CN114969994B - RV reducer fault diagnosis method and system based on single-measuring-point multidirectional data fusion - Google Patents
RV reducer fault diagnosis method and system based on single-measuring-point multidirectional data fusion Download PDFInfo
- Publication number
- CN114969994B CN114969994B CN202111317295.1A CN202111317295A CN114969994B CN 114969994 B CN114969994 B CN 114969994B CN 202111317295 A CN202111317295 A CN 202111317295A CN 114969994 B CN114969994 B CN 114969994B
- Authority
- CN
- China
- Prior art keywords
- model
- data
- size
- fusion
- constructed
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000004927 fusion Effects 0.000 title claims abstract description 54
- 238000003745 diagnosis Methods 0.000 title claims abstract description 49
- 239000003638 chemical reducing agent Substances 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 25
- 230000001133 acceleration Effects 0.000 claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 40
- 238000011176 pooling Methods 0.000 claims abstract description 24
- 238000012360 testing method Methods 0.000 claims abstract description 17
- 238000005070 sampling Methods 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 6
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 238000005259 measurement Methods 0.000 claims 1
- 238000000605 extraction Methods 0.000 description 4
- 238000009434 installation Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000008054 signal transmission Effects 0.000 description 3
- 238000013499 data model Methods 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000005299 abrasion Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Geometry (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computer Hardware Design (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a RV reducer fault diagnosis method and system based on single-measuring-point multidirectional data fusion, wherein the method comprises the following steps: synchronously acquiring acceleration signals of the RV reducer in three directions by using a triaxial acceleration sensor; performing data fusion on the three-dimensional acceleration signals by using an energy weighting method to obtain one-dimensional fusion data containing three-dimensional time sequence information; dividing one-dimensional fusion data into a training data set and a test data set; constructing a wavelet convolution energy pooling network model for RV reducer fault diagnosis by modifying the network structure of a classical deep convolution model; training the network model by using the training data set to obtain a diagnosis model; the test dataset is used to verify the correctness of the diagnostic model. The RV reducer fault diagnosis method has the advantages that only a test position is selected on the RV reducer at random, vibration signals of the position in three directions are collected, a diagnosis model is built through a data fusion training wavelet convolution energy pooling network, and the RV reducer fault diagnosis can be achieved.
Description
Technical Field
The invention relates to a fault diagnosis method and system for an RV reducer based on single-measuring-point multidirectional data fusion, and belongs to the technical field of rotary machine fault diagnosis.
Background
Due to the compact and complex structure, time-varying working condition and uncertain transmission path of the RV reducer, the running state of equipment cannot be comprehensively represented by data acquired by a single vibration sensor, and the synchronism of channel acquisition and the information representation of different channels to measuring points are different by adopting a method of acquiring a plurality of vibration sensors.
Because the RV reducer is complex in internal structure and time-varying in working condition, the traditional signal processing methods, such as short-time Fourier transform (Fourier transform, FT), empirical mode decomposition (EMPIRICAL MODE DECOMPOSITION, EMD) and the like, are difficult to effectively excavate fault characteristics, and the running state of the RV reducer cannot be determined; the time-frequency image fault diagnosis based on the classical CNN framework obtains good results, but is limited in that one-dimensional acceleration signals cannot be processed, and the extraction of the characteristics depends on the learning capability of the network and lacks prior knowledge of RV fault diagnosis.
Disclosure of Invention
The invention provides a fault diagnosis method and system for an RV reducer based on single-measuring-point multi-directional data fusion, and the fault diagnosis of the RV reducer is realized through the single-measuring-point multi-directional data fusion.
The technical scheme of the invention is as follows: a RV reducer fault diagnosis method based on single-measuring-point multidirectional data fusion comprises the following steps of 1: and (3) data acquisition: synchronously acquiring acceleration signals of the RV reducer in three directions by using a triaxial acceleration sensor;
step2: data fusion: based on the three-dimensional acceleration signals acquired by the Setp2, performing data fusion on the three-dimensional acceleration signals by using an energy weighting method to obtain one-dimensional fusion data containing three-dimensional time sequence information;
Step3: data set partitioning: dividing one-dimensional fusion data obtained based on Step3 into a training data set and a test data set;
Step4: and (3) constructing a network model: based on a classical depth convolution model, constructing a wavelet convolution energy pooling network model for RV reducer fault diagnosis by modifying the network structure of the classical depth convolution model;
step5: model training: training the network model by using the training data set to obtain a diagnosis model;
Step6: diagnosis and verification: the test dataset is used to verify the correctness of the diagnostic model.
And the triaxial acceleration sensor is arranged at a random position of the outer ring of the RV reducer.
The data fusion specifically comprises the following steps:
An acceleration sensor is set to collect acceleration signals of the RV reducer in the s direction within a sampling time t at a set sampling frequency, data values X s1,xs2,…,xsk of k discrete time points in the s direction are obtained, and an energy value X s in the direction within the time t is obtained according to the discrete data values; s=x, y, z;
determining an energy value coefficient lambda s of the acceleration signals in all directions;
Fusing data in three directions at the same time, and obtaining fused one-dimensional data by a data value x q＝∑λs*xsq of a q-th time point after fusion; wherein q=1, 2, …, k; x sq represents the data value of the q-th time point in the s direction, and x q represents the data value of the q-th time point after the three directions are fused.
The energy value is calculated as follows:
where x si represents the data value at the i-th time point in the s-direction.
The s-direction energy value coefficientWherein X x,Xy,Xz represents the energy values in the X, y and z directions respectively.
The wavelet convolution energy pooling network model is constructed by the following steps:
step4.1: the one-dimensional CNN structure is adopted as a model backbone, and the model comprises five layers of convolution layers and three layers of full-connection layers, wherein the arrangement structure comprises: conv1, conv2, conv3, conv4, conv5, full1, full2, full3;
Step4.2: the arrangement in Conv1 layer of the model is constructed as: the wavelets conv1, relu1, LPPool1; where wavelet conv1 represents wavelet convolution with size [1,27,55], LPPool D represents energy pooling with size [16,1], step size 55;
Step4.3: the arrangement in Conv2 layer of the model is constructed as: conv2, relu2, dropout2; wherein, the conv2 size is [27,27,55], and the Dropout parameter is 0.5;
Step4.4: the arrangement in Conv3 layer of the model is constructed as: conv3, relu3, maxPool2; wherein, the conv3 size is [27,27,55], the MaxPool2 size is [16,1], and the step length is 55;
Step4.5: the arrangement in Conv4 layer of the model is constructed as: conv4, relu4, dropout4; wherein, the conv4 size is [27,27,55], and the Dropout parameter is 0.5;
Step4.6: the arrangement in Conv5 layer of the model is constructed as: conv5, relu; wherein, the conv5 size is [27,27,55];
Step4.7: the arrangement in Full1 layer of the model is constructed as: linear1, relu, 6; wherein, the size of the Linear1 is [216,216];
step4.8: the arrangement in the Full2 layer of the model is constructed as: linear2, relu; wherein, the size of the Linear2 is [216,64];
Step4.9: the arrangement in Full3 layer of the model is constructed as: linear3, softmax; wherein the size of the Linear3 is [64,6].
The invention provides a RV reducer fault diagnosis system based on single-measuring-point multidirectional data fusion, which comprises the following components:
the data acquisition module is used for synchronously acquiring acceleration signals of the RV reducer in three directions by using a triaxial acceleration sensor;
the data fusion module is used for carrying out data fusion on the three-dimensional acceleration signals by using an energy weighting method based on the three-dimensional acceleration signals acquired by the data acquisition module to obtain one-dimensional fusion data containing three-dimensional time sequence information;
The data set dividing module is used for dividing the one-dimensional fusion data obtained based on the data fusion module into a training data set and a test data set;
the network model construction module is used for constructing a wavelet convolution energy pooling network model for RV reducer fault diagnosis by modifying the network structure of the classical depth convolution model based on the classical depth convolution model;
the model training module is used for training the network model by using the training data set to obtain a diagnosis model;
and the diagnosis verification module is used for verifying the correctness of the diagnosis model by using the test data set.
The beneficial effects of the invention are as follows: the invention has the advantages that the installation position of the sensor of the RV reducer is not restricted, the test position is selected randomly on the RV reducer, vibration signals of the position in three directions are collected, and the fault diagnosis of the RV reducer can be effectively realized by establishing a diagnosis model through data fusion training wavelet convolution energy pooling network.
Drawings
FIG. 1 is a system flow diagram of the present application;
FIG. 2 is a comparison of the same signal analysis results: (a) wavelet transformation; (b) wavelet convolution; (c) classical convolution;
FIG. 3 is a graph comparing training effects of a single channel+classical convolution model, a single channel+wavelet convolution energy pooling model, a data fusion+wavelet convolution energy pooling model (invention). .
Detailed Description
The invention will be further described with reference to the drawings and examples, but the invention is not limited to the scope.
Example 1: 1-3, a RV reducer fault diagnosis method based on single-measuring-point multi-directional data fusion comprises the following steps:
step1: and (3) data acquisition: synchronously acquiring acceleration signals of the RV reducer in three directions by using a triaxial acceleration sensor;
step2: data fusion: based on the three-dimensional acceleration signals acquired by the Setp2, performing data fusion on the three-dimensional acceleration signals by using an energy weighting method to obtain one-dimensional fusion data containing three-dimensional time sequence information;
Step3: data set partitioning: dividing one-dimensional fusion data obtained based on Step3 into a training data set and a test data set;
Step4: and (3) constructing a network model: based on a classical depth convolution model, constructing a wavelet convolution energy pooling network model for RV reducer fault diagnosis by modifying the network structure of the classical depth convolution model;
step5: model training: training the network model by using the training data set to obtain a diagnosis model;
Step6: diagnosis and verification: the test dataset is used to verify the correctness of the diagnostic model.
Alternatively, the single station installation is specifically: the triaxial acceleration sensor is arranged at a random position of the outer ring of the RV reducer; the installation mode does not restrict the installation position of the vibration sensor of the RV reducer, and one triaxial acceleration sensor synchronously collects acceleration signals in three directions of the RV reducer, so that signals in all directions are comprehensively collected, the problem that the sensor is not fully used for acquiring fault information due to uncertain signal transmission paths is solved, and meanwhile, the synchronism of the signals is ensured. The three-channel data acquired by the triaxial acceleration sensor are subjected to data fusion in an energy weighting mode, three-dimensional time sequence vibration information of measuring points is extracted, the speed and accuracy of diagnosis model training are greatly improved while fault information is complete, and the problems that the RV reducer is complex in structure, uncertain in signal transmission path, low in diagnosis accuracy of unidirectional data model training, large in data quantity of the multidirectional data model and poor in diagnosis effect due to long model training time are solved.
Optionally, the data fusion specifically includes:
An acceleration sensor is set to collect acceleration signals of the RV reducer in the s direction within a sampling time t at a set sampling frequency, data values X s1,xs2,…,xsk of k discrete time points in the s direction are obtained, and an energy value X s in the direction within the time t is obtained according to the discrete data values; s=x, y, z;
determining an energy value coefficient lambda s of the acceleration signals in all directions;
Fusing data in three directions at the same time, and obtaining fused one-dimensional data by a data value x q＝∑λs*xsq of a q-th time point after fusion; wherein q=1, 2, …, k; x sq represents the data value of the q-th time point in the s direction, and x q represents the data value of the q-th time point after the three directions are fused.
Optionally, the energy value is calculated as follows:
where x si represents the data value at the i-th time point in the s-direction.
Optionally, the s-direction energy value coefficientWherein X x,Xy,Xz represents the energy values in the X, y and z directions respectively.
Optionally, the constructing step of the wavelet convolution energy pooling network model is as follows:
step4.1: the one-dimensional CNN structure is adopted as a model backbone, and the model comprises five layers of convolution layers and three layers of full-connection layers, wherein the arrangement structure comprises: conv1, conv2, conv3, conv4, conv5, full1, full2, full3;
Step4.2: the arrangement in Conv1 layer of the model is constructed as: the wavelets conv1, relu1, LPPool1; where wavelet conv1 represents wavelet convolution with size [1,27,55], LPPool D represents energy pooling with size [16,1], step size 55;
Step4.3: the arrangement in Conv2 layer of the model is constructed as: conv2, relu2, dropout2; wherein, the conv2 size is [27,27,55], and the Dropout parameter is 0.5;
Step4.4: the arrangement in Conv3 layer of the model is constructed as: conv3, relu3, maxPool2; wherein, the conv3 size is [27,27,55], the MaxPool2 size is [16,1], and the step length is 55;
Step4.5: the arrangement in Conv4 layer of the model is constructed as: conv4, relu4, dropout4; wherein, the conv4 size is [27,27,55], and the Dropout parameter is 0.5;
Step4.6: the arrangement in Conv5 layer of the model is constructed as: conv5, relu; wherein, the conv5 size is [27,27,55];
Step4.7: the arrangement in Full1 layer of the model is constructed as: linear1, relu, 6; wherein, the size of the Linear1 is [216,216];
step4.8: the arrangement in the Full2 layer of the model is constructed as: linear2, relu; wherein, the size of the Linear2 is [216,64];
Step4.9: the arrangement in Full3 layer of the model is constructed as: linear3, softmax; wherein the size of the Linear3 is [64,6].
Specifically, the wavelet convolution is a wavelet Conv1 of a Conv1 layer, the size of the wavelet Conv1 is [1,27,55], the size means that the wavelet Conv1 shares 27 one-dimensional convolution check signals with the length of 55 for convolution; meanwhile, the first layer convolution kernel parameter can allow gradient update learning and can adaptively adjust wavelet base parameters, so that the wavelet convolution energy pooling network can adaptively extract frequency band information of vibration signals of the RV reducer, an end-to-end self-adaptive diagnosis mode is realized, data preprocessing is not needed, and the capability of extracting fault characteristics from RV reducer signals under complex working conditions of the network to realize fault diagnosis is improved; according to the application, parameters of wavelet convolution are directly assigned to a convolution kernel of a first layer, frequency domain information of signals is obtained through signal and wavelet convolution, the direction of network data mining is determined through wavelet analysis, the subsequent network structure further performs feature mining on frequency band information obtained by wavelet conv1, uncertainty of feature extraction of the traditional random convolution kernel (such as uncertainty caused by the traditional convolution kernel, initial parameters are uncertainty caused by random assignment, uncertainty of results obtained by convolution of signals is avoided), and RV fault feature mining efficiency is improved. The energy pooling is LPPool D of Conv1 layer, the root mean square value is a statistical characteristic index describing signal energy, and the method has the characteristics of good stability and repeatability in judging whether the running state of equipment is normal, diagnosing faults such as abrasion of parts of the equipment and the like. The application takes 2 on the norm_type parameter of LPPool D, the function of which is equivalent to the effective value, the wavelet conv1 obtains the frequency band information, and LPPool D further calculates each frequency band energy obtained in the wavelet conv1, thereby realizing that the network digs RV fault characteristics in the frequency band energy of the signal and improving the accuracy of network diagnosis.
As can be seen from fig. 2 (c), the analysis result of the random convolution kernel is disordered in the whole frequency domain, and the data features are irregularly distributed and have poor interpretation, so that the knowledge migration is not facilitated; the frequency band in which the energy is concentrated in fig. 2 (a) is consistent with the result of the wavelet convolution in fig. (b), which demonstrates that applying wavelet transformation to the wavelet convolution can effectively analyze the frequency domain information of the vibration signal.
Model training, namely importing a one-dimensional fusion data set into a network by using a tensorflow pd.read_csv module; the StratifiedShuffleSplit () function in the sklearn. Model_selection module is called to divide the dataset into a training dataset and a test dataset. The training data set accounts for 80% of the whole data set, the test data set accounts for 20% of the whole data set, and the model has good training effect under the dividing proportion through experimental verification.
Further setting the initial learning rate to be 0.001 and the attenuation rate to be 0.99; training data length is 4096; the training process adopts mini_batch learning, and the size is set to 128; when the precision of the test set reaches 95%, training is completed, and a fault diagnosis model of the RV reducer is obtained, so that fault diagnosis of the RV reducer is realized.
In the figure 3, three curves are shown, a double-line is a single-channel and classical convolution model, and a dash-dot line is a single-channel and wavelet convolution energy pooling model; the solid line is a data fusion+wavelet convolution energy pooling model (the invention); by comparing the single channel and classical convolution models with the single channel and wavelet convolution energy pooling model, the training accuracy of the wavelet convolution energy pooling model under the same 45 times of model training can be seen to be obviously higher than that of the classical CNN model, but the training accuracy is not remarkably improved to be stable at about 0.75, because the RV signal transmission path is complex, the data in a single direction contains incomplete fault information, the method provided by the invention not only fuses the data in the RV multi-direction, ensures the integrity of the fault information, but also greatly improves the feature extraction efficiency of the fault, and can be seen from fig. 3 that the accuracy is smooth and approaches to 1 when the training times reach 20 times, so that the method is far better than that of the CNN model in the RV fault feature extraction, and the effectiveness of the data fusion is verified.
While the present invention has been described in detail with reference to the drawings, the present invention is not limited to the above 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 (5)
1. A RV reducer fault diagnosis method based on single-measuring-point multidirectional data fusion is characterized in that: the method comprises the following steps:
step1: and (3) data acquisition: synchronously acquiring acceleration signals of the RV reducer in three directions by using a triaxial acceleration sensor;
Step2: data fusion: based on the three-dimensional acceleration signals acquired by the Setp1, performing data fusion on the three-dimensional acceleration signals by using an energy weighting method to obtain one-dimensional fusion data containing three-dimensional time sequence information;
step3: data set partitioning: dividing one-dimensional fusion data obtained based on Step2 into a training data set and a test data set;
Step4: and (3) constructing a network model: based on a classical depth convolution model, constructing a wavelet convolution energy pooling network model for RV reducer fault diagnosis by modifying the network structure of the classical depth convolution model;
step5: model training: training the network model by using the training data set to obtain a diagnosis model;
step6: diagnosis and verification: verifying the correctness of the diagnostic model using the test dataset;
the data fusion specifically comprises the following steps:
An acceleration sensor is set to collect acceleration signals of the RV reducer in the s direction within a sampling time t at a set sampling frequency, data values X s1,xs2,…,xsk of k discrete time points in the s direction are obtained, and an energy value X s in the direction within the time t is obtained according to the discrete data values; s=x, y, z;
determining an energy value coefficient lambda s of the acceleration signals in all directions;
Fusing data in three directions at the same time, and obtaining fused one-dimensional data by a data value x q＝∑λs*xsq of a q-th time point after fusion; wherein q=1, 2, …, k; x sq represents the data value of the q-th time point in the s direction, and x q represents the data value of the q-th time point after the three directions are fused;
the wavelet convolution energy pooling network model is constructed by the following steps:
step4.1: the one-dimensional CNN structure is adopted as a model backbone, and the model comprises five layers of convolution layers and three layers of full-connection layers, wherein the arrangement structure comprises: conv1, conv2, conv3, conv4, conv5, full1, full2, full3;
Step4.2: the arrangement in Conv1 layer of the model is constructed as: waveletconv1, relu1, LPPool D; where wavelet conv1 represents wavelet convolution with size [1,27,55], LPPool D represents energy pooling, and root mean square value is a statistical feature index describing signal energy with size [16,1], step size 55;
Step4.3: the arrangement in Conv2 layer of the model is constructed as: conv2, relu2, dropout2; wherein, the conv2 size is [27,27,55], and the Dropout parameter is 0.5;
Step4.4: the arrangement in Conv3 layer of the model is constructed as: conv3, relu3, maxPool2; wherein, the conv3 size is [27,27,55], the MaxPool2 size is [16,1], and the step length is 55;
Step4.5: the arrangement in Conv4 layer of the model is constructed as: conv4, relu4, dropout4; wherein, the conv4 size is [27,27,55], and the Dropout parameter is 0.5;
Step4.6: the arrangement in Conv5 layer of the model is constructed as: conv5, relu; wherein, the conv5 size is [27,27,55];
Step4.7: the arrangement in Full1 layer of the model is constructed as: linear1, relu, 6; wherein, the size of the Linear1 is [216,216];
step4.8: the arrangement in the Full2 layer of the model is constructed as: linear2, relu; wherein, the size of the Linear2 is [216,64];
Step4.9: the arrangement in Full3 layer of the model is constructed as: linear3, softmax; wherein the size of the Linear3 is [64,6].
2. The RV retarder fault diagnosis method based on single-station multi-directional data fusion of claim 1 wherein: and the triaxial acceleration sensor is arranged at a random position of the outer ring of the RV reducer.
3. The RV retarder fault diagnosis method based on single-station multi-directional data fusion of claim 1 wherein: the energy value is calculated as follows:
where x si represents the data value at the i-th time point in the s-direction.
4. The RV retarder fault diagnosis method based on single-station multi-directional data fusion of claim 1 wherein: the s-direction energy value coefficientWherein X x,Xy,Xz represents the energy values in the X, y and z directions respectively.
5. RV reduction gear fault diagnosis system based on single measurement point multidirectional data fusion, its characterized in that: comprising the following steps:
the data acquisition module is used for synchronously acquiring acceleration signals of the RV reducer in three directions by using a triaxial acceleration sensor;
the data fusion module is used for carrying out data fusion on the three-dimensional acceleration signals by using an energy weighting method based on the three-dimensional acceleration signals acquired by the data acquisition module to obtain one-dimensional fusion data containing three-dimensional time sequence information;
The data set dividing module is used for dividing the one-dimensional fusion data obtained based on the data fusion module into a training data set and a test data set;
the network model construction module is used for constructing a wavelet convolution energy pooling network model for RV reducer fault diagnosis by modifying the network structure of the classical depth convolution model based on the classical depth convolution model;
the model training module is used for training the network model by using the training data set to obtain a diagnosis model;
The diagnosis verification module is used for verifying the correctness of the diagnosis model by using the test data set;
the data fusion specifically comprises the following steps:
An acceleration sensor is set to collect acceleration signals of the RV reducer in the s direction within a sampling time t at a set sampling frequency, data values X s1,xs2,…,xsk of k discrete time points in the s direction are obtained, and an energy value X s in the direction within the time t is obtained according to the discrete data values; s=x, y, z;
determining an energy value coefficient lambda s of the acceleration signals in all directions;
Fusing data in three directions at the same time, and obtaining fused one-dimensional data by a data value x q＝∑λs*xsq of a q-th time point after fusion; wherein q=1, 2, …, k; x sq represents the data value of the q-th time point in the s direction, and x q represents the data value of the q-th time point after the three directions are fused;
the wavelet convolution energy pooling network model is constructed by the following steps:
step4.1: the one-dimensional CNN structure is adopted as a model backbone, and the model comprises five layers of convolution layers and three layers of full-connection layers, wherein the arrangement structure comprises: conv1, conv2, conv3, conv4, conv5, full1, full2, full3;
Step4.2: the arrangement in Conv1 layer of the model is constructed as: waveletconv1, relu1, LPPool D; where wavelet conv1 represents wavelet convolution with size [1,27,55], LPPool D represents energy pooling, and root mean square value is a statistical feature index describing signal energy with size [16,1], step size 55;
Step4.3: the arrangement in Conv2 layer of the model is constructed as: conv2, relu2, dropout2; wherein, the conv2 size is [27,27,55], and the Dropout parameter is 0.5;
Step4.4: the arrangement in Conv3 layer of the model is constructed as: conv3, relu3, maxPool2; wherein, the conv3 size is [27,27,55], the MaxPool2 size is [16,1], and the step length is 55;
Step4.5: the arrangement in Conv4 layer of the model is constructed as: conv4, relu4, dropout4; wherein, the conv4 size is [27,27,55], and the Dropout parameter is 0.5;
Step4.6: the arrangement in Conv5 layer of the model is constructed as: conv5, relu; wherein, the conv5 size is [27,27,55];
Step4.7: the arrangement in Full1 layer of the model is constructed as: linear1, relu, 6; wherein, the size of the Linear1 is [216,216];
step4.8: the arrangement in the Full2 layer of the model is constructed as: linear2, relu; wherein, the size of the Linear2 is [216,64];
Step4.9: the arrangement in Full3 layer of the model is constructed as: linear3, softmax; wherein the size of the Linear3 is [64,6].
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111317295.1A CN114969994B (en) | 2021-11-09 | 2021-11-09 | RV reducer fault diagnosis method and system based on single-measuring-point multidirectional data fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111317295.1A CN114969994B (en) | 2021-11-09 | 2021-11-09 | RV reducer fault diagnosis method and system based on single-measuring-point multidirectional data fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114969994A CN114969994A (en) | 2022-08-30 |
| CN114969994B true CN114969994B (en) | 2024-07-26 |
Family
ID=82974540
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111317295.1A Active CN114969994B (en) | 2021-11-09 | 2021-11-09 | RV reducer fault diagnosis method and system based on single-measuring-point multidirectional data fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114969994B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117992863B (en) * | 2024-04-07 | 2024-06-18 | 安徽大学 | Rotary machine fault diagnosis method based on interpretable stationary wavelet packet convolution network |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113361372A (en) * | 2021-06-02 | 2021-09-07 | 长江大学 | Main reducer multi-fault intelligent diagnosis method based on multi-channel data deep mining |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6609217B1 (en) * | 1998-03-30 | 2003-08-19 | General Electric Company | System and method for diagnosing and validating a machine over a network using waveform data |
| US9104650B2 (en) * | 2005-07-11 | 2015-08-11 | Brooks Automation, Inc. | Intelligent condition monitoring and fault diagnostic system for preventative maintenance |
| CA2874991A1 (en) * | 2012-05-31 | 2013-12-05 | University Of Connecticut | Methods and apparatuses for defect diagnosis in a mechanical system |
| CN110334740A (en) * | 2019-06-05 | 2019-10-15 | 武汉大学 | The electrical equipment fault of artificial intelligence reasoning fusion detects localization method |
| CN113469060A (en) * | 2021-07-02 | 2021-10-01 | 浙大城市学院 | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method |
-
2021
- 2021-11-09 CN CN202111317295.1A patent/CN114969994B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113361372A (en) * | 2021-06-02 | 2021-09-07 | 长江大学 | Main reducer multi-fault intelligent diagnosis method based on multi-channel data deep mining |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114969994A (en) | 2022-08-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108805269B (en) | Method for picking seismic facies arrival time based on LSTM recurrent neural network | |
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN110674604A (en) | Transformer DGA data prediction method based on multi-dimensional time sequence frame convolution LSTM | |
| CN110657985B (en) | Gearbox fault diagnosis method and system based on singular value spectrum manifold analysis | |
| CN111126471A (en) | Microseism event detection method and system | |
| CN113865859B (en) | Gear box state fault diagnosis method for multi-scale multi-source heterogeneous information fusion | |
| CN112200244A (en) | Intelligent detection method for anomaly of aerospace engine based on hierarchical countermeasure training | |
| CN107909118A (en) | A kind of power distribution network operating mode recording sorting technique based on deep neural network | |
| CN106198000A (en) | A kind of rocker arm of coal mining machine gear failure diagnosing method | |
| CN103941254A (en) | Soil physical property classification recognition method and device based on geological radar | |
| CN114048769A (en) | Multi-source multi-domain information entropy fusion and model self-optimization method for bearing fault diagnosis | |
| CN110632662A (en) | Algorithm for automatically identifying microseism signals by using DCNN-inclusion network | |
| CN114969994B (en) | RV reducer fault diagnosis method and system based on single-measuring-point multidirectional data fusion | |
| CN115586406A (en) | GIS partial discharge fault diagnosis method and system based on ultrahigh frequency signal | |
| CN114237046A (en) | Partial discharge pattern recognition method based on SIFT data feature extraction algorithm and BP neural network model | |
| CN114818815B (en) | Method and system for acquiring tip timing arrival time for blade vibration measurement | |
| CN115481666A (en) | Fault diagnosis method, system and equipment for small sample of gear box | |
| CN115826042A (en) | Edge-cloud-combined distributed seismic data processing method and device | |
| CN103267652A (en) | Intelligent online diagnosis method for early failures of equipment | |
| CN114330120A (en) | 24-hour PM prediction based on deep neural network2.5Method of concentration | |
| CN113111706B (en) | SAR target feature unwrapping and identifying method for azimuth continuous deletion | |
| CN110160778A (en) | Gearbox fault state identification method based on sequential hypothesis testing | |
| CN117349640A (en) | Vehicle-mounted vibration acceleration data analysis method and system | |
| CN117454100A (en) | Missing data reconstruction method for engineering structure | |
| CN115356599B (en) | Multi-mode urban power grid fault diagnosis method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |