various faults in the test set of embodiment 2 of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly and completely described in the following in conjunction with the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
As shown in fig. 1 and fig. 2, the fault diagnosis method for the coal mill unit based on the multi-layer graph convolution neural network provided by the invention comprises the following basic steps:
s1, collecting fault data of a coal mill unit under different working conditions by using an installed sensor;
in this embodiment, the measured variables of the sensor in step S1 mainly include: coal mill current, coal amount of a coal feeder, inlet air temperature of the coal mill, bearing temperature of an induced draft fan, primary air fan current, blower current, outlet flue gas temperature of an air preheater and the like; the fault categories of the coal mill set mainly comprise: the oil station of the blower A has high filter screen difference, the oil pump of the primary blower B has abnormal vibration, the coal mill A has large vibration, the bearing of the induced draft fan B has large vibration, the coal mill C has large vibration, and the like. In this embodiment, the amount of coal fed determines the operating conditions of the coal mill train.
S2, carrying out normalization processing on the collected sensor data, uniformly intercepting data fragments by utilizing a sliding window strategy to obtain N samples, and then mapping fault samples under different working conditions to the same feature space through a domain countermeasure model;
in this embodiment, the data of step S2 is processed by the maximum-minimum normalization method:
where x is the original data, x max And x min And respectively obtaining the maximum value and the minimum value in the original data, intercepting the characteristics of N data samples through a sliding window w, taking one working condition data as a source domain, taking the data of other working conditions as a target domain, and finally mapping the data into the same characteristic space through a domain countermeasure model.
S3, forming a feature matrix X by the features of the N mapped fault samples, constructing a multi-layer network through three different indexes, and obtaining a corresponding adjacent matrix A, wherein the fault label is the label of a node in the multi-layer network;
in this embodiment, step S3 includes mapping N feature matrices X ε R of the feature composition map of the failure samples N×w The fault sample features are node features, and then the distance and the similarity between the nodes are obtained through calculation by using an Euclidean distance formula and a cosine similarity formula, and the distance and the similarity can be expressed as follows:
where a and b are feature vectors of any two nodes. The adjacent matrix A corresponding to the graph can be obtained by calculating the distance and the similarity between the current node and each other node, sequencing each distance and each similarity, and then respectively selecting K nodes with the minimum distance and K nodes with the maximum similarity as the adjacent nodes of the node 1 ∈R N×N And A 2 ∈R N ×N . Meanwhile, in order to extract the time sequence relation between the nodes, the nodes at the previous time and the next time of the node are selected as neighbor nodes, and then the adjacent matrix A corresponding to the graph can be obtained 3 ∈R N×N And the class label of the fault is the label of the node.
S4, dividing N fault samples into a training set and a testing set, inputting a feature matrix X and different adjacent matrixes A into a graph convolution neural network, outputting a prediction label of the training set through interlayer and layer convolution, and finally performing iterative training of a model by using a cross entropy loss function;
in the present embodiment, step S4 combines the obtained feature matrix X with a different adjacent matrix a l (l=1, 2, 3) is input into a multi-layer graph convolution model, and information of neighbor nodes with different structures for node aggregation by using an intra-layer and inter-layer convolution mechanism of the model can be expressed as:
where iα represents an i-th node in the alpha-th layer network, intra-k represents a k-th intra-layer convolution, inter-k represents a k-th inter-layer convolution,representing the characteristics of node i after k+1 convolution aggregations, x iα Is an input feature of the node i of the alpha layer graph. And then, obtaining the aggregated node characteristics by summing the characteristic matrixes of the three graph networks:
H i ＝sum(H α ),α＝1,2,3
wherein H is α Is a feature matrix corresponding to different layers. The intra-and inter-layer convolution mechanism decouples intra-and inter-layer propagation by learning two sets of GCN parameters, enabling the model to learn different importance in the two propagation directions. And finally, performing iterative training of the model by using a softmax function and a cross entropy loss function.
And S5, inputting the test set into the trained multi-layer graph convolutional neural network to perform fault diagnosis, and measuring the overall classification accuracy of the model by using an accuracy (Acc) index and a Macro-F1 index.
In this embodiment, step S5 uses a test set to test the trained model, then outputs a fault class label of the test set, and finally measures the overall classification accuracy of the model through an Acc index and a Macro-F1 index, where the Acc index may be expressed as:
where TP is the number of positive samples predicted to be positive, FP is the number of negative samples predicted to be positive, TN is the number of negative samples predicted to be negative, and FN is the number of positive samples predicted to be negative. In the multi-classification problem, the Acc index is defined as the ratio of the number of correctly classified samples to the total number of samples.
In this embodiment, precision (Precision) and Recall (Recall) are a pair of contradictory metrics, and generally, when Precision is high, the Recall value tends to be low, and vice versa. To consider both metrics in combination, the overall performance of the fault diagnosis of the model is measured using the F1score metric, which can be expressed as:
however, the F1score index is only applicable to two classification problems, and as an improved form of the F1score index, the Macro-F1 index first calculates TP, FP, FN, TN of each class, then obtains respective F1score values on the basis of the TP, FP, FN, TN, and finally averages all the F1score values.
Example 2
As shown in fig. 3, in the present embodiment, the medium speed coal mill includes: the device comprises a coal dropping pipe 1, an outlet gas sealing system 2, a discharge valve and multiple outlets device 3, a separator top cover device 4, a reverse cone device 5, a separator body device 6, an inner cone device 7, a spring loading device 8, a grinding roller device 9, a side machine body device 10, a grinding bowl and impeller device 11, a scraping plate device 12 and a planetary gear reduction box 13. Raw coal in the medium-speed coal mill firstly falls onto a grinding bowl from a coal dropping pipe 1, the raw coal is ground into coal powder through a grinding roller device 9, the coal powder is thrown to the edge through rotation and is brought into a separator body device 6 by hot air for separation, the coarse powder falls into the grinding bowl from an inner cone for re-grinding, and the fine powder is collected along with air flow from a discharge valve and a multi-outlet device 3 for grinding. In this embodiment, step S1 in embodiment 1 is performed, and the coal mill unit is mainly composed of a coal mill A, B, C, D, a primary fan A, B, an induced fan A, B, a blower A, B, an air preheater A, B, and the like; the sensor data are mainly collected on key measuring points of each device, and comprise sensor data for monitoring 172 variables such as coal mill current, coal amount of a coal feeder, inlet air temperature of the coal mill, bearing temperature of a draught fan, primary fan current, blower current, outlet flue gas temperature of an air preheater and the like; the working condition of the coal mill unit can be divided into 5 working conditions by the coal feeding amount, wherein the coal feeding amount is 35-40 t/h in working condition 1, 40-45 t/h in working condition 2, 45-50 t/h in working condition 3, 50-55 t/h in working condition 4 and 55-60 t/h in working condition 5.
The time span of data collection of the coal mill unit is 15 months, normal data are collected once every 5 minutes, fault data are collected once every 1 second, 32 types of fault data including high differential pressure of an oil station filter screen of an A blower, abnormal vibration of an oil pump of a B primary blower, large vibration of a coal mill, large vibration of a bearing of a B induced draft fan, large vibration of a C coal mill and the like are collected together, the data comprise 5 working conditions, and all the data are derived from the actual operation process of a coal mill of a certain power plant in China east.
In this embodiment, step S2 in embodiment 1 is performed, the maximum and minimum normalization method is used to normalize 172 sensor features of the fault data, so as to eliminate the influence of dimension, and then a sliding window w (with a value of 1) is used to intercept and obtain the feature of the node, where each node feature represents the fault data corresponding to each time point, that is, the dimension of the node feature is 172 dimensions. And then taking each working condition as a source domain and the other working conditions as target domains in turn, and finally mapping the sample of the target domain into the same feature space of the source domain by using a domain countermeasure model.
In the domain countermeasure model, the domain LABEL of the source domain is set to 1, the domain LABEL of the target domain is set to 0, the feature dimension of mapping all sample data to the same feature space is set to 172 dimensions, and then the domain LABEL LOSS los_label and the edge distribution LOSS los_mmd are used to constrain the confusion degree between the source domain and the target domain, so that the LOSS function of the domain countermeasure model is:
LOSS＝LOSS_MMD+LOSS_LABEL
wherein n is s And n t Representing the number of samples of the source domain and the target domain, x, respectively i Is characteristic of the ith sample, D s And D t Sample space representing source domain and target domain, respectively, D c Representing domain discriminator, d i Representing the actual domain label of sample i, L () is a binary cross entropy loss function, Φ () represents a nonlinear mapping function that projects the original data into hilbert and space,is characteristic of the ith sample in the source domain, < +.>As a feature of the jth sample in the target domain, LOSS represents the total LOSS function in the domain countermeasure process.
In this embodiment, step S3 in embodiment 1 is performed, and the feature matrix H e R of the mapped N fault sample feature composition map is formed N×172 Then calculating to obtain the distance S between any two nodes by using the Euclidean distance formula and the cosine similarity formula 1 And similarity S 2 Then sequencing each distance and similarity, and selecting K nodes with the smallest distance and K nodes with the largest similarity as neighbor nodes of the nodes respectively to obtain an adjacent matrix A corresponding to the network 1 ∈R N×N And A 2 ∈R N×N . Meanwhile, in order to extract the time sequence relation between the nodes, the nodes at the previous time and the next time of the node are selected as neighbor nodes, and then the adjacent matrix A corresponding to the graph can be obtained 3 ∈R N×N . After respectively carrying out intra-layer convolution operation on the three-layer network with different structures, constructing an adjacent matrix of the inter-layer nodes, respectively connecting the same nodes of different layers two by two, and obtaining the adjacent matrix A of the inter-layer without connecting edges between the different nodes 4 ∈R 3N×3N And the class label of the failure sample is the label of the node.
In this embodiment, step S4 in embodiment 1 is performed, all the mapped fault samples are divided into a training set and a test set, the samples of the training set and the test set respectively account for 70% and 30% of all the samples, and then the training set is input into the MGCN. The MGCN model adopts two to input data H on three-layer diagramThe layer-picture-volume-layer (GCNConv), i.e. the first picture convolution layer (GCNConv 1) and the second picture convolution layer (GCNConv 2), performs space-time feature extraction, and the outputs of the two layer-picture-volume-layers can be expressed as H respectively α (1) And H α (2) ：
Wherein, the liquid crystal display device comprises a liquid crystal display device,representing the feature vector of node i in H in the alpha layer network, the node feature matrix of the three layers of the initially input network is the same, < >>And->The dimension of the two node feature matrices is 256-dimension and 128-dimension respectively, and ReLU represents an activation function, W α (1) And W is α (2) The laminated weight matrix is rolled up for two layers of the alpha layer graph.
After the first graph convolution is respectively carried out on each layer of network, the node characteristics on the three layers of networks are spliced into the following steps:
inter-layer feature representation H at acquisition node 4 Then, the interlayer adjacent matrix A is reused 4 And (3) carrying out interlayer graph rolling operation to obtain an output interlayer characteristic matrix, wherein the dimension is 128 dimensions:
meanwhile, after second graph convolution is respectively carried out on each layer of network, node features on three layers of networks are spliced according to rows, and an output intra-layer feature matrix can be obtained as follows:
and then summing the output interlayer characteristic matrix and the internal characteristic matrix to obtain an output characteristic matrix of the node in the three-layer network, wherein the output characteristic matrix is as follows:
H out ＝sum(H intra ,H inter ),H out ∈R 3N×128
after obtaining the node characteristic matrix of the three-layer graph, namely H 1 Is H out Matrix of the first N rows of (H) 2 Is H out Matrix of (n+1) th row to (2N) th row, H 3 Is H out The node characteristic matrix after the interlayer and the intra-layer convolution can be obtained by summing the characteristics of the same nodes of different layers of networks according to the matrix from 2N+1th row to 3N row:
two fully-connected layers are designed after an intra-layer and inter-layer convolution mechanism of the MGCN model, the first fully-connected layer reduces the dimension of the extracted features, the second fully-connected layer acts as a classifier in the model to classify the extracted features, and in addition, the nonlinear expression capacity of the model can be improved by designing the two fully-connected layers. Finally, carrying out iterative training of the model by utilizing a prediction label and a real label of the sample and through a cross entropy loss function, and observing the trend of the loss of the training set and the change of the Acc index and the Macro-F1 index of the test set to select the optimal parameters of the model.
As shown in fig. 4a, fig. 4b, fig. 5, and fig. 6, in this embodiment, step S5 in the embodiment is executed, domain antagonism training is performed by using the working condition 1 as a source domain to obtain a mapped fault sample, and the training sample is input into a multi-layer graph convolution model to perform model training, so that it can be observed that the Acc index and the Macro-F1 index of the test set show an ascending trend with the increase of the iteration times. In this embodiment, the working condition 1 coal feeding interval includes: 35-40 t/h. And then the overall classification accuracy of the model is measured through a confusion matrix of the classification results of the test set, and the classification accuracy of each type of faults in the model can be measured through the Acc indexes and the F1score indexes of each type of faults. Next, domain countermeasure training is performed by taking different working conditions as source domains, and results of Acc indexes and Macro-F1 indexes of the MGCN model and the GCN and MLP models are compared, in the embodiment, coal feeding amounts of working conditions 2-5 are respectively 40-45 t/h, 45-50 t/h, 50-55 t/h and 55-60 t/h, as shown in tables 1 and 2:
table 1 Acc index comparison of various models mapped to different conditions
| Working condition 1 | Working condition 2 | Working condition 3 | Working condition 4 | Working condition 5 | |
| MLP model | 0.7086 | 0.8188 | 0.8155 | 0.8067 | 0.7983 |
| GCN model | 0.8502 | 0.9241 | 0.9143 | 0.9511 | 0.9465 |
| MGCN model | 0.9293 | 0.9926 | 0.9784 | 0.9966 | 0.9934 |
TABLE 2 Macro-F1 index comparison of models mapped to different conditions
| Working condition 1 | Working condition 2 | Working condition 3 | Working condition 4 | Working condition 5 | |
| MLP model | 0.6916 | 0.8152 | 0.8021 | 0.8034 | 0.7890 |
| GCN model | 0.8485 | 0.9201 | 0.9130 | 0.9500 | 0.9452 |
| MGCN model | 0.9277 | 0.9926 | 0.9779 | 0.9965 | 0.9936 |
The comparison results of the three algorithms under different working conditions show that the overall Acc index of the MGCN model is about 20% higher than that of the MLP model and about 6% higher than that of the single-layer GCN model, and meanwhile, the overall Macro-F1 index is improved in the same way. The above results show that the MGCN model of the present invention has good fault diagnosis performance, high robustness and high practicability.
In summary, the invention solves the problem of fault diagnosis of the coal mill unit under different working conditions based on domain countermeasure and multi-layer graph convolutional neural network MGCN, and the domain countermeasure can map fault data under different working conditions to the same feature space, thereby being more close to the actual working scene of the coal mill unit and improving the robustness and practicability of the model. The multi-layer graph convolutional neural network MGCN can better represent various potential correlations among the nodes, and an interlayer and layer convolutional mechanism can aggregate information of different neighbor nodes for the nodes, so that fault diagnosis performance is improved.
The variable working condition fault diagnosis framework of the coal mill unit based on the multi-layer graph convolution neural network MGCN provided by the invention can process and analyze large-scale non-Euclidean structured data under different working conditions, overcomes the defects of insufficient data characteristics and difficult realization of working condition migration in the existing single-layer graph network method, and has better fault diagnosis performance and stronger method practicability.
The invention maps the data samples of different working conditions to the same feature space, and meanwhile, the intra-layer and inter-layer convolution in the MGCN can effectively extract and aggregate the information of different neighbor nodes to improve the training performance of the model, thereby improving the accuracy of fault diagnosis and the practicability of the method. According to the invention, the domain countermeasure strategy is utilized to map the fault data under different working conditions into the same feature space, so that the conversion between different working conditions can be realized, the method is more close to an actual working scene, and the robustness of the model can be improved.
The intra-layer and inter-layer convolution mechanism adopted by the invention decouples intra-layer and inter-layer propagation by learning two groups of GCN parameters, so that the model can learn different importance in two propagation directions. The multi-layer graph rolling network MGCN model adopted by the invention can better represent various correlations among samples by constructing a multi-layer network. And then, node characteristics are independently propagated by utilizing different GCN layers, and node representations of different layers are aggregated through intra-layer and inter-layer convolution, so that the diagnostic performance is improved. The invention solves the technical problems of poor diagnosis performance and low diagnosis precision under the variable working condition without considering the influence of various complex correlations among samples on fault diagnosis in the prior art.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention 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 invention.
Claims (10)
1. The fault diagnosis method of the coal mill unit based on the multi-layer graph convolution neural network is characterized by comprising the following steps of:
s1, collecting fault data of different working conditions of a coal mill set by using a sensor;
s2, normalizing the difference working condition fault data, intercepting average length data fragments by utilizing a sliding window strategy to obtain N difference working condition fault samples, and mapping the difference working condition fault samples to the same feature space by utilizing a domain countermeasure model to obtain mapped fault samples;
s3, constructing a feature matrix X by utilizing the characteristics of N mapping fault samples, and constructing a multi-layer network through at least 2 preset indexes to obtain an adjacent matrix A, thereby obtaining multi-layer network node labels of the mapping fault samples for updating model parameters by a minimized loss function;
s4, dividing N mapping fault samples into a training set and a testing set, inputting the feature matrix X and the difference adjacent matrix A into a graph convolution neural network to perform interlayer convolution and intra-layer convolution, outputting a prediction label of the training set, and performing iterative training by using a cross entropy loss function to obtain a multi-layer graph convolution neural network model;
s5, inputting the test set into an applicable multi-layer graph convolution neural network model to perform fault diagnosis, and measuring the classification precision of the applicable multi-layer graph convolution neural network model by using an Acc accuracy index and a Macro-F1 index.
2. The method for diagnosing a fault in a coal pulverizer set based on a multi-layer convolutional neural network according to claim 1, wherein the measured variables of the sensor in step S1 include: coal mill current, coal amount of a coal feeder, inlet air temperature of the coal mill, bearing temperature of an induced draft fan, primary air fan current, blower current and air preheater outlet flue gas temperature.
3. The method for diagnosing a fault in a coal pulverizer set based on a multi-layer convolutional neural network according to claim 1, wherein the step S2 comprises:
s21, processing the difference working condition fault data by using a maximum and minimum normalization method to obtain normalized fault data, and intercepting and obtaining the characteristics of N difference working condition fault samples from the normalized fault data through a sliding window w;
s22, taking one working condition data in the difference working condition fault sample as a source domain and the rest working condition data as a target domain;
s23, mapping the fault wool of the different working conditions to the same feature space by using a domain countermeasure model according to the source domain and the target domain, and obtaining the mapping fault sample.
4. The method for diagnosing a fault of a coal mill unit based on a multi-layer graph convolutional neural network according to claim 3, wherein in the step S21, the fault data of the different working conditions are processed by using the following maximum and minimum normalization logic:
wherein x is the original data, x max And x min Respectively, the maximum and minimum values in the original data.
5. The method for diagnosing a fault in a coal pulverizer set based on a multi-layer convolutional neural network according to claim 1, wherein the step S3 comprises:
s31, utilizing the characteristic matrix X epsilon R of the characteristic composition diagram of N mapping fault samples N×w ；
S32, calculating to obtain the distance between the current node and other nodes and the similarity between the nodes according to the feature matrix;
s33, sorting the nodes according to the distances among the nodes and the similarity among the nodes;
s34, taking K minimum distance nodes and K maximum similarity nodes as neighbor nodes of the current node, and obtaining a first graph corresponding adjacent matrix A according to the K minimum distance nodes and the K maximum similarity nodes 1 ∈R N×N Adjacent matrix A corresponding to the second graph 2 ∈R N×N ；
S35, taking the previous time node and the next time node of the current node as the neighbor nodes to obtain a third graph corresponding adjacent matrix A 3 ∈R N×N Thereby extracting the time sequence relation among the nodes.
6. The method for diagnosing a fault in a coal pulverizer set based on a multi-layer convolution neural network according to claim 5, wherein in said step S32, said inter-node distance S is calculated by using the following logic 1 Similarity S between the nodes 2 ：
Where a and b are feature vectors of any two nodes.
7. The method for diagnosing a fault in a coal pulverizer set based on a multi-layer convolutional neural network according to claim 1, wherein the step S4 comprises:
s41, the characteristic matrix X and the difference adjacent matrix A are processed l (l=1, 2, 3) input to the multi-layer graph convolutional network model;
s42, aggregating neighbor node information of different structures for the nodes by using an intra-layer convolution and an inter-layer convolution mechanism of the multi-layer graph convolution network model, and learning two groups of GCN parameters to decouple intra-layer and inter-layer propagation;
s42, summing the graph network feature matrixes in not less than 2 to obtain aggregate node features;
s43, performing iterative training by using a softmax function and a cross entropy loss function to obtain the applicable multi-layer graph convolution neural network model.
8. The method for diagnosing a fault of a coal mill unit based on a multi-layer graph convolutional neural network according to claim 1, wherein in the step S42, the following logic is used to represent the neighbor node information of the different structures:
where iα represents an i-th node in the alpha-th layer network, intra-k represents a k-th intra-layer convolution, inter-k represents a k-th inter-layer convolution,representing the characteristics of node i after k+1 convolution aggregations, x iα Is an input feature of the node i of the alpha layer graph.
9. The method for diagnosing a fault in a coal pulverizer set based on a multi-layer convolutional neural network as recited in claim 1, wherein in step S5,
the Acc index is represented by the following logic:
wherein TP is the number of positive samples predicted to be positive, FP is the number of negative samples predicted to be positive, TN is the number of negative samples predicted to be negative, FN is the number of positive samples predicted to be negative;
the Acc index is represented by the following logic:
where TP is the number of positive samples predicted to be positive, FP is the number of negative samples predicted to be positive, TN is the number of negative samples predicted to be negative, and FN is the number of positive samples predicted to be negative.
10. Coal mill unit fault diagnosis system based on multilayer graph convolution neural network, characterized in that, the system includes:
the acquisition module is used for collecting fault data of different working conditions of the coal mill unit by using the sensor;
the domain countermeasure module is used for normalizing the difference working condition fault data, intercepting average-length data fragments by utilizing a sliding window strategy to obtain N difference working condition fault samples, mapping the difference working condition fault samples to the same feature space by utilizing a domain countermeasure model to obtain mapped fault samples, and the domain countermeasure module is connected with the acquisition module;
the adjacent matrix acquisition module is used for constructing a feature matrix X by utilizing the characteristics of N mapping fault samples, constructing a multi-layer network through at least 2 preset indexes to obtain an adjacent matrix A, acquiring multi-layer network node labels of the mapping fault samples according to the adjacent matrix A, and updating model parameters by a minimized loss function, wherein the adjacent matrix acquisition module is connected with the domain countermeasure module;
the model iterative training module is used for dividing N mapping fault samples into training sets and test sets, inputting the feature matrix X and the difference adjacent matrix A into a graph convolution neural network to perform interlayer convolution and intra-layer convolution, outputting a prediction label of the training set, and performing iterative training by using a cross entropy loss function to obtain a multi-layer graph convolution neural network model, wherein the model iterative training module is connected with the adjacent matrix acquisition module;
the fault diagnosis and precision measurement module is used for inputting the test set into the applicable multi-layer graph convolution neural network model to perform fault diagnosis, and measuring the classification precision of the applicable multi-layer graph convolution neural network model by using an Acc accuracy index and a Macro-F1 index, and is connected with the model iteration training module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310606439.8A CN116821811A (en) | 2023-05-25 | 2023-05-25 | Coal mill unit fault diagnosis method and system based on multi-layer graph convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310606439.8A CN116821811A (en) | 2023-05-25 | 2023-05-25 | Coal mill unit fault diagnosis method and system based on multi-layer graph convolution neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116821811A true CN116821811A (en) | 2023-09-29 |
Family
ID=88117634
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310606439.8A Pending CN116821811A (en) | 2023-05-25 | 2023-05-25 | Coal mill unit fault diagnosis method and system based on multi-layer graph convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116821811A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117540175A (en) * | 2024-01-09 | 2024-02-09 | 海纳云物联科技有限公司 | Model training method, prediction method, device, equipment and medium |
-
2023
- 2023-05-25 CN CN202310606439.8A patent/CN116821811A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117540175A (en) * | 2024-01-09 | 2024-02-09 | 海纳云物联科技有限公司 | Model training method, prediction method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110361176B (en) | Intelligent fault diagnosis method based on multitask feature sharing neural network | |
| Zheng et al. | Research on rolling bearing fault diagnosis of small dataset based on a new optimal transfer learning network | |
| CN110929918A (en) | 10kV feeder line fault prediction method based on CNN and LightGBM | |
| CN112036089A (en) | Coal mill fault early warning method based on DPC-MND and multivariate state estimation | |
| Wang et al. | A fast abnormal data cleaning algorithm for performance evaluation of wind turbine | |
| CN110555478B (en) | Fan multi-fault diagnosis method based on depth measurement network of difficult sample mining | |
| CN116821811A (en) | Coal mill unit fault diagnosis method and system based on multi-layer graph convolution neural network | |
| CN109359662B (en) | Non-stationary analysis and causal diagnosis method for million-kilowatt ultra-supercritical unit | |
| CN114417971A (en) | Electric power data abnormal value detection algorithm based on K nearest neighbor density peak clustering | |
| CN111401126A (en) | Combustion condition monitoring method and device based on flame image and deep learning | |
| CN116562114A (en) | Power transformer fault diagnosis method based on graph convolution neural network | |
| CN114580934A (en) | Early warning method for food detection data risk based on unsupervised anomaly detection | |
| CN116467658A (en) | Equipment fault tracing method based on Markov chain | |
| CN111402249A (en) | Image evolution analysis method based on deep learning | |
| Yin et al. | Location of abnormal energy consumption and optimization of energy efficiency of hydraulic press considering uncertainty | |
| Ding et al. | Artificial intelligence based abnormal detection system and method for wind power equipment | |
| CN117725491A (en) | SCITET-based power system fault state detection and classification method | |
| CN111880090B (en) | Online fault detection method for million-kilowatt ultra-supercritical unit | |
| Tian et al. | Causal network construction based on convergent cross mapping (CCM) for alarm system root cause tracing of nonlinear industrial process | |
| CN112598057B (en) | Coal-fired power plant boiler fault diagnosis method based on FEKNN strategy | |
| CN108759479A (en) | Sintering machine based on the unilateral detection method of blind source separating leaks out on-line fault diagnosis method | |
| CN114548701A (en) | Process early warning method and system for analyzing and estimating coupling structure of full-scale measuring point | |
| CN112682269A (en) | Wind turbine generator state monitoring method based on OC-RKELM | |
| CN114358176A (en) | Method for controlling quality of original wind power data of wind power station based on LOF algorithm | |
| Shan et al. | Image feature-based for bearing health monitoring with deep-learning method |
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 |