CN111855202A - Gear box fault diagnosis method and system - Google Patents

Gear box fault diagnosis method and system Download PDF

Info

Publication number
CN111855202A
CN111855202A CN202010805158.1A CN202010805158A CN111855202A CN 111855202 A CN111855202 A CN 111855202A CN 202010805158 A CN202010805158 A CN 202010805158A CN 111855202 A CN111855202 A CN 111855202A
Authority
CN
China
Prior art keywords
signal
fault diagnosis
horizontal vibration
signals
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.)
Pending
Application number
CN202010805158.1A
Other languages
Chinese (zh)
Inventor
田国栋
左颖
乔文生
靳志军
陈志刚
康涛
王铁铮
赵赤兵
高鹤
丁斐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Aero Top Hi Tech Co ltd
Beihang University
Original Assignee
Beijing Aero Top Hi Tech Co ltd
Beihang University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing Aero Top Hi Tech Co ltd, Beihang University filed Critical Beijing Aero Top Hi Tech Co ltd
Priority to CN202010805158.1A priority Critical patent/CN111855202A/en
Publication of CN111855202A publication Critical patent/CN111855202A/en
Pending legal-status Critical Current

Links

Images

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
    • G01M13/028—Acoustic or vibration analysis
    • 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

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a method and a system for diagnosing faults of a gearbox, wherein the method comprises the steps of acquiring horizontal vibration signals and vertical vibration signals of a gearbox shaft through an acceleration sensor arranged on a wind driven generator, extracting characteristic data of the horizontal vibration signals and the vertical vibration signals of the gearbox shaft, inputting the extracted characteristic data into a deep convolutional neural network module to obtain a fault diagnosis result, wherein the deep convolutional neural network module is a fault diagnosis model which is learned and established in advance according to the characteristic data of the horizontal vibration signals and the vertical vibration signals of the gearbox shaft with different faults, and the characteristic data extraction of the horizontal vibration signals and the vertical vibration signals of the gearbox shaft comprises the following steps: and performing multi-scale time domain feature extraction and deep noise reduction feature extraction. The invention improves the accuracy and automation level of the fault diagnosis of the wind driven generator gear box, can monitor the working condition of the wind driven generator gear box in real time and reduces the labor intensity of personnel.

Description

Gear box fault diagnosis method and system
Technical Field
The invention relates to a gearbox fault diagnosis method and system, in particular to a gearbox fault diagnosis method and system with a multi-feature fusion convolutional neural network.
Background
The gearbox plays an important role in realizing functions of transmission, speed change, speed regulation and the like in the wind driven generator, and particularly along with the development of wind power technology, in order to better realize the conversion from wind energy to mechanical energy and then to electric energy, the gearbox is also developed towards the direction of complication and precision, so that the vibration signal generated at a fault point is complex, and the difficulty of fault diagnosis is increased. The traditional fault judgment method generally comprises the steps of firstly extracting features, for example, analyzing the time-frequency features of vibration signals by a time-frequency analysis method, extracting the features by Fourier transform, wavelet transform and other methods, and then classifying and diagnosing faults by an expert system, a support vector machine and other models.
Disclosure of Invention
The invention aims to provide a gearbox fault diagnosis method and system, and particularly relates to a gearbox fault diagnosis method and system with a multi-feature fusion convolutional neural network.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a gearbox fault diagnosis method is used for wind driven generator gearbox fault diagnosis, a gearbox shaft horizontal vibration signal and a gearbox shaft vertical vibration signal are obtained through an acceleration sensor arranged on a wind driven generator, gearbox shaft horizontal vibration signal and gearbox shaft vertical vibration signal feature data are extracted, the extracted feature data are input into a deep convolution neural network module to obtain a fault diagnosis result, the deep convolution neural network module is a fault diagnosis model which is learned and established in advance according to gearbox shaft horizontal and vertical vibration signal feature data of different faults of a gearbox, and the extraction of the feature data of the gearbox shaft horizontal vibration signal and the gearbox shaft vertical vibration signal comprises the following steps: performing multi-scale time domain feature extraction and deep noise reduction feature extraction:
the multi-scale time domain feature extraction step comprises:
firstly, decomposing a vibration signal into stable signal IMF components according to a set time length, obtaining the energy of each IMF component through a formula (1),
Figure BDA0002628862990000021
wherein: n denotes that the signal is divided into N equal parts, each equal part having a time length of Deltat, ci(tn) represents the amplitude of the ith component at the time tn, and h is the total number of IMF components;
secondly, selecting the first S IMF components with energy sorted from big to small, calculating the kurtosis value index of the S IMF components through a formula (2),
Figure BDA0002628862990000022
thirdly, taking the first P IMF components with the kurtosis value in a descending order
Figure BDA0002628862990000023
As a result of multi-scale time-domain feature extraction, where Nc is the length of the signal,
Figure BDA0002628862990000024
1xNc indicating that the IMF component is a real number dimension, P being less than S;
the deep noise reduction feature extraction comprises the following steps: and sending the vibration signal into a depth noise reduction self-encoder, and outputting a depth noise reduction characteristic through the depth noise reduction self-encoder.
The scheme is further as follows: the horizontal vibration signals include axial horizontal vibration signals and radial horizontal vibration signals.
The scheme is further as follows: the fault diagnosis model establishment steps are as follows:
firstly, respectively extracting the multi-scale time domain features and the deep noise reduction features according to different faults;
and secondly, fusing the multi-scale time domain features and the depth noise reduction features: splicing each IMF component in the multi-scale time domain features and the deep noise reduction features, stacking the IMF components into a multi-channel training sample according to a time sequence order, constructing and training a deep convolutional neural network, and completing the training of the convolutional neural network through the input of the training sample;
wherein each layer of convolution layer outputs a feature map of formula (3) to complete feature fusion:
Figure BDA0002628862990000031
wherein denotes a convolution, M i1, 2., P +1 denotes the convolution kernel of P +1 channels;
and thirdly, forming the probability of each fault type according to the feature fusion and outputting a fault diagnosis result.
The scheme is further as follows: the feature data extraction is to extract feature data of a plurality of sample signals of the vibration signal, wherein the sample signals are obtained by dividing the length of the vibration signal by a window with equal length.
The scheme is further as follows: the method further comprises the steps of marking the device name, the number of sampling points, the sensitivity and the acquisition time parameter in the acquisition process of each group of the multiple sample signals, and checking whether corresponding fields of each group of the multiple sample signals have empty values or not and whether the fields are complete or not.
The scheme is further as follows: the method for decomposing the vibration signal into the stable signal IMF component according to the set time length comprises the following steps:
firstly, defining the total times MAX _ ITER of EMD to be subjected to empirical mode decomposition;
secondly, white noise in normal distribution is added into the original signal, and the signal to be processed is decomposed through EMD to obtain IMF component Cij，CijRepresenting i IMF components decomposed in the j-th time, and repeating the operation until the number of empirical mode decomposition times reaches MAX _ ITER;
thirdly, calculating IMF components corresponding to each time point and calculating the average value according to a formula (4),
Figure BDA0002628862990000032
finally spliced according to time sequence into
Figure BDA0002628862990000033
As the final found stationary signal IMF component.
A gearbox fault diagnosis system implementing the method, comprising: the wind driven generator is characterized in that a bearing seat, a gear transmission box and a generator shell, which are butted with the wind driven rotating blade hub, of an input shaft are respectively provided with an acceleration sensor for sensing horizontal vibration signals and vertical vibration signals of a gear shaft, the acceleration sensor is connected with a fault diagnosis and analysis computer through a wire or a wireless way, a signal receiving interface card is arranged in the computer, the signal receiving interface card is connected with the acceleration sensor to send the received horizontal vibration signals and the received vertical vibration signals of the gear to a central processing unit, and a vibration signal preprocessing module, a characteristic data extraction module and a deep convolution neural network module are arranged in the central processing unit;
the vibration signal preprocessing module divides the lengths of the acquired horizontal vibration signals and the acquired vertical vibration signals of the gear into a plurality of sample signals by using windows with equal lengths, records the equipment name, the number of sampling points, the sensitivity and the acquisition time parameters of the corresponding vibration signals, and checks whether corresponding fields of each group of the plurality of sample signals have empty values or not and whether the fields are complete or not;
the characteristic data extraction module is used for extracting the characteristics of the sample signal provided by the preprocessing module;
and the deep convolutional neural network module is used for carrying out fault diagnosis on the characteristic data extracted by the characteristic data extraction module and outputting a diagnosis result.
The scheme is further as follows: the acceleration sensor includes: the wind driven generator comprises an input shaft axial horizontal vibration sensor arranged on the axial end face of a bearing seat, wherein the input shaft axial horizontal vibration sensor is butted with a hub of a wind driven rotating blade, an input shaft radial horizontal vibration sensor arranged on the horizontal radial end face of the bearing seat, a box body vertical vibration sensor and a box body radial horizontal vibration sensor arranged on a gear transmission box body, and a generator shaft radial horizontal vibration sensor respectively arranged on an in-out output shaft of a generator.
The scheme is further as follows: the box vertical vibration sensor is arranged in the middle of the upper side face of the box, the box radial horizontal vibration sensors are two, and the two box radial horizontal vibration sensors are arranged on two sides of the box vertical vibration sensor on the side face of the box respectively.
The scheme is further as follows: the acceleration sensor is fixed at a set position in an embedding mode.
The invention has the beneficial effects that:
the system can diagnose the fault of the wind driven generator gear box, improves the accuracy and automation level of fault diagnosis of the wind driven generator gear box, can monitor the working condition of the wind driven generator gear box in real time, and reduces the labor intensity of personnel.
The method extracts the fault characteristics in the vibration signal by different methods, decomposes the original signal and extracts IMF components with obvious fault characteristic information by adopting a method of ensemble empirical mode decomposition, and extracts the deep noise reduction characteristics in the original signal by utilizing a deep noise reduction self-encoder, thereby making up the defect of the characteristic extraction capability of a single method;
the method adopts the deep convolutional neural network to carry out feature fusion and fault diagnosis, because each component after feature extraction is difficult to uniformly model, the deep convolutional neural network can excavate the correlation relationship therein, establish the complex mapping with the fault type, update the model parameters through training, improve the accuracy and robustness of the model, and improve the accuracy of the fault diagnosis of the wind driven generator to a certain extent.
Drawings
FIG. 1 is a schematic diagram of the distribution of acceleration sensors of the system of the present invention;
FIG. 2 is a block diagram of a system of the present invention;
FIG. 3 is a flow chart of multi-scale feature extraction according to the present invention;
FIG. 4 is a multi-scale time domain feature extraction module outputting 5 IMF components;
FIG. 5 is a fused feature first layer convolution.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments, and all other embodiments obtained by a person skilled in the art based on the embodiments of the present invention belong to the protection scope of the present invention without creative efforts.
A technical scheme of a gearbox fault diagnosis method is used for wind turbine generator gearbox fault diagnosis, and the method is realized based on a gearbox fault diagnosis system, as shown in figures 1 and 2, the fault diagnosis system comprises: a gear transmission case 1 of the wind driven generator, an input shaft 101 of the gear transmission case 1 is connected with a wind power rotating blade hub 2, an output shaft 102 of the gear transmission case is connected with a generator 3, wherein, a bearing seat 8 of which the input shaft is butted with the hub of the wind power rotating blade, the gear transmission case 1 and the shell of the generator 3 are respectively provided with an acceleration sensor for sensing a horizontal vibration signal and a vertical vibration signal of a gear shaft, the acceleration sensor is connected with a fault diagnosis and analysis computer (not shown in the figure) through a wire or a wireless way, a signal receiving interface card is arranged in the computer, the signal receiving interface card is connected with an acceleration sensor to send the received gear horizontal vibration signal and the received gear vertical vibration signal to a central processing unit, as shown in figure 2, a vibration data signal acquisition and preprocessing module 5, a characteristic data extraction module 6 and a deep convolutional neural network module 7, namely a multi-characteristic fusion convolutional neural network module, are arranged in the central processing unit;
the vibration data signal acquisition and preprocessing module acquires vibration signals, divides the acquired lengths of the horizontal vibration signals and the vertical vibration signals of the gear shaft into a plurality of sample signals by using windows with equal length, records the equipment name, the number of sampling points, the sensitivity and the acquisition time parameters of the corresponding vibration signals, and checks whether corresponding fields of each group of the plurality of sample signals have empty values or not and whether the fields are complete or not;
the characteristic data extraction module performs multi-scale time domain IMF component characteristic extraction and deep noise reduction characteristic extraction on the sample signal provided by the preprocessing module;
and the deep convolutional neural network module is used for carrying out fault diagnosis on the characteristic data extracted by the characteristic data extraction module and outputting a diagnosis result.
The acceleration sensor includes: an input shaft axial horizontal vibration sensor 9 arranged on the axial end face of a bearing seat 8 of which the input shaft is butted with a hub of the wind power rotating blade, an input shaft radial horizontal vibration sensor 10 arranged on the horizontal radial end face of the bearing seat, a box body vertical vibration sensor 11 and a box body radial horizontal vibration sensor 12 arranged on a gear transmission box body, and a generator shaft radial horizontal vibration sensor 13 respectively arranged on an input shaft and an output shaft of a generator.
Wherein: the input shaft axial horizontal vibration sensor 9 that 8 axial end faces of bearing frame set up and the radial horizontal vibration sensor 10 of input shaft that the horizontal radial terminal surface of bearing frame set up differ the setting of 90 degrees angles, box vertical vibration sensor 11 sets up in the middle of the side on the box, box radial horizontal vibration sensor 12 has two, and two box radial horizontal vibration sensor sides do not set up the box vertical vibration sensor both sides in the box side. In order to fix firmly, the acceleration sensor is fixed on a set position in an embedding mode, namely the acceleration sensor is embedded and fixed by machining a groove.
According to the method, a horizontal vibration signal and a vertical vibration signal of a gearbox shaft are obtained through an acceleration sensor arranged on a wind driven generator, characteristic data of the horizontal vibration signal and the vertical vibration signal of the gearbox shaft are extracted, the extracted characteristic data are input into a deep convolutional neural network module to obtain a fault diagnosis result, the deep convolutional neural network module is a fault diagnosis model which is learned and established in advance according to the characteristic data of the horizontal vibration signal and the vertical vibration signal of the gearbox shaft with different faults, and the extraction of the characteristic data of the horizontal vibration signal and the vertical vibration signal of the gearbox shaft comprises the following steps: performing multi-scale time domain feature extraction and deep noise reduction feature extraction:
the multi-scale time domain feature extraction step comprises:
firstly, decomposing a vibration signal into stable signal IMF components according to a set time length, obtaining the energy of each IMF component through a formula (1),
Figure BDA0002628862990000071
wherein: n denotes that the signal is divided into N equal parts, each equal part having a time length of Deltat, ci(tn) represents the amplitude of the ith component at the time tn, and h is the total number of IMF components;
secondly, selecting the first S IMF components with energy sorted from big to small, calculating the kurtosis value index of the S IMF components through a formula (2), wherein the S value is less than 10, generally 6 to 8,
Figure BDA0002628862990000072
thirdly, taking the first P IMF components with the kurtosis value in a descending order
Figure BDA0002628862990000073
As a result of multi-scale time-domain feature extraction, where Nc is the length of the signal,
Figure BDA0002628862990000074
representing that the IMF component is 1xNc of real dimensionality, P is smaller than S, usually P is smaller than 2 to 3 of the S value, for example, S value is 8, and P value is 6 or 5;
the deep noise reduction feature extraction comprises the following steps: sending the vibration signal into a depth noise reduction self-encoder, and outputting the depth noise reduction characteristic through the depth noise reduction self-encoder
Figure BDA0002628862990000075
The horizontal vibration signals include axial horizontal vibration signals and radial horizontal vibration signals.
Wherein: the fault diagnosis model establishment steps are as follows:
firstly, respectively extracting the multi-scale time domain features and the deep noise reduction features according to different faults;
and secondly, fusing the multi-scale time domain features and the depth noise reduction features: splicing each IMF component in the multi-scale time domain features and the deep noise reduction features, stacking the IMF components into a multi-channel training sample according to a time sequence order, constructing and training a deep convolutional neural network, and completing the training of the convolutional neural network through the input of the training sample;
wherein each layer of convolution layer outputs a feature map of formula (3) to complete feature fusion:
Figure BDA0002628862990000076
wherein denotes a convolution, M i1, 2., P +1 denotes the convolution kernel of P +1 channels;
and thirdly, outputting a fault diagnosis result according to the probability of each fault type formed by the feature fusion.
Wherein: the feature data extraction is to extract feature data of a plurality of sample signals of the vibration signal, wherein the sample signals are obtained by dividing the length of the vibration signal by a window with equal length.
The method further comprises the steps of marking the device name, the number of sampling points, the sensitivity and the acquisition time parameter in the acquisition process of each group of the multiple sample signals, and checking whether corresponding fields of each group of the multiple sample signals have empty values or not and whether the fields are complete or not.
Wherein: the method for decomposing the vibration signal into the stable signal IMF component according to the set time length comprises the following steps:
firstly, defining the total times MAX _ ITER of EMD to be subjected to empirical mode decomposition;
secondly, white noise in normal distribution is added into the original signal, and the signal to be processed is decomposed through EMD to obtain IMF component Cij，CijRepresenting i IMF components decomposed in the j-th time, and repeating the operation until the number of empirical mode decomposition times reaches MAX _ ITER;
thirdly, calculating IMF components corresponding to each time point and calculating the average value according to a formula (4),
Figure BDA0002628862990000081
finally spliced according to time sequence into
Figure BDA0002628862990000082
As the final found stationary signal IMF component.
The following is a further detailed description of the performance of the above-described method in conjunction with fig. 2-5.
In fig. 2, the data acquisition and preprocessing module is responsible for data acquisition and data preprocessing, and is specifically implemented as follows:
the data enhancement processing is performed by overlapping the slices by dividing with windows of equal length, and acquiring more sample signals as much as possible. I.e. setting the step length of the sliding window to lwindowLess than a single sample signal length linputLength of lsignalCan be divided into Nums ═ (l)signal-linput)/lwindowA sample signal;
and marking relevant parameters such as equipment fault type, equipment name, sampling point number, sensitivity, acquisition time and the like in the acquisition process of each group of sample signals, checking whether a corresponding field of each group of sample signals has a null value or not, and checking whether the field is perfect or not, and finishing the preprocessing of data.
The feature data extraction module comprises a multi-scale time domain feature extraction module and a deep noise reduction feature extraction module, wherein:
the multi-scale time domain feature extraction module decomposes the acquired vibration signals into a series of stable IMF components with different feature scales and obvious fault features by adopting a set empirical mode decomposition method, and a specific flow of the module is shown in FIG. 3 and is specifically realized as follows:
1, defining the total times MAX _ ITER of EMD to be carried out;
2, adding white noise in normal distribution to the original signal, and decomposing the signal to be processed by EMD to obtain IMF component CijRepresenting i IMF components decomposed in the j-th time, and repeating the operation until the number of empirical mode decomposition times reaches MAX _ ITER;
3, calculating IMF components corresponding to each stage and calculating the average value thereof
Figure BDA0002628862990000091
As the stable IMF component finally obtained;
4, solving the energy size of each IMF component
Figure BDA0002628862990000092
In the formula, N represents that the signal is divided into N equal parts, and the time length of each equal part is delta t, ci(tn) represents the i-thThe amplitude of each component at the time tn, h is the total number of IMF components, the first 8 IMF components with larger energy are selected, and the kurtosis indexes of the 8 IMF components are calculated
Figure BDA0002628862990000093
Taking the 5 IMF components with larger kurtosis values,
Figure BDA0002628862990000094
as a result of multi-scale time-domain feature extraction, where Nc is the length of the signal,
Figure BDA0002628862990000095
1xNc representing the IMF component as a real dimension; the output results are shown in fig. 4.
The deep noise reduction feature extraction module trains a deep noise reduction self-encoder by adding noise, and the obtained model can effectively perform noise reduction filtering and feature extraction on an original vibration signal, and is specifically realized as follows:
the method comprises the steps of 1, adding random noise to train a deep noise reduction self-encoder, namely, taking a vibration signal with noise as an input and taking a vibration signal without noise as an output, training a plurality of self-encoders which are laminated together by changing a reconstruction error term of a loss function, and enabling the self-encoders to learn to remove the noise to obtain a real input which is not polluted by the noise. The input of one self-encoder is the output of the next self-encoder, L self-encoders are stacked to form L hidden layers, and the output of each layer of self-encoding network is
Figure BDA0002628862990000101
Wherein
Figure BDA0002628862990000102
Representing hidden layer activation function, Wl a,
Figure BDA0002628862990000103
Representing the parameters learned by the self-encoder during the training process, Xl-1Representing the output of the previous layer from the encoder;
2, after training, keeping original network parameters, inputting a vibration signal to be processed, and outputting deep noise reduction characteristics through a deep noise reduction self-encoder
Figure BDA0002628862990000104
The multi-feature fusion convolutional neural network module in fig. 2 inputs the IMF component and the deep noise reduction feature extracted by the first two modules into the deep convolutional neural network for training through feature combination, and finally outputs the probability value of occurrence of each fault to complete the fault diagnosis task, which is specifically implemented as follows:
1, the characteristics of the two modules, i.e. each IMF component
Figure BDA0002628862990000105
And depth noise reduction features
Figure BDA0002628862990000106
Splicing is carried out, the components are stacked into a multi-channel sample according to the time sequence, and the data dimension is increased by one dimension, namely the data dimension is changed from 1 × Nc to 1 × Nc × 6;
2, constructing and training a deep convolutional neural network, wherein the width of a convolutional kernel of a first layer is 1, the number of channels is 6, extracting convolution information of 5 IMF components and corresponding moments in 1 deep noise reduction feature, and outputting a feature map:
Figure BDA0002628862990000107
wherein denotes a convolution, Mi(i ═ 1, 2.., 6) denotes the convolution kernel for these 6 channels. The first layer convolution process is illustrated in fig. 5, with a first layer set of convolution kernels of dimension 1 × q × 6, and the resulting first layer convolution map of dimension 1 × (Nc-q +1) × 1. And subsequently establishing a multi-layer convolution layer, a batch normalization layer, an activation layer and a pooling layer, and finally establishing a Dropout layer, a full connection layer and a classifier to complete the construction of the neural network. Completing the training of a convolutional neural network in a supervision mode by inputting a training sample;
and 3, applying a deep convolutional neural network, inputting the combination characteristics, and outputting a diagnosis result, namely the probability of each fault type to finish a fault diagnosis task.
The following example is incorporated for gearbox fault diagnosis:
the collected gear fault states are four, including slight peeling of gears, severe peeling of gears, local abrasion and overall abrasion, and the fault samples are expanded from original 120 to 600 through the data collection and preprocessing module, so that the fault samples are enriched, a foundation is laid for training a deep network, and each sample contains 1024-point vibration data.
Decomposing an original signal into a plurality of IMF components through a multi-scale time domain feature extraction module, calculating energy values and kurtosis values of the IMF components, aiming at a sample signal calculation result of the gear severe spalling fault, as shown in table 1, selecting IMF components 1,2,3,4 and 6 as multi-scale time domain features representing the sample signal according to an energy value method and a kurtosis value method, and waveforms of the five components are shown in fig. 4.
TABLE 1 energy values and kurtosis values calculation results
Figure BDA0002628862990000111
In the deep noise reduction feature extraction module, a training deep noise reduction self-encoder is used, wherein noise parameters are set to be 0.5, the number of neurons in a hidden layer is 512,256,128,128,256,512, Dropout is 0.3, an activation function selects ReLu, a gradient descent method is adopted for training, and the trained deep noise reduction self-encoder can effectively filter noise and extract features of an original signal.
In the multi-feature fusion convolutional neural network module, parameters of each layer of a neural network are designed as shown in table 2, in addition, a Dropout parameter is 0.3, an activation function is ReLu, and a deep convolutional neural network is established through network model building as shown in table 2, so that the expression capability of the network is improved, and the probability of fault occurrence is obtained through diagnosis.
TABLE 2 convolutional neural network parameters
Figure BDA0002628862990000112
Through the construction of the modules, the test fault diagnosis precision is 95.32%, compared with the model diagnosis precision established by using a more traditional method SVM support vector machine which is 89.43%, the model fault diagnosis rate established by the method is greatly improved, and the network shows higher robustness by combining a multi-scale time domain characteristic and a deep noise reduction characteristic, so that a good classification effect is achieved.
In conclusion, the method and the device solve the problems of single fault feature and low diagnosis efficiency to a certain extent, and improve the fault diagnosis accuracy by combining the multi-scale time domain feature and the deep noise reduction feature and fusing the two features by using the deep convolution neural network.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.

Claims (10)

1. A gearbox fault diagnosis method is used for wind driven generator gearbox fault diagnosis, a gearbox shaft horizontal vibration signal and a gearbox shaft vertical vibration signal are obtained through an acceleration sensor arranged on a wind driven generator, characteristic data of the gearbox shaft horizontal vibration signal and the gearbox shaft vertical vibration signal are extracted, the extracted characteristic data are input into a deep convolution neural network module to obtain a fault diagnosis result, the deep convolution neural network module is a fault diagnosis model which is learned and established in advance according to the gearbox shaft horizontal and vertical vibration signal characteristic data of different faults of a gearbox, and the method is characterized in that the characteristic data extraction of the gearbox shaft horizontal vibration signal and the gearbox shaft vertical vibration signal comprises the following steps: performing multi-scale time domain feature extraction and deep noise reduction feature extraction:
the multi-scale time domain feature extraction step comprises:
firstly, decomposing a vibration signal into stable signal IMF components according to a set time length, obtaining the energy of each IMF component through a formula (1),
Figure FDA0002628862980000011
wherein: n denotes that the signal is divided into N equal parts, each equal part having a time length of Deltat, ci(tn) represents the amplitude of the ith component at the time tn, and h is the total number of IMF components;
secondly, selecting the first S IMF components with energy sorted from big to small, calculating the kurtosis value index of the S IMF components through a formula (2),
Figure FDA0002628862980000012
thirdly, taking the first P IMF components with the kurtosis value in a descending order
Figure FDA0002628862980000013
As a result of multi-scale time-domain feature extraction, where Nc is the length of the signal,
Figure FDA0002628862980000021
1xNc indicating that the IMF component is a real number dimension, P being less than S;
the deep noise reduction feature extraction comprises the following steps: and sending the vibration signal into a depth noise reduction self-encoder, and outputting a depth noise reduction characteristic through the depth noise reduction self-encoder.
2. The method of claim 1, wherein the horizontal vibration signals comprise axial horizontal vibration signals and radial horizontal vibration signals.
3. The method according to claim 1 or 2, characterized in that the step of establishing the fault diagnosis model is:
firstly, respectively extracting the multi-scale time domain features and the deep noise reduction features according to different faults;
and secondly, fusing the multi-scale time domain features and the depth noise reduction features: splicing each IMF component in the multi-scale time domain features and the deep noise reduction features, stacking the IMF components into a multi-channel training sample according to a time sequence order, constructing and training a deep convolutional neural network, and completing the training of the convolutional neural network through the input of the training sample;
wherein each layer of convolution layer outputs a feature map of formula (3) to complete feature fusion:
Figure FDA0002628862980000022
wherein denotes a convolution, Mi1, 2., P +1 denotes the convolution kernel of P +1 channels;
and thirdly, forming the probability of each fault type according to the feature fusion and outputting a fault diagnosis result.
4. The method according to claim 1 or 2, wherein the feature data extraction is feature data extraction of a plurality of sample signals of the vibration signal, the plurality of sample signals being obtained by dividing a length of the vibration signal by a window of equal length.
5. The method of claim 4, further comprising labeling device name, sampling point number, sensitivity, and acquisition time parameters during each of the sets of the multiple sample signals, and checking whether the corresponding field of each of the sets of the multiple sample signals has a null value and whether the field is complete.
6. A method according to claim 1 or 2, characterized in that the method of decomposing the vibration signal into stationary signal IMF components for a set length of time is:
firstly, defining the total times MAX _ ITER of EMD to be subjected to empirical mode decomposition;
secondly, white noise with normal distribution is added to the original signal, and the white noise is processed through an empirical modeDecomposing EMD to decompose the signal to be processed to obtain IMF component Cij，CijRepresenting i IMF components decomposed in the j-th time, and repeating the operation until the number of empirical mode decomposition times reaches MAX _ ITER;
thirdly, calculating IMF components corresponding to each time point and calculating the average value according to a formula (4),
Figure FDA0002628862980000031
finally spliced according to time sequence into
Figure FDA0002628862980000032
As the final found stationary signal IMF component.
7. A gearbox fault diagnosis system implementing the method of the preceding claims, comprising: the wind driven generator is characterized in that a bearing seat, a gear transmission box and a generator shell, which are butted with the wind driven rotating blade hub, of an input shaft are respectively provided with an acceleration sensor for sensing horizontal vibration signals and vertical vibration signals of a gear shaft, the acceleration sensor is connected with a fault diagnosis and analysis computer through a wire or a wireless way, a signal receiving interface card is arranged in the computer, the signal receiving interface card is connected with the acceleration sensor to send the received horizontal vibration signals and the received vertical vibration signals of the gear to a central processing unit, and a vibration signal preprocessing module, a characteristic data extraction module and a deep convolution neural network module are arranged in the central processing unit;
the vibration signal preprocessing module divides the lengths of the acquired horizontal vibration signals and the acquired vertical vibration signals of the gear into a plurality of sample signals by using windows with equal lengths, records the equipment name, the number of sampling points, the sensitivity and the acquisition time parameters of the corresponding vibration signals, and checks whether corresponding fields of each group of the plurality of sample signals have empty values or not and whether the fields are complete or not;
the characteristic data extraction module is used for extracting the characteristics of the sample signal provided by the preprocessing module;
and the deep convolutional neural network module is used for carrying out fault diagnosis on the characteristic data extracted by the characteristic data extraction module and outputting a diagnosis result.
8. The system of claim 7, wherein the acceleration sensor comprises: the wind driven generator comprises an input shaft axial horizontal vibration sensor arranged on the axial end face of a bearing seat, wherein the input shaft axial horizontal vibration sensor is butted with a hub of a wind driven rotating blade, an input shaft radial horizontal vibration sensor arranged on the horizontal radial end face of the bearing seat, a box body vertical vibration sensor and a box body radial horizontal vibration sensor arranged on a gear transmission box body, and a generator shaft radial horizontal vibration sensor respectively arranged on an in-out output shaft of a generator.
9. The system of claim 8, wherein the box vertical vibration sensor is disposed in the middle of the upper side of the box, and the two box radial horizontal vibration sensors are disposed on two sides of the box vertical vibration sensor on the side of the box.
10. The system of claim 8, wherein the acceleration sensor is fixed in a set position by means of a mosaic.
CN202010805158.1A 2020-08-12 2020-08-12 Gear box fault diagnosis method and system Pending CN111855202A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010805158.1A CN111855202A (en) 2020-08-12 2020-08-12 Gear box fault diagnosis method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010805158.1A CN111855202A (en) 2020-08-12 2020-08-12 Gear box fault diagnosis method and system

Publications (1)

Publication Number Publication Date
CN111855202A true CN111855202A (en) 2020-10-30

Family

ID=72972821

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010805158.1A Pending CN111855202A (en) 2020-08-12 2020-08-12 Gear box fault diagnosis method and system

Country Status (1)

Country Link
CN (1) CN111855202A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113255780A (en) * 2021-05-28 2021-08-13 润联软件系统(深圳)有限公司 Reduction gearbox fault prediction method and device, computer equipment and storage medium
CN113532835A (en) * 2021-08-18 2021-10-22 中国国家铁路集团有限公司 Railway contact net hard spot diagnosis method and device
CN114034481A (en) * 2021-11-15 2022-02-11 燕山大学 Fault diagnosis system and method for rolling mill gearbox
CN116417013A (en) * 2023-06-09 2023-07-11 中国海洋大学 Underwater propeller fault diagnosis method and system

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113255780A (en) * 2021-05-28 2021-08-13 润联软件系统(深圳)有限公司 Reduction gearbox fault prediction method and device, computer equipment and storage medium
CN113255780B (en) * 2021-05-28 2024-05-03 润联智能科技股份有限公司 Reduction gearbox fault prediction method and device, computer equipment and storage medium
CN113532835A (en) * 2021-08-18 2021-10-22 中国国家铁路集团有限公司 Railway contact net hard spot diagnosis method and device
CN114034481A (en) * 2021-11-15 2022-02-11 燕山大学 Fault diagnosis system and method for rolling mill gearbox
CN116417013A (en) * 2023-06-09 2023-07-11 中国海洋大学 Underwater propeller fault diagnosis method and system
CN116417013B (en) * 2023-06-09 2023-08-25 中国海洋大学 Underwater propeller fault diagnosis method and system

Similar Documents

Publication Publication Date Title
CN111855202A (en) Gear box fault diagnosis method and system
CN108344574B (en) A kind of Wind turbines Method for Bearing Fault Diagnosis based on depth joint adaptation network
CN109555566B (en) Steam turbine rotor fault diagnosis method based on LSTM
CN113255078A (en) Bearing fault detection method and device under unbalanced sample condition
CN109946080B (en) Mechanical equipment health state identification method based on embedded circulation network
CN113008559B (en) Bearing fault diagnosis method and system based on sparse self-encoder and Softmax
CN108827605A (en) A kind of mechanical breakdown characteristic automatic extraction method based on improvement sparseness filtering
CN112665852B (en) Variable working condition planetary gearbox fault diagnosis method and device based on deep learning
CN112541233A (en) Rotary machine fault diagnosis method based on improved convolutional neural network
CN112257528A (en) Wind power gear box fault diagnosis method based on wavelet transformation and dense connection expansion convolutional neural network
Chen et al. Deep residual network for identifying bearing fault location and fault severity concurrently
CN112305388B (en) On-line monitoring and diagnosing method for insulation partial discharge faults of generator stator winding
CN116358871B (en) Rolling bearing weak signal composite fault diagnosis method based on graph rolling network
CN112364706A (en) Small sample bearing fault diagnosis method based on class imbalance
CN113240022A (en) Wind power gear box fault detection method of multi-scale single-classification convolutional network
CN115587290A (en) Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network
CN116956215A (en) Fault diagnosis method and system for transmission system
CN112160877A (en) Fan bearing fault diagnosis method based on SFA and CNN
CN113916535B (en) Bearing diagnosis method, system, equipment and medium based on time frequency and CNN
CN212363649U (en) System for realizing fault diagnosis method of gearbox
CN111539381B (en) Construction method of wind turbine bearing fault classification diagnosis model
CN113758709A (en) Rolling bearing fault diagnosis method and system combining edge calculation and deep learning
CN113673442A (en) Variable working condition fault detection method based on semi-supervised single classification network
CN116484258A (en) Elevator traction machine bearing fault diagnosis method
CN109886109B (en) Behavior identification method based on deep learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination