CN116793666A - Wind turbine generator system gearbox fault diagnosis method based on LSTM-MLP-LSGAN model - Google Patents
Wind turbine generator system gearbox fault diagnosis method based on LSTM-MLP-LSGAN model Download PDFInfo
- Publication number
- CN116793666A CN116793666A CN202310441531.3A CN202310441531A CN116793666A CN 116793666 A CN116793666 A CN 116793666A CN 202310441531 A CN202310441531 A CN 202310441531A CN 116793666 A CN116793666 A CN 116793666A
- Authority
- CN
- China
- Prior art keywords
- data
- lsgan
- model
- lstm
- mlp
- 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.)
- Withdrawn
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 66
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000012549 training Methods 0.000 claims abstract description 36
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 25
- 238000012360 testing method Methods 0.000 claims abstract description 24
- 230000008569 process Effects 0.000 claims abstract description 13
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 230000008034 disappearance Effects 0.000 claims abstract description 6
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 19
- 230000002159 abnormal effect Effects 0.000 claims description 17
- 238000010586 diagram Methods 0.000 claims description 14
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000002790 cross-validation Methods 0.000 claims description 9
- 238000007637 random forest analysis Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 6
- 230000036541 health Effects 0.000 claims description 6
- 238000010200 validation analysis Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 2
- 238000013480 data collection Methods 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 4
- 230000007547 defect Effects 0.000 abstract 1
- 238000012423 maintenance Methods 0.000 description 8
- 230000008901 benefit Effects 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 4
- 210000004027 cell Anatomy 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 3
- 238000005299 abrasion Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000003749 cleanliness Effects 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 230000003750 conditioning effect Effects 0.000 description 1
- 230000001808 coupling effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 239000002803 fossil fuel Substances 0.000 description 1
- 239000007789 gas Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000003345 natural gas Substances 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000013450 outlier detection Methods 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000010792 warming Methods 0.000 description 1
Classifications
-
- 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
- G01M13/02—Gearings; Transmission mechanisms
- G01M13/021—Gearings
-
- 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
- G01M13/02—Gearings; Transmission mechanisms
- G01M13/028—Acoustic or vibration analysis
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a wind turbine generator system gearbox fault diagnosis method based on an LSTM-MLP-LSGAN model, which comprises the following steps: firstly, preprocessing data after obtaining vibration signal data of a fan gear box; secondly, generating a main model of an countermeasure network (Least Squares Generative Adversarial Networks, LSGAN) based on least squares, adopting a Long-short-term memory (Long-Short Term Memory, LSTM) neural network as a generator of GAN, using a multi-layer perceptron (Multilayer Perceptron, MLP) as a discriminator of GAN, and establishing a fault diagnosis model of an LSTM-MLP-LSGAN fan gearbox; finally, the data set is divided into a training set and a testing set, the fan gear box fault is diagnosed based on a data-driven diagnosis model, and the effectiveness and the superiority of the method are verified through comparison with other algorithms. The fault diagnosis model provided by the invention can overcome the defect of low accuracy of the conventional fan gear box fault diagnosis technology, can process time sequence data with any length, and can eliminate gradient disappearance and avoid over fitting.
Description
Technical Field
The invention belongs to the technical field of fan detection, and particularly relates to a wind turbine generator gearbox fault diagnosis method based on an LSTM-MLP-LSGAN model.
Background
With the increasing decrease of the world total storage of traditional fossil fuels such as coal, petroleum, natural gas and the like, renewable energy plays an increasingly important role in promoting national economy development and relieving global climate warming. Wind power generation occupies a very important position in all renewable energy sources due to the advantages of cleanliness, reproducibility, short building period, flexible unit scale and the like. Wind turbines generally operate in a harsher environment than steam turbines, gas turbines, used in conventional power plants, resulting in many unpredictable failures of the wind turbines. The fan fault not only increases the maintenance cost of equipment, but also can cause the load reduction of a unit and even unplanned shutdown, so that the grid-connected generated energy is reduced, and the economic benefit of power generation enterprises is further influenced. How to carry out high-efficient state monitoring and accurate fault diagnosis to wind generating set gear box, in time early warning and advance maintenance reduce the fan because of trouble down time, reduce wind power plant's maintenance and cost of maintenance, it is important to improve economic benefits.
In recent years, along with the rapid development of artificial intelligence technology, an artificial intelligence algorithm continuously makes new breakthroughs in various fields, and the artificial intelligence algorithm is applied to solving the fault diagnosis of a wind turbine generator, so that the method is not only suitable for fault diagnosis of a wind turbine generator gearbox in a small sample scene, but also can obtain new data through exploration of the environment, and repeatedly update and iterate the existing model by utilizing the new data, further can improve the fault diagnosis precision of the wind turbine generator and provide accurate fault early warning for the wind turbine generator, can effectively reduce the possibility of further development and expansion of faults, improve the running safety and reliability of fan equipment, and can also arrange and overhaul in advance, and strive for more initiative and fault processing time for field maintenance personnel, thereby reducing maintenance cost, reducing maintenance time and maximizing economic benefits. Therefore, how to ensure the stable operation of the fan, and establish accurate and effective fault diagnosis and emergency mechanism, thereby having important practical significance for maintaining the efficient and stable operation of the fan.
Disclosure of Invention
The traditional fault diagnosis method relies on expert experience knowledge, has low diagnosis accuracy, and provides a wind turbine generator gearbox fault diagnosis method based on an LSTM-MLP-LSGAN model based on an artificial intelligence technology for further improving the fault diagnosis accuracy of a wind turbine gearbox. Firstly, preprocessing data after obtaining vibration signal data of a fan gear box; secondly, generating a main model of an countermeasure network (Least Squares Generative Adversarial Networks, LSGAN) by least square, adopting a Long-short-term memory (Long-Short Term Memory, LSTM) neural network as a generator of GAN, using a multi-layer perceptron (Multilayer Perceptron, MLP) network as a discriminator of GAN, and establishing a fault diagnosis model of an LSTM-MLP-LSGAN fan gearbox; finally, the data set is divided into a training set and a testing set, the fan gear box fault is diagnosed based on a data-driven diagnosis model, and the effectiveness and the superiority of the method are verified through comparison with other algorithms.
The invention adopts the technical scheme that: a wind turbine generator system gearbox fault diagnosis method based on an LSTM-MLP-LSGAN model comprises the following steps:
s1: acquiring a vibration signal data set of a gear box of a wind turbine generator and preprocessing data;
s2: constructing a wind turbine generator gearbox fault diagnosis model based on LSTM-MLP-LSGAN;
s3: and training a diagnosis model by using a training sample set, performing fault diagnosis test by using a test set, and comparing and verifying the effectiveness and superiority of the method by using an algorithm.
Specifically, the step S1: and acquiring an original vibration signal data set of the wind turbine gearbox and performing data preprocessing. Poor data such as missing values and abnormal values, which are possibly caused by the abnormality of the data acquisition equipment or the sensor, exist in the original data set, the data needs to be preprocessed, and then normalization processing is carried out on the data, so that the reduction of prediction accuracy of the deep learning model due to different dimensions is avoided. The part specifically comprises:
(1) And filling missing values in the original data set by adopting a random forest algorithm. For a data set of m rows and n columns (n represents the number of features and m represents the length of a feature sequence), wherein the feature i contains a missing value, other feature data of the row where the missing value is located is taken as prediction input data, the missing value is taken as an object to be predicted, the data of the row where the other n-1 features do not contain the missing value is taken as a training set x_train, the non-missing data in the i is taken as y_train, and then a model is trained and regression prediction is carried out on the missing value;
(2) And detecting abnormal data by using the box line graph and filling by adopting a random forest algorithm. And detecting the abnormal value of the original data by using a box diagram principle, then regarding the abnormal value as a missing value, and filling the missing value by using a random forest algorithm. The case diagram judges the condition of the abnormal value as follows:
wherein x is a Representing outliers; q (Q) 1 、Q 3 Respectively representing the upper quartile and the lower quartile of the box diagram; IQR represents the quarter-bit spacing, i.e., iqr=q 1 -Q 3 ；
(3) And (5) data normalization processing. If the method is directly used for model training, the capability of the model for learning nonlinear characteristics is weakened, and the data needs to be normalized to the [0,1] interval. The data is normalized by adopting maximum and minimum standardization, and the calculation formula is as follows:
wherein: x is x i (k) The original value, x, of the kth sample of feature i i,max 、x i,min Respectively the minimum value and the maximum value in the characteristic i, x i ' (k) is a normalized value.
Specifically, the step S2: and constructing a wind turbine generator gearbox fault diagnosis model based on LSTM-MLP-LSGAN. The diagnostic master model employs an LSGAN, wherein the generator (G) and the discriminator (D) employ an LSTM network and an MLP network, respectively. Original real data X real For vibration signals measured under the states of normal gear, abrasion, tooth breakage and the like of a gear box, category labels and random noise are input into an LSTM network to generate false data X fake X is then added fake And true value X real Respectively input toIn the discriminator MLP network, the output is true when the discriminator identifies the real data, otherwise false is output. In order to prevent the whole model from being over fitted, a Dropout strategy is adopted, and batch standardization is adopted to coordinate the updating of the multi-layer structure parameters in the model, so that the convergence speed is accelerated.
The conventional generation of the countermeasure network has the problem that the gradient disappears in the training process, and if the discriminator is not sufficiently trained, the generator cannot effectively learn the gradient. In order to solve the problem, an LSGAN is adopted to establish an LSTM-MLP-LSGAN combined prediction model, the LSGAN network structure is the same as a GAN network, but the LSGAN gives more punishment to larger errors, and the problem of gradient disappearance in the training process can be solved. The loss function of the traditional GAN network generator and discriminator in the parameter updating process is:
wherein E represents a mathematical expectation of the loss function; g (-) represents the data generated by the generator; d (·) represents the output from which the discriminator discriminates whether the generated data is true or false.
The LSGAN is based on L2 regularization, the discriminator uses alpha-beta codes, the alpha and the beta respectively represent real labels and false labels, in the training process, the discriminator is continuously forced to strengthen the capability of discriminating real data by the aid of game countermeasures between the generator and the discriminator, and the generator is forced to generate predicted data which is more and more close to the real data to deceive the discriminating model, so that Nash equilibrium is finally achieved. The loss function of LSGAN can be written as:
where γ represents the value that the generator wishes the discriminator to trust the dummy data.
Specifically, the step S3: and training a diagnosis model by using a training sample set, performing fault diagnosis test by using a test set, and comparing and verifying the effectiveness and superiority of the method by using an algorithm. Dividing the original data set into a training set and a testing set according to a ratio of 3:1, inputting the model into an original vibration signal of the gear box, and outputting the original vibration signal into the healthy state of the gear box. Comparing the diagnosis model with other algorithms, verifying the effectiveness and superiority of the method in diagnosis accuracy, and then obtaining a confusion matrix by adopting K-fold cross verification, wherein each row of the confusion matrix represents a diagnosis label, and each column represents a real label. The K-fold cross validation divides the original data into K groups, each subset data is respectively used as a validation set, the rest K-1 groups of subset data are used as training sets, K models are obtained, the K models are respectively evaluated in the validation sets, and the final mean square error is added and averaged to obtain the cross validation error.
The invention has the beneficial effects that: the invention provides a wind turbine generator gearbox fault diagnosis method based on an LSTM-MLP-LSGAN model, which aims to solve the problems that the existing fan gearbox fault diagnosis technology depends on expert experience knowledge and has lower diagnosis accuracy. Compared with the traditional prediction method, the provided fault diagnosis model can achieve higher diagnosis accuracy. The method has the advantages that the fan gear box faults are accurately diagnosed, and the method has important significance for improving the safety and reliability of the wind turbine generator.
Drawings
FIG. 1 is a flow chart of a fan fault diagnosis implementation of the present invention;
fig. 2 is a diagram of a conventional GAN network structure according to the present invention;
FIG. 3 is a block diagram of an LSTM cell of the present invention;
FIG. 4 is a schematic diagram of the present invention employing LSTM as an LSGAN network generator;
FIG. 5 is a schematic diagram of the present invention employing MLP as an LSGAN network discriminator;
FIG. 6 is a block diagram of an LSTM-MLP-LSGAN network of the invention;
FIG. 7 is a 5-fold cross-validation schematic of the present invention;
FIG. 8 is a graph showing the diagnostic accuracy of the comparison algorithm of the present invention as a function of iteration number;
FIG. 9 is a diagnostic confusion matrix for the comparison algorithm of the present invention over a test set.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. All other embodiments, which can be made by those skilled in the art without making any inventive effort, based on the embodiments of the present invention are within the scope of the present invention, and are specifically described below in connection with the embodiments.
The invention comprises the following steps:
according to fig. 1, specifically, in step S1, the data is preprocessed after being acquired, including detecting and filling missing values and abnormal values of the data, and then normalizing the data. The part specifically comprises:
(1) And filling missing values in the original data set by adopting a random forest algorithm. For a data set of m rows and n columns (n represents the number of features and m represents the length of a feature sequence), wherein the feature i contains a missing value, other feature data of the row where the missing value is located is taken as prediction input data, the missing value is taken as an object to be predicted, the data of the row where the other n-1 features do not contain the missing value is taken as a training set x_train, the non-missing data in the i is taken as y_train, and then a model is trained and regression prediction is carried out on the missing value;
(2) And detecting abnormal data by using the box line graph and filling by adopting a random forest algorithm. Firstly, carrying out outlier detection on original data by using a box diagram principle, then regarding the outlier as a missing value, and finally filling the missing value by adopting a random forest algorithm. The case diagram judges the condition of the abnormal value as follows:
wherein x is a Representing outliers; q (Q) 1 、Q 3 Respectively representing the upper quartile and the lower quartile of the box diagram; IQR represents the quarter-bit spacing, i.e., iqr=q 1 -Q 3 ；
(3) And (5) data normalization processing. If the method is directly used for model training, the capability of the model for learning nonlinear characteristics is weakened, and the data needs to be normalized to the [0,1] interval. The data is normalized by adopting maximum and minimum standardization, and the calculation formula is as follows:
wherein: x is x i (k) The original value, x, of the kth sample of feature i i,max 、x i,min Respectively the minimum value and the maximum value in the characteristic i, x' i (k) Is a normalized value.
Specifically, the step S2: and constructing a wind turbine generator gearbox fault diagnosis model based on LSTM-MLP-LSGAN. The diagnostic master model employs an LSGAN, wherein the generator (G) and the discriminator (D) employ an LSTM neural network and an MLP network, respectively. FIG. 3 shows an LSTM cell structure in which the internal state c t Specially making linear circulation information transfer, at the same time non-linearly outputting information to external state of hidden layer, x t Input data representing the current time; c t-1 、c t Respectively representing the internal states of the previous moment and the current moment; h is a t-1 、h t Respectively representing the hidden layer states of the previous moment and the current moment; ☉ represents the multiplication of matrix elements,representing matrix element addition; s is a Sigmoid activation function, and tanh is a tanh activation function; z is the cell state; z f Is in a forgetting door gating state and is used for controlling the internal state c at the last moment t-1 The amount of information that needs to be forgotten; z i Representing an input doorThe state control is used for controlling how much information of the current time unit state z needs to be stored; z o Representing the gate control state of the output gate, and controlling the internal state c at the current moment t How much information needs to be output to the external state h t . According to the information flow direction in the graph, the following formula holds:
z＝tanh(W[x t h t-1 ] Τ )
z f ＝s(W f [x t h t-1 ] Τ )
z i ＝s(W i [x t h t-1 ] Τ )
z o ＝s(W o [x t h t-1 ] Τ )
h t ＝z o e tanh(c t )
in W, W f 、W i W is provided o The weight matrices representing the cell state update, forget gate, input gate and output gate, respectively.
According to fig. 4, the raw data X comprises vibration signals measured in four health states of the gear box, namely normal state, pitting state, abrasion state and tooth breakage state, the four states are marked, category label data and random noise are input into an LSTM network, and then h of the LSTM network is input t As input to the fully connected layer and outputAlong with class label data as dummy data X fake 。
The method comprises the following steps:
h t ＝L(X)
wherein L (·) represents the output of LSTM; g (·) represents the output of the GAN generator; delta is a ReLU activation function; w (W) h 、b h The weight matrix and the bias term of the full connection layer are respectively.
According to FIG. 5, y is the labeled data, X fake And true value X real Respectively inputting the data into the discriminator MLP network, outputting true when the discriminator identifies the true data, otherwise outputting false, and classifying the diagnosed health state of the fan gearbox. The output of the discriminator to discriminate the true or false of the data can be expressed as:
True＝σ[m(X fake )]
False＝σ[m(X real )]
where m (·) represents the output of the MLP and σ represents the Sigmoid activation function.
In order to prevent the whole model from being over fitted, a Dropout strategy is adopted, and batch standardization is adopted to coordinate the updating of the multi-layer structure parameters in the model, so that the convergence speed is accelerated.
The problem of gradient disappearance occurs in the training process of the conventional GAN network, and if the discriminator is not sufficiently trained, the generator cannot effectively learn the gradient. To solve this problem, an LSTM-MLP-LSGAN fault diagnosis model is built based on LSGAN, and the whole structure is shown in figure 6. The LSGAN network structure is the same as the GAN network, but LSGAN gives more punishment to larger errors and can solve the problem of gradient disappearance in the training process. The loss function of the traditional GAN network generator and discriminator in the parameter updating process is:
wherein E represents a mathematical expectation of the loss function; g (-) represents the data generated by the generator; d (·) represents the output from which the discriminator discriminates whether the generated data is true or false.
The LSGAN is based on L2 regularization, the discriminator uses alpha-beta codes, the alpha and the beta respectively represent real labels and false labels, in the training process, the discriminator is continuously forced to strengthen the capability of discriminating real data by the aid of game countermeasures between the generator and the discriminator, and the generator is forced to generate predicted data which is more and more close to the real data to deceive the discriminating model, so that Nash equilibrium is finally achieved. The loss function of LSGAN can be written as:
where γ represents the value that the generator wishes the discriminator to trust the dummy data.
Specifically, the step S3: and training a diagnosis model by using a training sample set, performing fault diagnosis test by using a test set, and comparing and verifying the effectiveness and superiority of the method by using an algorithm. Dividing the original data set into a training set and a testing set according to a ratio of 3:1, inputting the model into an original vibration signal of the gear box, and outputting the original vibration signal into the healthy state of the gear box. The diagnostic model is compared with other algorithms to verify the effectiveness and superiority of the methods presented herein in terms of diagnostic accuracy. The invention adopts 5-fold cross validation to obtain the confusion matrix to observe the diagnosis condition of the algorithm on the test set, each row in the confusion matrix represents the diagnosis label, and each column represents the real label. According to fig. 7, the adopted 5-fold cross validation divides the original data into 5 groups, each subset data is respectively processed into a validation set, the rest 4 groups of subset data are processed as training sets, so that 5 models are obtained, the 5 models evaluate results in the validation sets respectively, and the final mean square error sum and the average are processed to obtain the cross validation error.
The validity of the present invention is verified as follows:
according to fan gear box fault record of new energy company for many years, the fault of gear box is occurred at high-speed shaft, planetary gear and middle-speed shaft, after three types of faults are occurred, vibration signal of each frequency band can be changed differently, and it is difficult to extract fault characteristics due to complex working condition and coupling effect, so that a high-accuracy diagnosis method is required. The economic losses caused by the faults of the high-speed shaft and the medium-speed shaft are relatively small, and the maintenance difficulty of the faults of the planetary gears is high, so that more manpower and material resources are needed. The invention adopts the LSTM-LSGAN model to diagnose the faults of the planetary gears, the adopted data set is derived from actual recorded data of vibration signals of a fan gear box of a certain company, the rotating speed of the fan is 1700-1800 rpm, and the rated power is 1500kW. The data were marked for normal, pitting, wear, tooth breakage, and each label contained 3560 pieces of data for a total of 14240 pieces of data. According to the diagnosis flow, the data set is preprocessed, and the data set is selected to have no missing value and abnormal data, so that only the data is normalized, and the data preprocessing is essential for practical engineering application. On this basis, the data set is divided into a training set and a test set according to a ratio of 3:1, wherein the training set contains 10680 pieces of data, and the test set contains 3560 pieces of data.
The training is input into a diagnosis model, the model is tested after model training, in order to verify the effectiveness and superiority of the method, the method is compared with LSTM and Artificial Neural Network (ANN) diagnosis accuracy, an Adam optimizer is selected to update LSTM network parameters, a mean square error is adopted as a loss function, the iteration times of the parameters are set to be 200, the learning interest rate is 0.001, the hidden layer number is 2, and the number of hidden layer neurons is 50 and 100 respectively through an artificial experience method; ANN adopts Adam and MSE as an optimizer and a loss function respectively, and the learning rate is 0.001; MLP adopts Adam and MSE as an optimizer and a loss function respectively, regularization parameter alpha=0.0001, and exponential decay rate beta of first-order moment vector 1 =0.9, the exponential decay rate β of the second moment vector 2 As can be seen from the graph, the three methods all converge after 150 iterations, wherein LSTM-MLP-LSGAN begins to converge after 100 iterations with an accuracy of 98.45%, LSTM begins to converge after 125 iterations with an accuracy of 97.17%, ANN begins to converge after 150 iterations with an accuracy of 94.91%, and LSTM-MLP-LSGAN has a higher diagnostic accuracy and a faster convergence rate than LSTM and ANN by comparisonFault diagnosis performance is improved.
And then, a confusion matrix for three algorithm diagnoses is obtained through five-fold cross validation, as shown in fig. 9, wherein (a) is an ANN fault diagnosis result, (b) is an LSTM fault diagnosis result, and (c) is an LSTM-MLP-LSGAN fault diagnosis result, and labels 1, 2, 3 and 4 in the figure represent four states of normal planetary gear, pitting, abrasion and tooth breakage respectively. The diagnostic accuracy of the proposed method for four health states of the planetary gear on the test set is 99.63%, 98.43%, 98.06%, 97.67%, the fault diagnosis is carried out by adopting LSTM, the accuracy on the test set is 99.41%, 96.77%, 96.63%, 95.84%, the fault diagnosis is carried out by adopting ANN, and the diagnostic accuracy for four states on the test set is 97.89%, 94.94%, 94.83%, 92.02%. By contrast, the diagnostic accuracy of the method on the test set is highest, compared with an LSTM network, the diagnostic accuracy of the four health states is respectively 0.22%, 1.66%, 1.43% and 1.83%, compared with an ANN network, the diagnostic accuracy of the four health states is respectively 1.74%, 3.49%, 3.23% and 5.65%, and the effectiveness of the method is verified, and the method has higher fault diagnosis performance.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Claims (7)
1. A wind turbine generator system gearbox fault diagnosis method based on an LSTM-MLP-LSGAN model is characterized by comprising the following steps:
s1: acquiring a vibration signal data set of a gear box of a wind turbine generator and preprocessing data;
s2: constructing a wind turbine generator fault diagnosis model based on LSTM-MLP-LSGAN;
s3: and training a diagnosis model by using a training sample set, performing fault diagnosis test by using a test set, and comparing and verifying the effectiveness and superiority of the method by using an algorithm.
2. The method for diagnosing faults of a wind turbine gearbox based on an LSTM-MLP-LSGAN model according to claim 1, wherein in step S1, the missing value and the abnormal value are caused to exist in original data by considering poor operation or abnormal equipment in the experimental test and the data collection process, the missing value and the abnormal value are required to be detected and filled in the original data set, and then the data is normalized, and the method comprises the following specific implementation steps:
(1) Filling missing values in the original data set by adopting a random forest algorithm; for a data set of m rows and n columns (n represents the number of features and m represents the length of a feature sequence), wherein the feature i contains a missing value, other feature data of the row where the missing value is located is taken as prediction input data, the missing value is taken as an object to be predicted, the data of the row where the other n-1 features do not contain the missing value is taken as a training set x_train, the non-missing data in the i is taken as y_train, and then a model is trained and regression prediction is carried out on the missing value;
(2) Detecting abnormal data by using a box line graph and filling the abnormal data by adopting a random forest algorithm; detecting an abnormal value of the original data by utilizing a box diagram principle, then regarding the abnormal value as a missing value, and filling the missing value by adopting a random forest algorithm; the case diagram judges the condition of the abnormal value as follows:
wherein x is a Representing outliers; q (Q) 1 、Q 3 Respectively representing the upper quartile and the lower quartile of the box diagram; IQR represents the quarter-bit spacing, i.e., iqr=q 1 -Q 3 ；
(3) Carrying out data normalization; if the method is directly used for model training, the capability of the model for learning nonlinear characteristics is weakened, and the data needs to be normalized to a [0,1] interval; the data is normalized by adopting maximum and minimum standardization, and the calculation formula is as follows:
wherein: x is x i (k) The original value, x, of the kth sample of feature i i,max 、x i,min Respectively the minimum value and the maximum value in the characteristic i, x i ' (k) is a normalized value.
3. The wind turbine gearbox fault diagnosis method based on the LSTM-MLP-LSGAN model according to claim 1 is characterized in that in the step S2, a wind turbine planetary gearbox fault diagnosis model based on the LSTM-MLP-LSGAN is constructed;
the main model of the fault diagnosis adopts a least squares generation countermeasure network ((Least Squares Generative Adversarial Networks, LSGAN)) structure, wherein the generator (G) and the discriminator (D) respectively adopt a Long-short-term memory (Long-Short Term Memory, LSTM) neural network and a multi-layer perceptron (Multilayer Perceptron, MLP) network.
4. A wind turbine gearbox fault diagnosis method based on LSTM-MLP-LSGAN model according to claims 1 and 3, characterized in that in step S2, further, the raw data X includes vibration signals measured in the healthy state of the gearbox gear in normal, pitting, wearing, tooth breakage 4, marking these four states, inputting category label data and random noise into LSTM network, then inputting h of LSTM network t As input to the fully connected layer and outputAlong with class label data as dummy data X fake ；
The method comprises the following steps:
h t ＝L(X)
wherein L (·) represents the output of LSTM; g (·) represents the output of the GAN generator; delta is a ReLU activation function; w (W) h 、b h The weight matrix and the bias term of the full connection layer are respectively.
5. The method for diagnosing faults of a wind turbine gearbox based on an LSTM-MLP-LSGAN model according to claim 1, 3 and 4, wherein in step S2, further, X is set to be fake And true value X real Respectively inputting the real data into an MLP network of the discriminator, outputting true when the discriminator recognizes the real data, otherwise outputting false, and classifying the diagnosed health state of the fan gear box; the output of the discriminator to discriminate the true or false of the data can be expressed as:
True＝σ[m(X fake )]
False＝σ[m(X real )]
where m (·) represents the output of the MLP and σ represents the Sigmoid activation function;
in order to prevent the whole model from being over fitted, a Dropout strategy is adopted, and batch standardization is adopted to coordinate the updating of the multi-layer structure parameters in the model, so that the convergence speed is accelerated.
6. The method for diagnosing faults of a wind turbine gearbox based on an LSTM-MLP-LSGAN model according to claims 1, 3, 4 and 5, wherein in step S2, further, the problem of gradient disappearance occurs in the conventional GAN network during training, and if the discriminator fails to be sufficiently trained, the generator cannot learn the gradient effectively; to solve this problem, an LSTM-MLP-LSGAN fault diagnosis model is built based on LSGAN; the LSGAN network structure is the same as the GAN network, but LSGAN gives more punishment to larger errors and can solve the problem of gradient disappearance in the training process; the loss function of the traditional GAN network generator and discriminator in the parameter updating process is:
wherein E represents a mathematical expectation of the loss function; g (-) represents the data generated by the generator; d (·) represents the output of the discriminator discriminating whether the generated data is true or false;
the LSGAN is based on L2 regularization, the discriminator uses alpha-beta codes, wherein alpha and beta respectively represent real and false labels, in the training process, the discriminator is continuously forced to strengthen the capability of discriminating real data by the aid of game countermeasures between the generator and the discriminator, and the generator is forced to generate predicted data which is more and more close to the real data to deceive the discriminating model, so that Nash equilibrium is finally achieved; the loss function of LSGAN can be written as:
where γ represents the value that the generator wishes the discriminator to trust the dummy data.
7. The method for diagnosing faults of the wind turbine generator gearbox based on the LSTM-MLP-LSGAN model according to claim 1 is characterized in that in the step S3, a training sample set is utilized to train a diagnosis model, a test set is utilized to carry out fault diagnosis test, and the effectiveness and the superiority of the method are verified through algorithm comparison; dividing an original data set into a training set and a testing set according to a ratio of 3:1, inputting the model into an original vibration signal of the gear box, and outputting the original vibration signal into a healthy state of the gear box; comparing the diagnosis model with other algorithms, verifying the effectiveness and superiority of the method in diagnosis accuracy, and then adopting K-fold cross verification to obtain a confusion matrix, wherein each row of the confusion matrix represents a diagnosis label, and each column represents a real label; the K-fold cross validation divides the original data into K groups, each subset data is respectively used as a validation set, the rest K-1 groups of subset data are used as training sets, K models are obtained, the K models are respectively evaluated in the validation sets, and the final mean square error is added and averaged to obtain the cross validation error.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310441531.3A CN116793666A (en) | 2023-04-23 | 2023-04-23 | Wind turbine generator system gearbox fault diagnosis method based on LSTM-MLP-LSGAN model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310441531.3A CN116793666A (en) | 2023-04-23 | 2023-04-23 | Wind turbine generator system gearbox fault diagnosis method based on LSTM-MLP-LSGAN model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116793666A true CN116793666A (en) | 2023-09-22 |
Family
ID=88045375
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310441531.3A Withdrawn CN116793666A (en) | 2023-04-23 | 2023-04-23 | Wind turbine generator system gearbox fault diagnosis method based on LSTM-MLP-LSGAN model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116793666A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117113234A (en) * | 2023-10-12 | 2023-11-24 | 济南泉晓电气设备有限公司 | Power transmission line fault detection method and system based on machine learning |
-
2023
- 2023-04-23 CN CN202310441531.3A patent/CN116793666A/en not_active Withdrawn
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117113234A (en) * | 2023-10-12 | 2023-11-24 | 济南泉晓电气设备有限公司 | Power transmission line fault detection method and system based on machine learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Anomaly detection and critical SCADA parameters identification for wind turbines based on LSTM-AE neural network | |
| CN111237134B (en) | Offshore double-fed wind driven generator fault diagnosis method based on GRA-LSTM-stacking model | |
| Zhao et al. | Intelligent fault diagnosis of multichannel motor–rotor system based on multimanifold deep extreme learning machine | |
| Dai et al. | Machinery health monitoring based on unsupervised feature learning via generative adversarial networks | |
| Udo et al. | Data-driven predictive maintenance of wind turbine based on SCADA data | |
| CN107657250B (en) | Bearing fault detection and positioning method and detection and positioning model implementation system and method | |
| CN112033463B (en) | Nuclear power equipment state evaluation and prediction integrated method and system | |
| CN115758290A (en) | Fan gearbox high-speed shaft temperature trend early warning method based on LSTM | |
| CN111103137A (en) | Wind turbine gearbox fault diagnosis method based on deep neural network | |
| CN116793666A (en) | Wind turbine generator system gearbox fault diagnosis method based on LSTM-MLP-LSGAN model | |
| CN112731890A (en) | Power plant equipment fault detection method and device | |
| Zhang et al. | Fault diagnosis with bidirectional guided convolutional neural networks under noisy labels | |
| Zhao et al. | A capsnet-based fault diagnosis method for a digital twin of a wind turbine gearbox | |
| CN116226679A (en) | Wind turbine generator gearbox abnormality detection method considering similarity of running states of multiple units | |
| CN116821730A (en) | Fan fault detection method, control device and storage medium | |
| CN114897292B (en) | Fault diagnosis method and system based on multi-scale residual convolution variational network | |
| CN116521406A (en) | Method for detecting anomaly of non-overrun flight parameter data of aero-engine based on residual gate GRU-VAE model | |
| Sohaidan et al. | Remaining useful life prediction of turbofan engine using long-short term memory | |
| Sun et al. | Adversarial representation learning for intelligent condition monitoring of complex machinery | |
| Souza et al. | Evaluation of data based normal behavior models for fault detection in wind turbines | |
| CN115560962A (en) | Fan abnormal state detection method based on multi-level space-time diagram neural network | |
| Song et al. | Anomaly detection of wind turbine generator based on temporal information | |
| Zhang et al. | A flexible monitoring framework via dynamic-multilayer graph convolution network | |
| CN114046816A (en) | Sensor signal fault diagnosis method based on lightweight gradient lifting decision tree | |
| Yang et al. | Machinery fault detection using autoencoder and online sequential extreme learning machine |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20230922 |
|
| WW01 | Invention patent application withdrawn after publication |