CN113865870A - Bearing fault edge diagnosis method and system based on wavelet improved MobileNet network - Google Patents
Bearing fault edge diagnosis method and system based on wavelet improved MobileNet network Download PDFInfo
- Publication number
- CN113865870A CN113865870A CN202111159275.6A CN202111159275A CN113865870A CN 113865870 A CN113865870 A CN 113865870A CN 202111159275 A CN202111159275 A CN 202111159275A CN 113865870 A CN113865870 A CN 113865870A
- Authority
- CN
- China
- Prior art keywords
- same
- layer
- wavelet
- diagnosis
- improved
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 76
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 40
- 240000007651 Rubus glaucus Species 0.000 claims abstract description 10
- 235000011034 Rubus glaucus Nutrition 0.000 claims abstract description 10
- 235000009122 Rubus idaeus Nutrition 0.000 claims abstract description 10
- 230000004913 activation Effects 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 20
- 230000001133 acceleration Effects 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 238000002372 labelling Methods 0.000 claims description 8
- 238000004891 communication Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 239000006185 dispersion Substances 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000004806 packaging method and process Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 abstract description 4
- 238000013136 deep learning model Methods 0.000 abstract description 2
- 238000005096 rolling process Methods 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
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
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a bearing fault edge diagnosis method and a system based on a wavelet improved MobileNet network, wherein the method comprises the steps of issuing collected bearing vibration data; inputting the collected historical vibration data into an improved MobileNetV3-Small network for training to obtain a diagnosis model; and inputting the vibration data acquired in real time into a diagnosis model for fault diagnosis. The method simplifies a large convolution model trained in a Tensorflow environment by utilizing a wavelet-based improved MoblieNetV3-small network, and then places the simplified deep learning model into a raspberry group which has poorer calculation power and is dozens of times cheaper than a computer for real-time diagnosis.
Description
Technical Field
The invention relates to a bearing fault edge diagnosis method and system based on a wavelet improved MobileNet network, and belongs to the field of fault diagnosis of mechanical equipment.
Background
In the field of fault diagnosis of mechanical equipment, a Convolutional Neural Network (CNN) is widely applied to fault diagnosis and condition monitoring of equipment, the diagnosis precision is higher and higher, and a qualitative leap is achieved in the field of fault of bearings. However, in order to pursue the accuracy of diagnosis excessively, the depth of the convolutional neural network is getting deeper, and the model is getting more complicated, such as the residual error network (ResNet) has as many as 152 layers, the parameter amount is calculated in billions, if it is stored, it needs more than five trillions of disk space, not to mention that the operation amount is more than three billions. Such large diagnostic models are typically deployed to run on high-performance, computationally intensive CPUs/GPUs, but take a long time. Due to the limitation of hardware resources and computing power, the mobile device and the embedded system are difficult to directly deploy the complex CNN model. In addition, the low-delay and quick-response requirements for real-time evaluation of the equipment state at present make the diagnosis and training time of a large model difficult to meet the real-time requirement; therefore, how to reduce the model volume and improve the model operation speed while ensuring the model accuracy and how to deploy on an embedded platform becomes a necessary requirement for application and popularization of the neural network in engineering.
Disclosure of Invention
The invention provides a bearing fault edge diagnosis method and system based on a wavelet improved MobileNet network, which are used for training the wavelet improved MobileNet network according to collected bearing vibration real-time waveform data and further realizing edge fault diagnosis.
The technical scheme of the invention is as follows: a bearing fault edge diagnosis method based on a wavelet improved MobileNet network comprises the following steps:
issuing the collected bearing vibration data;
inputting the collected historical vibration data into an improved MobileNetV3-Small network for training to obtain a diagnosis model;
and inputting the vibration data acquired in real time into a diagnosis model for fault diagnosis.
The step of issuing the vibration data of the collected bearing specifically comprises the following steps: setting channel configuration and sampling configuration parameters for data acquisition through an upper computer, and acquiring vibration data of the bearing acquired by a sensor on a real-time controller by using an acquisition card; extracting a trend index of a signal from the collected vibration data, packaging the trend index and real-time waveform data of the vibration data together, transmitting the packaged data to an upper computer in a network release form through a shared variable, and displaying the packaged data in real time through the upper computer; meanwhile, a UDP communication protocol is adopted to upload the packed data to a database for storage; and meanwhile, the packed data is uploaded to the raspberry pi embedded diagnosis system by adopting a TCP communication protocol.
The real-time controller adopts a compact RIO embedded platform, and the sensor adopts an IEPE acceleration sensor; the IEPE acceleration sensor is connected with the acquisition card through a sensor connecting wire, the IEPE acceleration sensor is arranged above the tested bearing, and one side of the acquisition card is inserted into a compact RIO embedded platform slot; and connecting the upper computer with the compactRIO embedded platform through a network cable.
The method for obtaining the diagnosis model by inputting the collected historical vibration data into an improved MobileNetV3-Small network for training specifically comprises the following steps: the method comprises the steps of downloading historical waveform data stored in a database through an upper computer, preprocessing the historical waveform data, dividing a training set and a verification set, and importing the preprocessed training set and the preprocessed test set into an improved MobileNet V3-Small network for training to obtain a diagnosis model.
The pretreatment operation specifically comprises the following steps: and (6) normalizing and labeling.
The normalization is dispersion normalization; the labeling specifically comprises the following steps: the unique identifier is used to represent different bearing states.
In training, the input data length is 2048, all parameters are updated through back propagation and an Adam optimization algorithm, a cross entropy loss function is used for calculating a loss function, the initial learning rate is set to be 0.001, and the attenuation rate is 0.98; the training process adopts mini-batch learning, and the size is set to be 32.
The improved MobileNet V3-Small network has a wavelet convolution layer, 5 bottleeck layers and two full-connection layers; convolution kernels in the first layer of convolution layers are constructed using Daubechies wavelet function structures.
The improved MobileNet V3-Small network comprises the following specific parameters:
a first layer: wavelet convolution layer: filter 27, kernel _ size 55, threads 1, padding same as each other as same as each other as same;
a second layer: bottleeck layer: filter 16, kernel _ size 3, strings 16, padding same as same, activation same as ReLu, pool same as MaxPooling, pool _ size 2, input _ shape 2048 × 27, and output _ shape 64 × 16;
and a third layer: bottleeck layer: filter 32, kernel _ size 3, threads 1, padding same as same, activation same as ReLu, pool same as MaxPooling, pool _ size 2, input _ shape 64 × 16output _ shape 32 × 32;
a fourth layer: bottleeck layer: filter 64, kernel _ size 3, threads 1, padding same as each other as same as each other as same;
and a fifth layer: bottleeck layer: filter 64, kernel _ size 5, threads 1, padding same as same, activation same as h-shock, pool same as MaxPooling, pool _ size 2, input _ shape 16 × 64, and output _ shape 8 × 64;
a sixth layer: bottleeck layer: filter 64, kernel _ size 5, threads 1, padding same as same, activation same as h-swish, pool same as MaxPooling, pool _ size 2, input _ shape 8 × 64, and output _ shape 4 × 64;
a seventh layer: full connection layer: unity is 100; activation ═ ReLu';
an eighth layer: full connection layer: unity is 4 and activation is 'softmax'.
A bearing fault edge diagnostic system based on a wavelet improved MobileNet network comprises:
the acquisition and release module is used for releasing the vibration data of the acquired bearing;
the model training module is used for inputting the collected historical vibration data into an improved MobileNet V3-Small network for training to obtain a diagnosis model;
and the fault diagnosis module is used for inputting the vibration data acquired in real time into the diagnosis model for fault diagnosis.
The invention has the beneficial effects that:
the invention aims to solve the bottleneck problem that the current neural network is difficult to be applied in engineering practice due to poor instantaneity, a large-scale convolution model trained in a Tensflow environment is simplified by utilizing a wavelet-based improved MoblieNet V3-small network, and then the simplified deep learning model is put into a raspberry group with poor calculation power and price which is dozens of times lower than that of a computer for real-time diagnosis. A new idea is provided for the subsequent edge fault diagnosis of the lightweight embedded system.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a graph comparing the structure of a conventional MobileNet V3-Small model with a modified MobileNet V3-Small model based on wavelets;
FIG. 3 is a schematic diagram of a process for training a MobileNet V3-Small model based on wavelet improvement.
Detailed Description
The invention will be further described with reference to the following figures and examples, without however restricting the scope of the invention thereto.
Example 1: as shown in fig. 1 to 3, a wavelet-based improved MobileNet network bearing fault edge diagnosis method includes:
issuing the collected bearing vibration data;
inputting the collected historical vibration data into an improved MobileNetV3-Small network for training to obtain a diagnosis model;
and inputting the vibration data acquired in real time into a diagnosis model for fault diagnosis.
Optionally, the method further comprises: after the model is trained, in order to ensure the accuracy of the model, the model needs to be periodically updated by using recently acquired data, so as to ensure the accuracy of diagnosis.
Optionally, the publishing the vibration data of the collected bearing specifically includes: setting parameters of channel configuration (signal type, sampling frequency and sampling point number) and sampling configuration (continuous sampling or interval sampling) of data acquisition through an upper computer, and acquiring vibration data of the bearing acquired by a sensor on a real-time controller by using an acquisition card; the collected vibration data are subjected to noise reduction and filtering treatment, trend indexes (which can be effective values, kurtosis, frequency spectrums and envelope spectrums) of signals are extracted, the trend indexes and real-time waveform data of the vibration data are packed together, the packed data are transmitted to an upper computer in a network release mode through shared variables, and real-time display is performed through the upper computer (information obtained through a real-time displayed trend graph can verify that the result diagnosed from a model is correct, and diagnosis misjudgment can be prevented to a certain extent); meanwhile, the packed data is uploaded to a MySQL database for storage by adopting a UDP communication protocol, and the data stored in the MySQL database is downloaded by an upper computer; and meanwhile, the packed data is uploaded to the raspberry pi embedded diagnosis system by adopting a TCP communication protocol.
Optionally, the real-time controller adopts a CompactRIO embedded platform, and the sensor adopts an IEPE acceleration sensor; the IEPE acceleration sensor is connected with the acquisition card through a sensor connecting wire, the IEPE acceleration sensor is arranged above the tested bearing, and one side of the acquisition card is inserted into a compact RIO embedded platform slot; and connecting the upper computer with the compactRIO embedded platform through a network cable. Can be improved by adopting a specific model of embedded platform
Optionally, the step of inputting the collected historical vibration data into an improved MobileNetV3-Small network for training to obtain a diagnostic model specifically includes: the method comprises the steps of downloading historical waveform data stored in a MySQL database through an upper computer, preprocessing the historical waveform data, dividing a training set and a verification set, and importing the preprocessed training set and the preprocessed testing set into an improved MobileNet V3-Small network for training to obtain a diagnosis model.
Optionally, the preprocessing operation specifically includes: and (6) normalizing and labeling.
Optionally, the normalization is dispersion normalization; the labeling specifically comprises the following steps: adopting unique identification to represent different bearing states; the unique identification can be a number or a letter number. The training set and validation set were divided in a ratio of 0.7: 0.3.
Optionally, in model training, the input data length is 2048, all parameters are updated through back propagation and Adam optimization algorithm, a cross entropy loss function is used for calculation of a loss function, the initial learning rate is set to 0.001, and the attenuation rate is 0.98; the training process adopts mini-batch learning, and the size is set to be 32.
Optionally, the modified MobileNetV3-Small network shares one wavelet convolution layer, 5 bottleeck layers, and two fully connected layers; convolution kernels in the first layer of convolution layers are constructed using Daubechies wavelet function structures.
Optionally, the specific parameters of the modified MobileNetV3-Small network are as follows:
a first layer: wavelet convolution layer: filter 27, kernel _ size 55, threads 1, padding same as each other as same as each other as same;
a second layer: bottleeck layer: filter 16, kernel _ size 3, strings 16, padding same as same, activation same as ReLu, pool same as MaxPooling, pool _ size 2, input _ shape 2048 × 27, and output _ shape 64 × 16;
and a third layer: bottleeck layer: filter 32, kernel _ size 3, threads 1, padding same as same, activation same as ReLu, pool same as MaxPooling, pool _ size 2, input _ shape 64 × 16output _ shape 32 × 32;
a fourth layer: bottleeck layer: filter 64, kernel _ size 3, threads 1, padding same as each other as same as each other as same;
and a fifth layer: bottleeck layer: filter 64, kernel _ size 5, threads 1, padding same as same, activation same as h-shock, pool same as MaxPooling, pool _ size 2, input _ shape 16 × 64, and output _ shape 8 × 64;
a sixth layer: bottleeck layer: filter 64, kernel _ size 5, threads 1, padding same as same, activation same as h-swish, pool same as MaxPooling, pool _ size 2, input _ shape 8 × 64, and output _ shape 4 × 64;
a seventh layer: full connection layer: unity is 100; activation ═ ReLu';
an eighth layer: full connection layer: unity is 4 and activation is 'softmax'.
The model described above can be trained in a TensorFlow environment under pycharm. Because the rotary machine vibration signal has the characteristics of serious interference of time domain vibration noise and very rich frequency domain information, and people are more concerned about the expression of the vibration signal in the frequency domain rather than the time domain, the wavelet convolution layer is adopted to replace a standard convolution layer, so that a network model can be more suitable for the processing of the rotary machine vibration signal, and furthermore, the parameters of the wavelet convolution layer and a bottleneck layer are specifically set, so that the invention can avoid the adverse conditions that the generalization capability of the model is weak and the conversion result is excessively depended on due to the fact that one-dimensional signals are converted into two-dimensional signals to be diagnosed by a MobileNet V3-small network for image processing; furthermore, by deleting the bottleeck layer, on the basis of ensuring the training precision, the parameter quantity of the model is greatly reduced, the diagnosis efficiency is improved, and the over-fitting phenomenon is avoided.
Optionally, the step of inputting the vibration data acquired in real time into the diagnosis model to perform fault diagnosis specifically includes: the raspberry group embedded diagnosis system analyzes the data packet, loads the model, loads the analyzed real-time waveform data into the diagnosis model for diagnosis, obtains a label diagnosis result output by the model, packages the analyzed time and channel information together with the diagnosis result, and sends the packaged result to an upper computer for result display, and triggers a fault alarm when the label result is identified as a fault.
A bearing fault edge diagnosis system based on wavelet improved MobileNet network is characterized in that: the method comprises the following steps:
the acquisition and release module is used for releasing the vibration data of the acquired bearing;
the model training module is used for inputting the collected historical vibration data into an improved MobileNet V3-Small network for training to obtain a diagnosis model;
and the fault diagnosis module is used for inputting the vibration data acquired in real time into the diagnosis model for fault diagnosis.
The invention provides a rolling bearing fault edge diagnosis method and system based on wavelet improved MobileNet network, aiming at the problems that a large model trained based on a deep learning method cannot be deployed to an embedded system with poor computing power, the diagnosis real-time response is poor and the like. The method has the core that wavelet convolution which is more effective in extracting the fault characteristics of the original data aiming at the vibration signals is introduced to replace a first standard convolution layer in the traditional MoblieNet V3-small network, so that the model is more suitable for the vibration signals, and the diagnosis precision is improved. And the original two-dimensional MobileNetV3-Small is subjected to dimension reduction aiming at the one-dimensional vibration signal, so that the one-dimensional data can be directly processed, and the generalization capability of the model is improved. And the improved model is deployed in a raspberry embedded system, and real-time fault diagnosis of the rolling bearing is realized at the edge end. As shown in particular in figure 1.
MobileNetV3-Small is a lightweight attention model for raspberry pi-oriented embedded platforms, a network structure generated using a Neural Architecture Search (NAS) technique (searching for global network structures by optimizing each network block) and a NetAdapt algorithm complementary thereto (searching for the number of filters per layer). And finding an optimized model through the combination of the two.
The conventional MobileNetV3-Small version has 11 bottomtech layers, one standard convolutional layer, two fully-connected layers, introduces a 5 × 5 deep convolution instead of a partial 3 × 3 deep convolution, and introduces a Squeeze-and-excitation (se) module and h-swish (hs) activation function to improve model accuracy.
The SE module provides a mechanism which can enable the network model to calibrate the characteristics, so that the effective weight is large, and the ineffective or small effect is small; and the swish activation function can effectively improve the accuracy of the network, but the swish calculation amount is too large, and the swish activation function is not suitable for a lightweight neural network. Mobilenet 3 finds a similar but much less computationally expensive alternative activation function h-swish (hard version of swish), whose expression is as follows:
wherein: the purpose of ReLU6 is to suppress the maximum value, and to have a good numerical resolution even at low precision of the moving end, i.e. when x >6, the derivative is also 0.
The following information is given for the improved MobileNetV3-Small network model training:
1. the normalization is a dispersion normalization, and the data is linearly transformed so that the result value is mapped between [0-1 ]. The transfer function is as follows:
wherein xiFor each sample data, max { xjIs the maximum value of the sample data, min { x }jIs the minimum value of the sample data.
After normalization operation is finished, data sets are divided into a training set and a verification set in a sampling mode according to the proportion of 0.7:0.3, then labeling operation is finished, wherein in labeling, a label 0 represents normal, 1 represents bearing inner ring fault, 2 represents bearing rolling element fault, and 3 represents bearing outer ring fault. In order to improve the accuracy of the fault, the input data length is 2048, all parameters are updated through back propagation and an Adam optimization algorithm, a cross entropy loss function is used for calculating a loss function, the initial learning rate is set to be 0.001, and the attenuation rate is 0.98. The training process adopts mini-batch learning, and the size is set to be 32. And (5) verifying the model by adopting a verification set while training until the training is finished, and storing the trained model. And issuing the model to the edge node formed by the raspberry group. The specific model training process is shown in fig. 3.
2. The essence of wavelet convolution is wavelet transformation, and by selecting different scales of basis functions psi (f, t) to make convolution on signals f (t), and by constructing proper basis functions, more ideal time-frequency analysis results can be obtained. The wavelet transform is as follows:
S(a,τ)＝f(t)*ψf,t)＝∫f(t)ψτ-t,a)dt
the specific operation in the convolutional layer is the convolution operation between the input and the convolutional kernel, which is specifically as follows:
in the formula: i represents the ith convolution kernel of the kth layer, z (i) represents a feature map obtained by learning the ith convolution kernel,
which is representative of the input signal(s),
which represents the kernel of the convolution,
representing the bias of the convolution kernel.
The general expression of wavelet convolution can be obtained according to the two formulas: g (i) ═ x · ψ (f, t); the present invention utilizes Daubechies-based wavelets to construct a convolution kernel with different wavelet structures.
3. And (4) the full connection layer is fitted by a Softmax function, and probability distribution prediction is carried out on the input data, wherein the state represented by the label with the highest probability represents the state of the bearing at the moment. The formula of the Softmax function is as follows:
in the formula: assuming a total of n neurons in the output layer, ykIs the output of the kth neuron, akTo input data.
The model is tested by selecting bearing vibration signal data disclosed by the university of Kaiser Sichu, the bearing selected in the experiment is an SKF bearing, the rotating speed of a motor is about 1772r/min, the sampling frequency is 12kHz under the working condition of 1HP load, and collected driving end (DE end) vibration data. The data set comprises four state data of a normal bearing, an inner ring fault, a rolling body fault and an outer ring fault, and the fault diameter is 0.1778 mm. The results obtained with 15 training sessions with the above parameter settings are shown in the following table:
as can be seen from the table, the improved model of the invention can improve the accuracy and obviously improve the running speed by jointly matching wavelet convolution replacement and deletion of the bottleeck layer. Therefore, the improved MobileNet network based on wavelet improvement is deployed in a raspberry group embedded system, real-time fault diagnosis of a rolling bearing is realized at the edge end, and the advantage of edge diagnosis can be effectively exerted.
While the present invention has been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art.
Claims (10)
1. A bearing fault edge diagnosis method based on wavelet improved MobileNet network is characterized in that: the method comprises the following steps:
issuing the collected bearing vibration data;
inputting the collected historical vibration data into an improved MobileNetV3-Small network for training to obtain a diagnosis model;
and inputting the vibration data acquired in real time into a diagnosis model for fault diagnosis.
2. The wavelet improved MobileNet network based bearing fault edge diagnosis method according to claim 1, wherein: the step of issuing the vibration data of the collected bearing specifically comprises the following steps: setting channel configuration and sampling configuration parameters for data acquisition through an upper computer, and acquiring vibration data of the bearing acquired by a sensor on a real-time controller by using an acquisition card; extracting a trend index of a signal from the collected vibration data, packaging the trend index and real-time waveform data of the vibration data together, transmitting the packaged data to an upper computer in a network release form through a shared variable, and displaying the packaged data in real time through the upper computer; meanwhile, a UDP communication protocol is adopted to upload the packed data to a database for storage; and meanwhile, the packed data is uploaded to the raspberry pi embedded diagnosis system by adopting a TCP communication protocol.
3. The wavelet improved MobileNet network based bearing fault edge diagnosis method according to claim 2, wherein: the real-time controller adopts a compact RIO embedded platform, and the sensor adopts an IEPE acceleration sensor; the IEPE acceleration sensor is connected with the acquisition card through a sensor connecting wire, the IEPE acceleration sensor is arranged above the tested bearing, and one side of the acquisition card is inserted into a compact RIO embedded platform slot; and connecting the upper computer with the compactRIO embedded platform through a network cable.
4. The wavelet improved MobileNet network based bearing fault edge diagnosis method according to claim 1, wherein: the method for obtaining the diagnosis model by inputting the collected historical vibration data into an improved MobileNetV3-Small network for training specifically comprises the following steps: the method comprises the steps of downloading historical waveform data stored in a database through an upper computer, preprocessing the historical waveform data, dividing a training set and a verification set, and importing the preprocessed training set and the preprocessed test set into an improved MobileNet V3-Small network for training to obtain a diagnosis model.
5. The wavelet improved MobileNet network based bearing fault edge diagnosis method according to claim 4, wherein: the pretreatment operation specifically comprises the following steps: and (6) normalizing and labeling.
6. The wavelet improved MobileNet network based bearing fault edge diagnosis method according to claim 5, wherein: the normalization is dispersion normalization; the labeling specifically comprises the following steps: the unique identifier is used to represent different bearing states.
7. The wavelet improved MobileNet network based bearing fault edge diagnosis method according to claim 1, wherein: in training, the input data length is 2048, all parameters are updated through back propagation and an Adam optimization algorithm, a cross entropy loss function is used for calculating a loss function, the initial learning rate is set to be 0.001, and the attenuation rate is 0.98; the training process adopts mini-batch learning, and the size is set to be 32.
8. The wavelet improved MobileNet network based bearing fault edge diagnosis method according to claim 1, wherein: the improved MobileNet V3-Small network has a wavelet convolution layer, 5 bottleeck layers and two full-connection layers; convolution kernels in the first layer of convolution layers are constructed using Daubechies wavelet function structures.
9. The wavelet improved MobileNet network based bearing fault edge diagnosis method according to claim 8, wherein: the improved MobileNet V3-Small network comprises the following specific parameters:
a first layer: wavelet convolution layer: filter 27, kernel _ size 55, threads 1, padding same as each other as same as each other as same;
a second layer: bottleeck layer: filter 16, kernel _ size 3, strings 16, padding same as same, activation same as ReLu, pool same as MaxPooling, pool _ size 2, input _ shape 2048 × 27, and output _ shape 64 × 16;
and a third layer: bottleeck layer: filter 32, kernel _ size 3, threads 1, padding same as same, activation same as ReLu, pool same as MaxPooling, pool _ size 2, input _ shape 64 × 16output _ shape 32 × 32;
a fourth layer: bottleeck layer: filter 64, kernel _ size 3, threads 1, padding same as each other as same as each other as same;
and a fifth layer: bottleeck layer: filter 64, kernel _ size 5, threads 1, padding same as same, activation same as h-shock, pool same as MaxPooling, pool _ size 2, input _ shape 16 × 64, and output _ shape 8 × 64;
a sixth layer: bottleeck layer: filter 64, kernel _ size 5, threads 1, padding same as same, activation same as h-swish, pool same as MaxPooling, pool _ size 2, input _ shape 8 × 64, and output _ shape 4 × 64;
a seventh layer: full connection layer: unity is 100; activation ═ ReLu';
an eighth layer: full connection layer: unity is 4 and activation is 'softmax'.
10. A bearing fault edge diagnosis system based on wavelet improved MobileNet network is characterized in that: the method comprises the following steps:
the acquisition and release module is used for releasing the vibration data of the acquired bearing;
the model training module is used for inputting the collected historical vibration data into an improved MobileNet V3-Small network for training to obtain a diagnosis model;
and the fault diagnosis module is used for inputting the vibration data acquired in real time into the diagnosis model for fault diagnosis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111159275.6A CN113865870A (en) | 2021-09-30 | 2021-09-30 | Bearing fault edge diagnosis method and system based on wavelet improved MobileNet network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111159275.6A CN113865870A (en) | 2021-09-30 | 2021-09-30 | Bearing fault edge diagnosis method and system based on wavelet improved MobileNet network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113865870A true CN113865870A (en) | 2021-12-31 |
Family
ID=79001112
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111159275.6A Pending CN113865870A (en) | 2021-09-30 | 2021-09-30 | Bearing fault edge diagnosis method and system based on wavelet improved MobileNet network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113865870A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210073692A1 (en) * | 2016-06-12 | 2021-03-11 | Green Grid Inc. | Method and system for utility infrastructure condition monitoring, detection and response |
| CN113052062A (en) * | 2021-03-23 | 2021-06-29 | 杭州电子科技大学上虞科学与工程研究院有限公司 | Bearing fault diagnosis method based on one-dimensional convolutional neural network |
| CN113240671A (en) * | 2021-06-16 | 2021-08-10 | 重庆科技学院 | Water turbine runner blade defect detection method based on YoloV4-Lite network |
| CN113313679A (en) * | 2021-05-21 | 2021-08-27 | 浙江大学 | Bearing surface defect detection method based on multi-source domain depth migration multi-light source integration |
-
2021
- 2021-09-30 CN CN202111159275.6A patent/CN113865870A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210073692A1 (en) * | 2016-06-12 | 2021-03-11 | Green Grid Inc. | Method and system for utility infrastructure condition monitoring, detection and response |
| CN113052062A (en) * | 2021-03-23 | 2021-06-29 | 杭州电子科技大学上虞科学与工程研究院有限公司 | Bearing fault diagnosis method based on one-dimensional convolutional neural network |
| CN113313679A (en) * | 2021-05-21 | 2021-08-27 | 浙江大学 | Bearing surface defect detection method based on multi-source domain depth migration multi-light source integration |
| CN113240671A (en) * | 2021-06-16 | 2021-08-10 | 重庆科技学院 | Water turbine runner blade defect detection method based on YoloV4-Lite network |
Non-Patent Citations (2)
| Title |
|---|
| MENGLIANG LIAO ET AL.: "Research on a Rolling Bearing Fault Detection Method With Wavelet Convolution Deep Transfer Learning", 《ACCESS》 * |
| 周国宪 等: "基于cRIO的滚动轴承故障诊断系统设计", 《昆明理工大学学报(自然科学版)》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112084974B (en) | Multi-label rolling bearing fault diagnosis method based on meta-learning | |
| Wang et al. | A new intelligent bearing fault diagnosis method using SDP representation and SE-CNN | |
| Han et al. | An enhanced convolutional neural network with enlarged receptive fields for fault diagnosis of planetary gearboxes | |
| CN112417954B (en) | Bearing fault mode diagnosis method and system for small sample data set | |
| CN112001306A (en) | Electroencephalogram signal decoding method for generating neural network based on deep convolution countermeasure | |
| CN112418013B (en) | Complex working condition bearing fault diagnosis method based on meta-learning under small sample | |
| Gao et al. | ASM1D-GAN: An intelligent fault diagnosis method based on assembled 1D convolutional neural network and generative adversarial networks | |
| CN112507785A (en) | Rolling bearing fault analysis based on CNN and LSTM | |
| CN111680642A (en) | Terrain classification method and device | |
| CN114861740B (en) | Self-adaptive mechanical fault diagnosis method and system based on multi-head attention mechanism | |
| CN114937461A (en) | Live pig sound event detection method and device based on channel attention and residual gating convolution | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| CN112926728A (en) | Small sample turn-to-turn short circuit fault diagnosis method for permanent magnet synchronous motor | |
| CN113988123A (en) | Electroencephalogram fatigue prediction method based on self-weighted increment RVFL network | |
| CN115114965A (en) | Wind turbine generator gearbox fault diagnosis model, method, equipment and storage medium | |
| CN116561517A (en) | Unbalanced intelligent fault quantitative diagnosis method based on reward optimization deep reinforcement learning | |
| CN111803028A (en) | Individualized epileptic seizure detection device based on network parameter migration | |
| Zhang et al. | Intelligent fault diagnosis using image representation of multi-domain features | |
| CN117312962A (en) | Long-service elevator guide rail fault diagnosis method based on transfer learning and data driving | |
| CN113865870A (en) | Bearing fault edge diagnosis method and system based on wavelet improved MobileNet network | |
| CN116399592A (en) | Bearing fault diagnosis method based on channel attention dual-path feature extraction | |
| CN115251909B (en) | Method and device for evaluating hearing by electroencephalogram signals based on space-time convolutional neural network | |
| CN111539381A (en) | Construction method of wind turbine bearing fault classification diagnosis model | |
| Pan et al. | Intelligent fault diagnosis of rolling bearing via deep-layerwise feature extraction using deep belief network | |
| CN115905827A (en) | Steam turbine fault diagnosis method and device based on neural network |
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 |