CN112163474B - Intelligent gearbox diagnosis method based on model fusion - Google Patents
Intelligent gearbox diagnosis method based on model fusion Download PDFInfo
- Publication number
- CN112163474B CN112163474B CN202010966319.5A CN202010966319A CN112163474B CN 112163474 B CN112163474 B CN 112163474B CN 202010966319 A CN202010966319 A CN 202010966319A CN 112163474 B CN112163474 B CN 112163474B
- Authority
- CN
- China
- Prior art keywords
- model
- probability
- classification
- sample
- gearbox
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 47
- 230000004927 fusion Effects 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 36
- 101100519162 Arabidopsis thaliana PCR6 gene Proteins 0.000 claims description 12
- 238000013527 convolutional neural network Methods 0.000 claims description 10
- 238000007499 fusion processing Methods 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 150000001875 compounds Chemical class 0.000 abstract description 4
- 238000005096 rolling process Methods 0.000 abstract description 4
- 238000002474 experimental method Methods 0.000 abstract description 3
- 239000002131 composite material Substances 0.000 abstract description 2
- 230000000007 visual effect Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 238000012549 training Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 238000007500 overflow downdraw method Methods 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000035755 proliferation Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Acoustics & Sound (AREA)
- Signal Processing (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention provides a model fusion fault diagnosis method based on a proportion conflict distribution rule aiming at the problem that the diagnosis accuracy of a rolling bearing composite fault in a gearbox is low by a single visual angle characteristic and a single model. The method comprises the following steps: 1) And extracting features of vibration signals of the gearbox, and constructing features from the time domain and the time domain aiming at complex compound faults. 2) And sending the multi-view features into a plurality of sub-models for preliminary diagnosis to obtain a diagnosis result with strong complementarity. 3) The classification probabilities output by the model are fused by class 6 example conflict allocation rules. Experiments prove that the result obtained by the fusion model has higher stability, and the accuracy of fault diagnosis can be improved under certain conditions.
Description
Technical Field
The invention belongs to the field of intelligent diagnosis of big data of gearboxes, and particularly relates to a model fusion gearbox diagnosis method based on multiple artificial intelligent diagnosis models.
Background
In recent years, the development of wind power industry in China is rapid, and the wind power generation technology is increasingly widely applied. Besides the main bearing of the wind turbine generator, the rolling bearing is used as a key component and is arranged in a wind power gear box, the propagation path of the collected signals is complex, and the coupling faults of other components are often generated. Therefore, it is of great importance how to identify faults of the rolling bearing and other components of the gearbox from the composite faults. In addition to common bearing failures, gearbox failures include: broken teeth, pitting, tooth surface wear, etc. The existing more effective fault diagnosis method for the rolling bearing and the gearbox is vibration monitoring and spectrum analysis. However, these conventional methods often require a great deal of expertise, and are not able to meet the current requirements for real-time and accuracy of diagnosis. Many studies have combined signal processing techniques with artificial intelligence models to achieve satisfactory results.
With the proliferation of machine learning models, related studies for different model combinations have also emerged. The field of fault diagnosis utilizes the complementarity of the model to analyze the difference between different states of the equipment, and indicates a direction for diagnosing early faults and compound faults. The model fusion is to synchronously and comprehensively analyze the equipment information obtained from a plurality of models to form more objective and clearer diagnosis on the equipment state. The key of model fusion is the comprehensive treatment and analysis of a plurality of models, so that the obtained diagnosis result is more definite and more real. The single model is unfavorable for positioning faults and timely replacement of components because false alarms and missing alarms often occur in diagnosis of complex equipment. The multi-model fault fusion diagnosis provides a new way for improving the equipment diagnosis accuracy. In the multi-model fusion diagnosis process, the reliability of diagnosis is greatly improved, and compared with a single diagnosis model, the method has the advantages that the requirement on performance optimization of the single model is reduced by utilizing the diversity and complementarity of a model diagnosis mechanism. The fusion increases the stability of the results, and the model is more reliable in the use process for diagnosis of complex equipment. The fault diagnosis method based on the method has become a hot spot and trend of current research, so that the fault diagnosis tends to be more intelligent.
Disclosure of Invention
The invention aims to solve the problems of low accuracy and insufficient stability in the prior diagnosis method for gearbox fault diagnosis, in particular to the diagnosis of compound faults, and provides a fusion frame capable of fusing multiple big data methods.
The technical scheme is as follows: the sixth class of example conflict reassignment method (PCR 6) using evidence theory in the fusion process comprises the following main steps:
(1) Inspecting the vibration signal of the gearbox, and extracting the time domain characteristics of the vibration signal of the gearbox through the statistical characteristics;
(2) And (2) inspecting the vibration signal of the gear box, automatically extracting the time-frequency characteristic of the gear box by combining continuous wavelet transformation and a convolutional neural network, and combining the step with the step (1) to construct the Gao Weite symptom of the vibration signal of the gear box.
(3) Inputting the high-dimensional feature set constructed in the step (1) and the step (2) into a sub-model to train the sub-model, performing fault diagnosis on a sample by using the trained sub-model, and outputting respective classification probabilities by the model;
(4) The input of the step is the sample classification probability of a plurality of sub-models in the step (3), and the classification probability output by the model is subjected to equal reliability adjustment;
(5) And the adjusted multi-model output is input into the PCR6 for model fusion, so that a final diagnosis result is obtained.
Preferably, in step (1): and (3) extracting and inspecting the original signals of the vibration sensor by using the time domain features, and constructing a time domain feature set by using a statistical formula. These statistical formulas include both common statistics and features constructed for the timing signal, such as kurtosis and skewness.
Preferably, in step (2): the wavelet transform is a convolution operation of the original signal with a wavelet function, which in effect measures the degree of similarity of the signal to the wavelet function. Thus, by selecting different wavelet basis functions, the content of the component close to the wavelet shape in the signal can be detected, and the method can be used for detecting the characteristic component in the signal. By combining the convolutional neural network, the time-frequency information in the signal can be efficiently extracted through back propagation by introducing sparse interaction and parameter sharing. In the time-frequency characteristic extraction, a vibration time domain signal is sent to a time-frequency analysis by utilizing continuous wavelet transformation to obtain a two-dimensional matrix of time-frequency coefficients, then the two-dimensional matrix is input into a convolutional neural network to obtain the time-frequency characteristic that the output of a full-connection layer is the most vibration of a gear box, and the time-frequency characteristics of different dimensions can be obtained by setting different numbers of neurons of the full-connection layer.
Preferably, in the step (3), the multi-view feature set is input into the submodel for classification, so that a preliminary diagnosis result can be obtained. Suitable sub-models need to have both diversity and stability. A diagnostic model as described in 4 below is commonly used. Support vector machines (Support Vector Machine, SVM), logistic regression (Logistic Regression, LR), deep belief networks (Deep Belief Network, DBN), and gradient boost trees (Gradient Boosting Decision Tree, GBDT), respectively. The models have mature application in classification problems, and model principles and optimization modes are also characterized. In the primary diagnosis of the sub-model, the input of each model is the high-dimensional characteristic of the vibration signal of the gear box (comprising the time domain characteristic extracted in the step (1) and the time-frequency characteristic extracted in the step (2)). After the model is trained, the classification probability of each sample is output, taking 6 classification as an example, and each sample corresponds to the classification probability of 6 classification, such as [0.1112,0.001,0.256,0.050,0.101,0.480]. It should be noted that the model classification probabilities are normalized.
Preferably, in step (4), before the models are fused, the probability result output by each model needs to be adjusted. The PCR6 is a method for fusing multi-source information with equal reliability, and in order to ensure that sub-model results have the same reliability and avoid the phenomenon of one-ticket overruling in the fusion process, probability matching and probability reliability discount are required to be carried out on classification probability of each model before model fusion is carried out. Both steps are performed after normalization of the classification probability. The adjustment mode is as follows:
1) Probability matching the probability distribution derived by the sub-models is different due to differences in model computation mechanisms. Based on the SVM model result, the probability of the model is adjusted. The sample classification probability given by the SVM remains unchanged, and other models are matched by taking the result of the SVM as a reference. For the adjustment model, firstly, calculating the average value of the maximum value of the classification probability of each sample of the model and the SVM model, and taking the ratio of the average value of the maximum probability of the samples of the two models as the difference between the model and the SVM model. In the adjustment, the maximum classification probability in each classification sample probability of the model is firstly adjusted by using the ratio, and then the classification probability of each sample is normalized, so that the adjustment is completed. The specific formula is as follows:
wherein w is * -maximum classification probability of the sample; *
w new -maximum classification probability of the sample after adjustment; w (w) i -other probabilities of the sample than the maximum classification probability, taking 6 classification as an example, the other classification probabilities being 5 classes; w (w) i new -other probabilities of the sample after adjustment than the maximum classification probability; lambda-adjusting the maximum classification probability average value of the model sample; lambda (lambda) SVM -mean value of maximum classification probability of SVM samples.
2) Probability discount in the PCR6 fusion process, the classification probabilities of the sub-classifiers should have the same reliability. For classifiers with different classification accuracy, it is obviously unreasonable to assign the same reliability. Therefore, according to the difference of the accuracy of the verification set, the model classification probability needs to be correspondingly discounted. In the adjustment, according to the accuracy of the verification set of the model, the classification probability of the model needs to be adjusted to the random classification result, and the specific calculation method is as follows:
wherein w is i -classification probabilities of different classes; w (w) i new -adjusting the classification probabilities of the different classes after; n is the number of classification categories, taking 6 classification as an example, n is 6; e (E) error Mould-dieVerification set error rate of type.
Preferably, in the step (5), in the fusion process, weights of output probability values of different models are set, then the probability values of all the different models are multiplied by the weights to sum up to obtain a final result of the fault probability, and the sum of the weights of all the models is 1.
The method focuses on the establishment of a model fusion framework, and utilizes an evidence theory to fuse fault classification of each model, so that the stability and accuracy of diagnosis are improved. In the feature engineering and sub-model diagnosis in the first and second steps, the method and the model are not limited to the above description, and can be adjusted according to the characteristics of the equipment, so that the latest model is introduced into the fusion framework.
In the gear box fault diagnosis method based on model fusion, the characteristics of the fusion framework comprise: gearbox time domain and time frequency domain multi-view features, multi-model preliminary diagnosis and sixth class example conflict reassignment methods.
The beneficial effects are that: compared with the prior art, the invention has the advantages and effects that:
1. the characteristics of the time domain and the time frequency domain of the vibration signal of the gear box are extracted by combining the statistical characteristics and the convolution neural network with the continuous wavelet change, so that the characteristic structure is simplified, and the interference of subjective factors in the characteristic extraction process is reduced.
2. Some common sub-models with large differences between model principles and training modes are selected for preliminary diagnosis of faults, and fault differences in multi-view feature sets are better mastered.
3. The method for performing equal reliability adjustment on the model result is provided, and the theoretical basis of PCR6 fusion is ensured.
4. By adopting the fusion method of PCR6, all training sets can be used in model fusion, and more excellent results are obtained than the common Stacking and voting methods.
Drawings
FIG. 1-a model fusion-based gearbox fault diagnosis flow chart;
FIG. 2-a gearbox time-frequency feature extraction flow chart;
FIG. 3-a time-frequency diagram of successive wavelet transforms for different gearboxes;
FIG. 4-convolutional neural network block diagram;
FIG. 5-test set accuracy.
Detailed Description
FIG. 1 is a fuzzy synthetic decision flow diagram based on validity analysis. The technical solution according to the invention is further elucidated below in connection with an actual gearbox.
1. Acquisition of multiple view characteristics of gearbox vibration signals
A gearbox of Qianpeng company is used for disclosing a data set, the input rotating speed is 880r/min, and the sampling frequency of a vibration signal is 5120Hz. The gearbox conditions include normal, pitting wear, tooth breakage, wear, tooth breakage wear, forming a 6-classification dataset, where pitting wear and tooth breakage wear are compound faults. In order to better reflect the fusion effect, samples of 3 cases were constructed altogether, namely balanced, unbalanced and small sample cases. The ratio of the training sample to the test sample is 2:1, and the sample sampling points are 2000. The number of samples and the partitioning are shown in table 1.
Table 1 sample quantity table
| Sample distribution | Normal state | Pitting corrosion | Point mill | Broken tooth | Breaking mill | Wear and tear |
| Balancing | 106 | 106 | 106 | 106 | 106 | 106 |
| Imbalance of | 100 | 40 | 40 | 40 | 40 | 40 |
| Small sample | 10 | 10 | 10 | 10 | 10 | 10 |
And (3) extracting and inspecting the original signals of the vibration sensor by using the time domain features, and constructing a time domain feature set by using a statistical formula. The time domain characteristics of the vibration signals are visual and concise in state evaluation of the gearbox, low in calculation complexity and easy to be interfered by noise. 63-dimensional features were extracted in the experiment using tsfresh, and the specific sequence feature extraction method can be referred to in table 2.
Table 2 time domain feature name table
The time-frequency domain feature investigation original vibration signal is subjected to wavelet analysis to obtain a wavelet two-dimensional coefficient matrix, and a time-frequency feature set is constructed by utilizing the function of extracting features of the CNN convolution layer. The time frequency of the continuous wavelet transform is shown in figure 2. The structure of CNN used in the experiment is shown in FIG. 3. The whole convolutional neural network model comprises 11 layers, and three convolutional layers are all arranged. Each convolution layer is followed by a normalization layer, a dropout layer and a pooling layer to prevent model overfitting and feature integration. The convolutional pooling layer would then have two fully connected layers, containing 24635, 120 neurons, respectively. Each neuron of the full-connection layer is connected with all nodes of the upper layer, the matrix characteristics after convolution pooling are converted into one-dimensional vector characteristics, and the characteristic dimensions are reduced through nonlinear combination. In the classification task, a convolutional neural network is usually connected with a Softmax layer at the top layer to give a classified label. In the convolution process, the convolution kernel moves in a two-dimensional matrix, and 8@3 ×3 (i.e., a total of 8 convolution kernels of 3×3) are used for three convolution layers, 16@3×3 and 8@3 ×3.
2. Sub-model preliminary diagnosis
Four fault classification models used in the present invention are SVM, LR, DBN and GBDT. And determining key super parameters of the model by adopting a grid parameter optimizing and cross-validation mode. The model specific parameters are shown in table 3. Before the models are fused, the probability results output by the models are required to be adjusted. The PCR6 is a method for fusing multi-source information with equal reliability, and in order to ensure that sub-model results have the same reliability and avoid the phenomenon of one-ticket overruling in the fusion process, probability matching and probability reliability discount are required to be carried out on classification probability of each model before model fusion is carried out. Both steps are performed after normalization of the classification probability.
TABLE 3 preliminary diagnosis of submodel parameters
| Sub-model | Super ginseng 1 | Super ginseng 2 |
| SVM | Penalty term C10 | Kernel function rbf |
| LR | Solver:L-BFGS | Penalty term C:1 |
| DBN | Iteration number: 450 | Activation function: reLU (ReLU) |
| GBDT | Learning rate 10 e-3 | Number of regression trees 200 |
3. Reliability adjustment for model output and the like
Table 4 illustrates the probability adjustment and fusion process using one wear-out failure sample as an example. Under balanced sample conditions, classification accuracy rates of SVM, LR, DBN and GBDT were 94.97%, 94.97%, 94.18% and 95.28%, respectively. Lambda was 85.61%, 91.38%, 97.53% and 95.57%, respectively. As can be seen from table 4, in the case that the classification result of 3 models is wrong, PCR6 gives the correct fault classification according to the classification probability of the model. This shows that in high conflict samples, the fusion strategy used in the present invention gives better results than voting.
TABLE 4 wear sample model fusion process (%)
4. PCR6 model fusion and result analysis
The complete experimental results are shown in table 5. As the sample size decreases, the classification accuracy of each model decreases. Under the condition of sample balance, the accuracy difference of the submodels is smaller and is between 92 and 95 percent. At this time, the fusion model not only realizes the grabbing of the optimal submodel, but also obtains a classification result which is more excellent than the optimal submodel. Under the conditions of unbalanced samples and small samples, the DBN classification accuracy is greatly reduced, and the SVM and GBDT accuracy do not fluctuate to a large extent, so that the accuracy difference between sub-models is overlarge, and the difficulty is brought to model fusion. However, even in unbalanced samples, the fusion model achieves accuracy equivalent to that of the optimal submodel.
Table 5 model test set accuracy (%)
| SVM | LR | DBN | GBDT | Mode number | Stacking | PCR6 | |
| Balancing | 94.97 | 94.97 | 94.18 | 95.28 | 95.13 | 95.44 | 95.44 |
| Imbalance of | 93.00 | 90.33 | 91.00 | 89.00 | 92.33 | 92.67 | 93.00 |
| Small sample | 88.33 | 85.00 | 81.67 | 88.33 | 86.67 | 88.33 | 90.00 |
FIG. 5 (a) compares the test set accuracy of the 4 sub-models with the PCR6 fusion model. The graph shows that the accuracy of the model is basically consistent with the variation trend of the PCR6 result under different sample conditions. Under the conditions of unbalanced samples and small samples, the misclassified samples of each model are large in difference, and the fusion effect is obviously improved. Comparing the accuracy of the test set of the method of the present invention with that of other fusion methods, it can be seen that the fusion method of the present invention has the highest classification accuracy under different sample conditions. For this reason, only the classified label is considered in the method for classifying the labels, and the finer classification probability which can be output by the model is not considered, so that the fusion effect is the worst. The main idea of Stacking is to train a model to learn the predictions using the underlying learner. As can be seen from the results of the graph, the fusion results of the method are better than the stacking results in the case of the three constructed sample sets. Compared with the stacking method, under the fusion framework provided by the invention, each sub-model can completely learn the probability distribution on the training set and the difference of all training samples, and in the stacking method, in order to prevent label leakage, each iteration, the model only learns part of training samples, and the generalization of the obtained fusion result on the test set is reduced.
While the invention has been shown and described with respect to the preferred embodiments, it will be understood by those skilled in the art that various changes and modifications may be made therein without departing from the scope of the invention as defined in the following claims.
Claims (4)
1. A gearbox intelligent diagnosis method based on model fusion is characterized by comprising the following steps:
(1) According to the gearbox vibration signal, extracting the time domain characteristics of the gearbox vibration signal through statistical characteristics;
(2) According to the vibration signal of the gear box, combining continuous wavelet transformation and a convolutional neural network to automatically extract the time-frequency characteristic of the gear box, combining the step with the step (1) to construct Gao Weite collection of the vibration signal of the gear box;
(3) Inputting the high-dimensional feature set constructed in the step (1) and the step (2) into a sub-model to train the sub-model, performing fault diagnosis on a sample by using the trained sub-model, and outputting respective classification probabilities by the model;
(4) The input of the step is the sample classification probability of a plurality of sub-models in the step (3), and the classification probability output by the model is subjected to equal reliability adjustment;
(5) The adjusted multi-model output is input into a PCR6 for model fusion, and a final diagnosis result is obtained;
the classification probability of the model output in the step (4) is subjected to equal reliability adjustment, and the method comprises the following steps:
1) Probability matching, namely adjusting the probability of a model based on the result of an SVM model, wherein the sample classification probability given by the SVM is kept unchanged, other models are matched based on the result of the SVM, for the adjustment model, the average value of the maximum value of each sample classification probability of the model and the SVM model is calculated at first, the ratio of the average value of the maximum probabilities of the samples of the two models is used as the difference between the model and the SVM model, in the adjustment, the maximum classification probability in each classification sample probability of the model is firstly adjusted by using the ratio, and then normalization is carried out on each sample classification probability, so that the adjustment is completed, and the specific formula is as follows:
wherein w is * -maximum classification probability of the sample; w (w) * new -maximum classification probability of the sample after adjustment; w (w) i -other probabilities of the sample than the maximum classification probability; w (w) i new -other probabilities of the sample after adjustment than the maximum classification probability; lambda-adjusting the maximum classification probability average value of the model sample; lambda (lambda) SVM -a mean value of the maximum classification probabilities of the SVM samples;
2) The probability discount, the classification probability of the sub-classifier should have the same reliability, the model classification probability needs to be correspondingly discounted according to the difference of the accuracy of the verification set, in the adjustment, the classification probability of the model needs to be adjusted to the random classification result according to the accuracy of the verification set of the model, and the specific calculation method is as follows:
wherein w is j -classification probabilities of different classes;-adjusting the classification probabilities of the different classes after; n-number of classification categories, E error -verification set error rate of model.
2. The intelligent diagnosis method of the gearbox based on model fusion according to claim 1, wherein in the step (2), in time-frequency feature extraction, vibration time domain signals are sent to a time-frequency analysis by continuous wavelet transformation to obtain a two-dimensional matrix of time-frequency coefficients, then the two-dimensional digital matrix is input into a convolutional neural network to obtain the time-frequency feature that the output of a full-connection layer is the vibration of the gearbox, and different neuron numbers of the full-connection layer are set to obtain the time-frequency features of different dimensions.
3. The intelligent diagnosis method of the gearbox based on model fusion according to claim 2, wherein in the step (3), in the preliminary diagnosis of the models, the input of each model is the high-dimensional characteristic of the vibration signal of the gearbox, the method comprises the time domain characteristic extracted in the step (1) and the time-frequency characteristic extracted in the step (2), and the classification probability of each sample is output after the models are trained.
4. The intelligent diagnosis method for the gearbox based on model fusion according to claim 3, wherein in the step (5), weights of output probability values of different models are set in the fusion process, then the probability values of all the different models are multiplied by the weights to sum so as to obtain a final fault probability result, and the sum of the weights of all the models is 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010966319.5A CN112163474B (en) | 2020-09-15 | 2020-09-15 | Intelligent gearbox diagnosis method based on model fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010966319.5A CN112163474B (en) | 2020-09-15 | 2020-09-15 | Intelligent gearbox diagnosis method based on model fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112163474A CN112163474A (en) | 2021-01-01 |
| CN112163474B true CN112163474B (en) | 2024-03-26 |
Family
ID=73857526
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010966319.5A Active CN112163474B (en) | 2020-09-15 | 2020-09-15 | Intelligent gearbox diagnosis method based on model fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112163474B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112985574B (en) * | 2021-02-26 | 2022-02-01 | 电子科技大学 | High-precision classification identification method for optical fiber distributed acoustic sensing signals based on model fusion |
| CN115371988A (en) * | 2022-10-27 | 2022-11-22 | 北谷电子有限公司 | Engineering machinery fault diagnosis method and system based on multi-feature fusion |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110617960A (en) * | 2019-10-12 | 2019-12-27 | 华北电力大学 | Wind turbine generator gearbox fault diagnosis method and system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109800861A (en) * | 2018-12-28 | 2019-05-24 | 上海联影智能医疗科技有限公司 | A kind of equipment fault recognition methods, device, equipment and computer system |
-
2020
- 2020-09-15 CN CN202010966319.5A patent/CN112163474B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110617960A (en) * | 2019-10-12 | 2019-12-27 | 华北电力大学 | Wind turbine generator gearbox fault diagnosis method and system |
Non-Patent Citations (1)
| Title |
|---|
| 基于CNN-SVM和特征融合的齿轮箱故障诊断;饶雷;唐向红;陆见光;;组合机床与自动化加工技术;20200820(第08期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112163474A (en) | 2021-01-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106168799B (en) | A method of batteries of electric automobile predictive maintenance is carried out based on big data machine learning | |
| CN108334948B (en) | Mechanical bearing fault diagnosis technology based on wide residual error network learning model | |
| CN112418277B (en) | Method, system, medium and equipment for predicting residual life of rotating machine parts | |
| CN112036547B (en) | Rolling bearing residual life prediction method combining automatic feature extraction with LSTM | |
| CN113505655B (en) | Intelligent bearing fault diagnosis method for digital twin system | |
| CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
| CN110929847A (en) | Converter transformer fault diagnosis method based on deep convolutional neural network | |
| Su et al. | Small sample fault diagnosis method for wind turbine gearbox based on optimized generative adversarial networks | |
| CN108256556A (en) | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network | |
| Wu et al. | A transformer-based approach for novel fault detection and fault classification/diagnosis in manufacturing: A rotary system application | |
| de Paula Monteiro et al. | A hybrid prototype selection-based deep learning approach for anomaly detection in industrial machines | |
| CN112163474B (en) | Intelligent gearbox diagnosis method based on model fusion | |
| CN111768000A (en) | Industrial process data modeling method for online adaptive fine-tuning deep learning | |
| CN110907732A (en) | Phase modulator fault diagnosis method based on PCA-RBF neural network | |
| CN112904810B (en) | Process industry nonlinear process monitoring method based on effective feature selection | |
| Lu et al. | Joint training of a predictor network and a generative adversarial network for time series forecasting: A case study of bearing prognostics | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| CN118194136B (en) | Fault diagnosis method for mechanical equipment | |
| Li et al. | Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy | |
| CN112763215A (en) | Multi-working-condition online fault diagnosis method based on modular federal deep learning | |
| US20240184678A1 (en) | Deep Learning Method Integrating Prior Knowledge for Fault Diagnosis | |
| Gao et al. | Design and application of a fault diagnosis and monitoring system for electric vehicle charging equipment based on improved deep belief network | |
| CN110554667A (en) | convolutional Neural Network (CNN) based intermittent industrial process fault diagnosis | |
| Kong et al. | Vibration fault analysis of hydropower units based on extreme learning machine optimized by improved sparrow search algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |