CN116358871B - Rolling bearing weak signal composite fault diagnosis method based on graph rolling network - Google Patents

Rolling bearing weak signal composite fault diagnosis method based on graph rolling network Download PDF

Info

Publication number
CN116358871B
CN116358871B CN202310321870.8A CN202310321870A CN116358871B CN 116358871 B CN116358871 B CN 116358871B CN 202310321870 A CN202310321870 A CN 202310321870A CN 116358871 B CN116358871 B CN 116358871B
Authority
CN
China
Prior art keywords
fault
graph
fault diagnosis
entropy
matrix
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
Application number
CN202310321870.8A
Other languages
Chinese (zh)
Other versions
CN116358871A (en
Inventor
王亚萍
高圣延
许迪
侯德康
葛江华
张祺松
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.)
Harbin University of Science and Technology
Original Assignee
Harbin University of Science and Technology
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 Harbin University of Science and Technology filed Critical Harbin University of Science and Technology
Priority to CN202310321870.8A priority Critical patent/CN116358871B/en
Publication of CN116358871A publication Critical patent/CN116358871A/en
Application granted granted Critical
Publication of CN116358871B publication Critical patent/CN116358871B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01MTESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00Testing of machine parts
    • G01M13/04Bearings
    • G01M13/045Acoustic or vibration analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/241Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/0464Convolutional networks [CNN, ConvNet]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • YGENERAL 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
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02TCLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00Enabling 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)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Acoustics & Sound (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a rolling bearing weak signal composite fault diagnosis method of a graph rolling network, which adopts a multi-scale entropy method, signals are segmented into N sections through a sliding window, then the time domain signals of each section are subjected to fine composite multi-scale entropy feature extraction, and different weights are defined for adjacent matrixes in different forms by using a Gaussian kernel function. To achieve fault diagnosis at the graph level, a representation of the entire graph is obtained using a GCN with a pooling layer and a readout layer, upon which standard machine learning techniques can be performed. Experimental verification shows that compared with a node construction method of a time domain or a frequency domain, the fine composite multi-scale entropy considers the cross correlation of all entropy values to evaluate the dynamic conditions in fault detection, and the edge connection mode enables fault information to be transmitted and updated only among fault nodes of the same type, so that the composite fault type of the rolling bearing can be effectively identified, and the fault diagnosis efficiency is improved.

Description

Rolling bearing weak signal composite fault diagnosis method based on graph rolling network
Technical Field
The invention belongs to the technical field of fault diagnosis of rotary machinery, relates to a rolling bearing fault diagnosis method, and in particular relates to a rolling bearing weak signal compound fault diagnosis method based on a graph rolling network.
Background
Rolling bearings are used as key components in large rotary machinery equipment, and are widely used in mechanical structures such as compressors, fans, turbines, generators, gas turbines, aero-generators, and various motors. Slight faults of key parts such as rolling bearings and the like can indirectly influence the operation of a system, cause a series of chain reactions, and further cause equipment performance attenuation so as to cause system-level faults. Predictive and health management (Prognostics and Health Management, PHM) aims to monitor and analyze the health status of diagnostic devices through data and predict the occurrence of faults, thereby greatly improving the efficiency of status maintenance. Intelligent diagnostics and prognostics have been widely used for monitoring rotating machinery, such as aircraft engines, wind turbines, helicopters, high speed trains, and the like, as two key components of PHM systems. In mechanical equipment, a composite fault is a typical fault, and a machine learning algorithm is also widely applied to intelligent diagnosis of composite faults of mechanical equipment, and intensive research is conducted. The intelligent diagnosis method of the composite fault can be developed from two aspects of machine learning and deep learning.
The traditional machine learning method mainly comprises the following steps: k-nearest neighbor algorithm, bayes classifier, support vector machine and artificial neural network. For many years, students at home and abroad are exploring and applying the traditional machine learning method to conduct intelligent fault diagnosis on mechanical equipment. Gu Min et al combine the variational modal decomposition with permutation entropy to identify composite faults of the bearing by k-nearest neighbors. S-nearest neighbor learning and random forest combined composite fault diagnosis method is proposed by S-nearest z et al, and high-quality characteristics are screened through characteristic sequencing, so that rolling bearing and gear box composite fault diagnosis is realized. Asr et al propose a compound fault diagnosis method of non-naive bayes analysis, which can accurately identify single and compound faults of the bearing. The Tangbaoping et al propose a rotary machine multi-fault diagnosis method combining orthogonal supervision linear local cutting space and least square support vector machine, and realize fault identification of bearings. Wu Jun and the like realize intelligent diagnosis of the composite fault of the rotary machine by integrating two extreme learning machines and a classifier. However, in the face of industrial big data, the traditional machine learning model has low model complexity, the diagnosis accuracy depends on feature extraction and selection, and the effectiveness of feature extraction depends largely on expert knowledge, which limits the application of intelligent diagnosis methods in industry.
In recent years, intelligent diagnosis methods with deep learning as a core lead the hot trend of industrial innovative application in fault diagnosis and predictive maintenance. Currently, deep neural networks (Deep Neural Networks, DNNs) mainly include convolutional neural networks, automatic encoders, deep belief networks, and recurrent neural networks. Sohaib et al propose a two-dimensional convolutional neural network composite fault diagnosis model capable of accurately compounding faults of bearings under variable working conditions. Li Zhinong et al propose an intelligent fault diagnosis method for deep convolutional neural networks, which realizes bearing multi-fault recognition. Pecht et al propose an unsupervised sparse feature learning mechanism to realize the compound fault diagnosis of the planetary gear box. Shen Changqing et al propose an improved convolutional deep belief network model that enables bearing composite fault diagnosis through a multi-layer feature fusion technique. In order to overcome uncertainty generated by manual feature extraction, selection and separation, the deep learning-based compound fault diagnosis method can realize end-to-end learning from raw data to the health state of mechanical equipment. Most of them ignore the interdependencies between data.
The graph neural networks (Graph Neural Networks, GNNs) can mine relationships between nodes from irregular data, unlike conventional deep learning methods. The graph data in which the relationship between nodes is reflected on the edges of the connection and the weight of the edges reflects the strength of the relationship is intended to be modeled. Convolutions in convolutional neural networks (Convolutional Neural Networks, CNNs) also have the same function. GNNs have been widely used in the health monitoring of mechanical equipment due to their ability to model local structures and interact nodes in graph data. Typical representations of Graph neural networks are primarily Graph convolution networks (Graph Convolutional Networks, GCNs), graph recurrent neural networks (Graph Recurrent Neural Networks, GRNNs), graph Auto-encodings (GAEs), and Graph annotation force networks (Graph Attention Network, GAT). GNNs can model three tasks, including: 1) Node classification, wherein GNNs are used to obtain a representation of each node and classify the node; 2) Graph classification, using GNNs to obtain a representation of the entire graph, and then performing graph classification; 3) Edge prediction, in which GNNs are used to mine the relationship between each node and predict missing edges. Recently, GNNs have been increasingly applied to PHM by researchers as they can model the interdependencies between data and embed them into extracted features. The wavelet packet is used by et al to decompose the vibration signal of the wind turbine gearbox and obtain a graphical dataset, and then fault classification is achieved by the proposed fast depth GCNs. Plum et al convert the vibration signal of the rolling bearing into a horizontal visibility map, and then apply GNNs to model the map data and implement fault classification. Gao et al propose a rolling bearing fault diagnosis method combining a weighted level visibility graph (Weighted Horizontal Visibility Graph, WHOG) and a graph Fourier transform method, which can effectively identify that most of fault impact components have stronger anti-interference capability. Li et al use WHOG method to convert to graphic data, while weighting by the difference between sampling indices, to propose a graph convolution network in combination with WHOG for bearing failure diagnosis. Yu et al combine inherent time scale decomposition with graph signal processing to accurately and effectively identify composite faults of an aeroengine rolling bearing. Yang et al acquire the frequency spectrum of fault signals by utilizing multi-sensor data, construct a label relation diagram of an adjacent matrix, and diagnose compound faults under various working conditions by proposing a deep capsule diagram convolution network. The method comprises the steps that Zhou et al extracts singular values from multi-sensor vibration signal samples to serve as sample node representations to construct graph data, and GCNs are utilized to extract features from the graph data to perform feature fusion to achieve fault classification. Chen et al propose a GCNs fault diagnosis method combining available measured values with priori knowledge, converting the pre-diagnosis result into a correlation chart and introducing a weight coefficient into a GCNs model to adjust the influence of the measured values and the priori knowledge, so that fault diagnosis under part of fault labels can be realized. Wang et al provides a graph convolution network fault diagnosis method based on vibration indexes aiming at the problem that a large amount of data is unlabeled data in an actual scene, and verifies that the method has good application prospect for bearing fault diagnosis with less labeled data. It can be seen from a review of the literature that the main difference between DNNs-based and GNNs-based methods is the construction of the graph and the design of the model. Therefore, how to convert the original data into a graph structure and design a proper GNNs network to be suitable for solving the rolling bearing composite fault is a faced problem.
In summary, in the mechanical power transmission system, the occurrence frequency of the faults is high due to the characteristics of randomness, concurrency, secondary and the like of the faults. The traditional deep learning method mainly aims at a single fault to build a model. When a composite fault occurs, the fault characteristics of the composite fault show that the intensity distribution is uneven and mutually overlapped and mutually interfered, so that the traditional deep learning method is difficult to judge which fault type. Whether traditional machine learning or deep learning is based on relevant research of Euclidean structure data, all sample data are regarded as a whole to learn so as to achieve the purpose of classification. The algorithm based on the graph rolling network can effectively dig the correlation information among samples which cannot be utilized by the traditional deep learning method. Therefore, it is important to develop a rolling bearing compound fault diagnosis method based on a graph rolling network.
Disclosure of Invention
The invention aims to construct a fine composite multi-scale entropy diagram model through a road diagram topological structure, and combine the fine composite multi-scale entropy with a diagram rolling network to provide a rolling bearing weak signal composite fault diagnosis method of the diagram rolling network.
The invention aims at realizing the following technical scheme:
a rolling bearing weak signal compound fault diagnosis method of a graph rolling network comprises the following steps:
step one, segmenting a time domain signal, extracting fine composite multi-scale entropy characteristics of each segment of signal, solving entropy values under multiple scales, taking an entropy sequence of each segment of signal as a node of a road map, and establishing a road map model based on fine composite multi-scale entropy; different weight definitions are carried out on adjacent matrixes in different forms by adopting Gaussian kernel function weighting, and a Laplacian matrix is calculated through a degree matrix and the adjacent matrix and is used as matrix representation of a graph;
and secondly, collecting vibration signal data of the rolling bearing in different fault states, calculating entropy values in multiple scales, taking a sample as a vertex to construct a road map, calculating a corresponding adjacent matrix, and inputting the disturbed data set and the adjacent matrix into the GCN to realize image-level fault diagnosis.
Compared with the prior art, the invention has the following advantages:
1. the road map accords with the fact that the vibration signal belongs to a continuous time sequence signal, and the regularity of the vibration signal along with the change of time is met, so that the vibration impact rule of the bearing can be obtained from the signal.
2. The GCN model based on the graph task is higher in GCN recognition accuracy than the GCN recognition accuracy of the node task, the graph task is to train the whole graph as a model, regularity among internal nodes is included, similarity and difference among the graphs are analyzed, and characteristic information of signals in a section of interval can be obtained.
3. The graph rolling network based on the road graph topological structure can obtain higher diagnosis accuracy under the condition of less data quantity, and the data sample constructed by the multi-scale entropy can still obtain better diagnosis accuracy under a small amount of sample data, and can accurately identify the composite fault type of the bearing.
4. Experimental verification shows that compared with a node construction method of a time domain or a frequency domain, the fine composite multi-scale entropy considers the cross correlation of all entropy values to evaluate the dynamic conditions in fault detection, and the edge connection mode enables fault information to be transmitted and updated only among fault nodes of the same type, so that the composite fault type of the rolling bearing can be effectively identified, and the fault diagnosis efficiency is improved.
Drawings
FIG. 1 is a graph-convolution-based network rolling bearing composite fault diagnosis model;
FIG. 2 is a graph showing waveforms of vibration signals in time and frequency domains in 10 states of a rolling bearing;
FIG. 3 shows classification accuracy (single fault diagnosis) of different fault diagnosis methods;
FIG. 4 is a confusion matrix (single fault diagnosis) for different fault diagnosis methods;
FIG. 5 is a powertrain bench;
FIG. 6 is a graph of classification accuracy (composite fault diagnosis) for different fault diagnosis methods;
fig. 7 is a confusion matrix (composite fault diagnosis) for different fault diagnosis methods.
Detailed Description
The following description of the present invention is provided with reference to the accompanying drawings, but is not limited to the following description, and any modifications or equivalent substitutions of the present invention should be included in the scope of the present invention without departing from the spirit and scope of the present invention.
The invention provides a rolling bearing weak signal composite fault diagnosis method of a graph rolling network, which comprises the steps of firstly providing a road graph model construction method based on fine composite multi-scale entropy, segmenting signals into N sections through a sliding window by adopting the multi-scale entropy method, extracting the characteristic of the fine composite multi-scale entropy of time domain signals of each section, and defining different weights for adjacent matrixes in different forms by using a Gaussian kernel function. Finally, a rolling bearing composite fault diagnosis model construction based on a graph rolling network is proposed, in order to realize fault diagnosis at the graph level, a representation of the whole graph is obtained using a GCN with a graph pooling layer and a readout layer, and after the graph representation is obtained, standard machine learning techniques can be performed on the representation. The specific implementation steps are as follows:
step one, constructing a road map model based on fine composite multi-scale entropy
Step one, constructing a road map model based on fine composite multi-scale entropy:
the time domain signal is segmented into N segments by sliding windows,l is the sample length of the original data, n is the sub-sample length, then the time domain signal of each section is subjected to fine composite multi-scale entropy feature extraction, the entropy values under a plurality of scales are obtained, and each time domain signal is subjected to entropy value extractionThe entropy sequence under the sliding window is used as a node of the road map, and a road map model based on the fine composite multi-scale entropy is established.
Step two, a neighbor matrix weighting mode of a road map model:
and different weight definitions are carried out on adjacent matrixes in different forms by adopting Gaussian kernel function weighting, the connection edge weight under the Gaussian kernel function weighting can effectively reduce the difference between vertexes, and a Laplace matrix is calculated through a degree matrix and the adjacent matrix, so that the graph is represented.
Step two, constructing a rolling bearing composite fault diagnosis model based on a graph rolling network
Step two, collecting data and calculating an adjacency matrix:
collecting vibration signal data of the rolling bearing in different fault states, and performing fine composite multi-scale entropy processing on the vibration signal data to obtain entropy values in multiple scales; and constructing a road map by taking all samples as vertexes, calculating a corresponding adjacency matrix by utilizing a Gaussian kernel function distance operation formula, dividing a data set into a training set and a testing set, and randomly disturbing.
Step two, rolling bearing compound fault diagnosis of the graph rolling network:
and step two, inputting the adjacency matrix and the data set in the step two into a trained GCN network, performing fault diagnosis on the rolling bearing test sample of the unknown label by using a forward propagation model, and outputting the final fault type through a Softmax classifier to realize image level fault diagnosis.
Examples:
in this embodiment, the rolling bearing weak signal composite fault diagnosis method based on the graph rolling network includes the following steps:
step one, constructing a road map model based on fine composite multi-scale entropy;
and secondly, constructing a rolling bearing composite fault diagnosis model based on a graph rolling network.
The rolling bearing composite fault diagnosis model based on the graph rolling network provided by the embodiment is shown in figure 1, and adopts the bearing fault data in the CWRU data set, kessi Chu Da bearingThe experimental platform mainly comprises an asynchronous motor, a torsion sensor, an alternating current power dynamometer and an indicator. The measured bearing is a SKF deep groove ball bearing near the drive end 6205 of the motor. The fault damage sizes of the inner ring, the outer ring and the rolling bodies of the bearing are 0.1778mm, 0.3556mm and 0.5334mm, the rotating speed is 1797r/min, and the sampling frequency is 12kHz. 200 groups were acquired for each type of fault condition, each group containing 2048 sample points. Firstly, a maximum and minimum normalization method is adopted to perform normalization processing, and the normalization processing is performed according to a formula pi= [ (x) 1 ,y 1 ),(x 2 ,y 2 ),…,(x N ,y N )]Dividing the input signal into several sub-samples, wherein:n is the obtained sample set, x represents the subsamples, y represents the labels, N represents the number of subsamples, l is the sample length of the original data, and N is the subsamples length. Each sub-sample is assigned a corresponding label and the original signal is truncated without overlapping using a sliding window of length 2048. Wherein the length d of each sub-sample is set to 2048. Then, 10 samples can be obtained. In the construction process of the road map, one map data is constructed every 10 sub-samples, numbered in the order from 0 to 9, each sub-sample being connected in time series. The original data is subjected to fine composite multi-scale entropy to obtain a total sample number of 50000, wherein the total sample number comprises 10 types of bearing faults, and each fault comprises 5000 weighted entropy samples. The data of the road map structure is expressed as a map composed of 10 nodes and 9 edges, and one node is composed of a vector of 1×25 dimensions, so that the total number of nodes of the sample is 2000. Each dataset was randomly extracted with 80% as training set and the remaining 20% as test set. The iteration number is 100, the initial learning rate is 0.01, and SGD is adopted as an optimization algorithm, wherein the momentum of the SGD is 0.9. The batch scale is 64, each model is trained for 100 cycles for fault diagnosis, the learning rate decay strategy is also used for adjusting the learning rate, the weight decay value is initialized to 0.0005, and the loss function is used for carrying out feature extraction experimental analysis on different fault positions and different fault sizes of cross entropy. The specific parameters and failure characteristic frequencies of the bearings are shown in table 1. Different fault conditionsThe experimental parameters of the state are set forth in table 2. The time domain and frequency domain waveforms for 10 different fault types are shown in fig. 2.
Table 1 6205 bearing parameters and failure characteristic frequencies
TABLE 2 Experimental parameter settings for different fault states
(1) Rolling bearing single fault diagnosis comparison experiment verification under different classification models
In order to verify the effectiveness of the method, six diagnostic methods, namely a support vector machine (support vector machines, SVM), a Nearest Neighbor (KNN) model, a multi-layer perceptron (Multilayer Perceptron, MLP), a CNN model and a Long Short-Term Memory (LSTM) model, and a GCN model, are adopted for verification. The following three cases are classified: in the first case, in order to verify the traditional machine learning models SVM and KNN, the fine composite multi-scale entropy is adopted to construct a graph model through a Gaussian kernel function, and then the SVM and KNN classification is carried out. In the second case, in order to verify the deep learning model MLP, CNN, LSTM, a one-dimensional data set is constructed by adopting a fine composite multi-scale entropy, and fault classification is performed on a data structure with data samples of 1×100. In the third case, in order to verify the method provided by the invention, a graph model is constructed through a fine composite multi-scale entropy through a road graph, and fault classification is carried out by adopting a GCN model. The classification accuracy of the different fault diagnosis methods is shown in fig. 3.
From the classification accuracy rate under 10 iterations of different fault diagnosis methods, as shown in fig. 3, the recognition accuracy rate of fault diagnosis in the GCN model is 100% by taking the fine composite multi-scale entropy as the input of the graph model. Compared with the SVM, KNN, MLP, CNN, LSTM model, the iterative convergence speed is faster, and the model is more stable. The fault recognition accuracy of the deep learning method based on the fine composite multi-scale entropy is improved. The feature extraction mode of the fine composite multi-scale entropy enables different bearing fault types to keep fluctuating in a certain entropy value interval, and the node-node relation considered by the topological structure of the sampling road map considers the change condition of the whole bearing fault data on all scales. In the fault diagnosis of the image level, better results can be obtained by adopting a GCN model.
In order to intuitively evaluate the classification effect of the model, a confusion matrix is adopted for visual analysis. The test set diagnosis accuracy rates of the different fault diagnosis methods are shown in table 3 according to the classification condition of each fault class, and the confusion matrix of the test samples of the different fault diagnosis methods is drawn as shown in fig. 4.
TABLE 3 diagnostic accuracy of different fault diagnosis methods
As can be seen from fig. 4 and table 3, on the SVM model, the rolling element moderate fault samples were erroneously recognized as rolling element mild faults, and the rolling element moderate fault recognition accuracy was 51%. The inner ring moderate fault sample is erroneously identified as an inner ring severe fault, the inner ring moderate fault identification precision is 84%, and the overall model identification precision is 93.5%. On the KNN model, the rolling element mild failure sample was erroneously identified as a moderate failure, the rolling element mild failure identification accuracy was 92%, and the model overall identification accuracy was 99.19%. On the MLP model, the inner ring severe fault sample was erroneously identified as normal, with an inner ring severe fault identification accuracy of 90%. The outer ring severe fault sample is erroneously identified as a rolling body mild fault, the outer ring severe fault identification precision is 90%, and the overall model identification precision is 97%. On the CNN model, the rolling element moderate fault samples were erroneously identified as outer ring moderate faults, and the rolling element moderate fault identification accuracy was 90%. The outer ring severe fault sample is erroneously identified as a rolling body weight fault, the outer ring severe fault identification precision is 90%, and the overall model identification precision is 98%. On the LSTM model, the rolling body weight failure samples were erroneously identified as an outer ring medium failure, and the rolling body weight failure identification accuracy was 80%. The inner ring severe fault sample is wrongly identified as the outer ring moderate fault, the inner ring severe fault identification precision is 90%, the outer ring severe fault sample is wrongly identified as the outer ring moderate fault, the outer ring severe fault identification precision is 90%, and the overall model identification precision is 96%. The GCN model has 100% of overall and local fault recognition accuracy, and has high recognition accuracy.
(2) Rolling bearing composite fault diagnosis comparison experiment verification under different classification models
An XJTU Gecarbox experiment platform is adopted, and the experimental platform consists of a driving motor, a controller, a planetary gear box, a parallel shaft gear box and a brake, as shown in figure 5. Wherein, the motor type is three-phase 3 horsepower motor, and the power is three-phase alternating current. The two-stage planetary gear box is composed of two-stage planetary gear trains, wherein the first stage is provided with 3 planetary gears of a solar belt, the second stage is provided with 4 planetary gears of the solar belt, and the overall transmission ratio is 27:1, the planetary gear bearing is a rolling bearing 61800, and the measured bearing is arranged on a planetary gear supporting seat of the planetary gear box. Two 1 acceleration sensor PCBs 352C04 are mounted in the X and Y directions of the planetary gear box for collecting vibration signals and using Y-direction signals. In experiments, four types of bearing failure modes were prefabricated on the planetary gear box, including rolling element failure, inner ring failure, outer ring failure, and compound failure. Together with the normal state, 5 vibration signals were collected in total. Further, during the experiment, the motor speed was set to 1800r/min and the sampling frequency was set to 20.48kHz.
The sample selection mode is consistent with the CWRU data set, the total sample number obtained by the original data through fine composite multi-scale entropy is 25000, the sample selection mode comprises 5 types of bearing faults, and each fault comprises 5000 entropy samples. The data of the road map structure is expressed as a map composed of 10 nodes and 9 edges, and one node is composed of a vector of 1×25 dimensions, so that the total number of nodes of the sample is 1000. Each dataset was randomly extracted with 80% as training set and the remaining 20% as test set. The number of iterations is 100. And respectively adopting SVM, KNN, MLP, CNN, LSTM, GCN model and adopting six diagnostic methods to make verification. The following three cases are classified: in the first case, in order to verify the effect of combining the traditional machine learning method SVM, KNN and the graph model, the graph model is constructed by adopting a Gaussian kernel function by adopting a fine composite multi-scale entropy, and then fault classification is carried out. In the second case, in order to verify the deep learning model MLP, CNN, LSTM, a one-dimensional data set is constructed by adopting a fine composite multi-scale entropy, and fault classification is performed on a data structure with data samples of 1×100. In the third case, to validate the method presented herein, a graph model is constructed by fine-composite multi-scale entropy through the road map, and the GCN model is employed for fault classification. The classification accuracy of the different fault diagnosis methods is shown in fig. 6.
As can be seen from fig. 6, the fine composite multi-scale entropy as an input to the graph model achieves 100% recognition accuracy in the fault diagnosis in the GCN model. Compared with the SVM, KNN, MLP, CNN, LSTM model, the iterative convergence speed is faster, and the model is more stable. The characteristic extraction mode of the fine composite multi-scale entropy enables different bearing fault types to keep fluctuating in a certain entropy value interval, the relation between nodes is considered by the topological structure of the road map, and the change condition of the whole bearing fault data on all scales is considered. In the fault diagnosis of the image level, better results can be obtained by adopting a GCN model. In order to intuitively show the classification condition of each fault category, the confusion matrix of test samples drawn with different fault diagnosis methods is shown in fig. 7, and the diagnosis accuracy of test sets of different fault diagnosis methods is shown in table 4.
TABLE 4 diagnostic accuracy of different fault diagnosis methods
As can be seen from fig. 7 and table 4, the rolling element failure samples were erroneously recognized as a composite failure on the SVM model, and the rolling element failure recognition accuracy was 98%. The outer ring fault sample is wrongly identified as a composite fault, and the outer ring fault identification precision is 95%. The composite fault sample is erroneously identified as an outer ring fault, and the composite fault identification accuracy is 94%. The overall recognition accuracy of the model was 97.38%. On the KNN model, the composite fault sample is wrongly identified as an inner ring fault, the composite fault identification precision is 99%, and the overall identification precision of the model is 99.75%. On the MLP model, the rolling element failure samples were erroneously identified as an outer ring failure, and the rolling element failure identification accuracy was 90%. The inner ring failure sample was erroneously identified as a rolling element failure, and the inner ring failure identification accuracy was 90%. The composite fault sample is erroneously identified as an outer ring fault, the composite fault identification precision is 90%, and the overall model identification precision is 94%. On the CNN model, the rolling element failure sample was erroneously identified as a composite failure, the rolling element failure identification accuracy was 90%, and the model overall identification accuracy was 98%. On the LSTM model, the rolling element failure samples were erroneously identified as a composite failure, the rolling element failure identification accuracy was 90%, and the model overall identification accuracy was 98%. The GCN model has 100% of overall and local fault recognition accuracy, and has high recognition accuracy.

Claims (1)

1. A rolling bearing weak signal composite fault diagnosis method of a graph rolling network is characterized by comprising the following steps:
step one, segmenting a time domain signal, extracting fine composite multi-scale entropy characteristics of each segment of signal, solving entropy values under multiple scales, taking an entropy sequence of each segment of signal as a node of a road map, and establishing a road map model based on fine composite multi-scale entropy; different weight definitions are carried out on adjacent matrixes in different forms by adopting Gaussian kernel function weighting, and a Laplacian matrix is calculated through a degree matrix and the adjacent matrix and is used as matrix representation of a graph, and the method comprises the following specific steps of:
step one, constructing a road map model based on fine composite multi-scale entropy:
the time domain signal is segmented into N segments by sliding windows,l is the sample length of the original data, n is the sub-sample lengthCarrying out fine composite multi-scale entropy feature extraction on the time domain signals of each section, solving entropy values under a plurality of scales, taking the entropy sequence under each sliding window as a node of a road map, and establishing a road map model based on fine composite multi-scale entropy;
step two, a neighbor matrix weighting mode of a road map model:
different weight definitions are carried out on adjacent matrixes in different forms by adopting Gaussian kernel function weighting, the connection edge weight under the Gaussian kernel function weighting can effectively reduce the difference between vertexes, and a Laplace matrix is calculated through a degree matrix and the adjacent matrix, so that the graph is represented;
collecting vibration signal data of the rolling bearing in different fault states, calculating entropy values under multiple scales, constructing a road map by taking a sample as a vertex, calculating a corresponding adjacent matrix, inputting a disturbed data set and the adjacent matrix into the GCN, and realizing map-level fault diagnosis, wherein the method comprises the following specific steps of:
step two, collecting data and calculating an adjacency matrix:
collecting vibration signal data of the rolling bearing in different fault states, and performing fine composite multi-scale entropy processing on the vibration signal data to obtain entropy values in multiple scales; constructing a road map by taking all samples as vertexes, calculating a corresponding adjacency matrix by utilizing a Gaussian kernel function distance operation formula, dividing a data set into a training set and a testing set, and randomly disturbing the training set and the testing set;
step two, rolling bearing compound fault diagnosis of the graph rolling network:
and step two, inputting the adjacency matrix and the data set in the step two into a trained GCN network, performing fault diagnosis on the rolling bearing test sample of the unknown label by using a forward propagation model, and outputting the final fault type through a Softmax classifier to realize image level fault diagnosis.
CN202310321870.8A 2023-03-29 2023-03-29 Rolling bearing weak signal composite fault diagnosis method based on graph rolling network Active CN116358871B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310321870.8A CN116358871B (en) 2023-03-29 2023-03-29 Rolling bearing weak signal composite fault diagnosis method based on graph rolling network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310321870.8A CN116358871B (en) 2023-03-29 2023-03-29 Rolling bearing weak signal composite fault diagnosis method based on graph rolling network

Publications (2)

Publication Number Publication Date
CN116358871A CN116358871A (en) 2023-06-30
CN116358871B true CN116358871B (en) 2024-01-23

Family

ID=86919282

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310321870.8A Active CN116358871B (en) 2023-03-29 2023-03-29 Rolling bearing weak signal composite fault diagnosis method based on graph rolling network

Country Status (1)

Country Link
CN (1) CN116358871B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116842459B (en) * 2023-09-01 2023-11-21 国网信息通信产业集团有限公司 Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning
CN117191385B (en) * 2023-09-27 2024-09-13 重庆凯测试验设备股份有限公司 Coaxial double-motor-based wind power gear box testing method
CN117892080A (en) * 2023-12-19 2024-04-16 南京天洑软件有限公司 Method and device for establishing and diagnosing fault diagnosis model of valve body of oilfield fracturing pump
CN117668469B (en) * 2024-01-31 2024-04-02 深圳华诚包装科技股份有限公司 Printing machine working state management method and system
CN118427689B (en) * 2024-06-25 2024-09-27 华东交通大学 Transmission rolling bearing compound fault diagnosis method and system

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104849050A (en) * 2015-06-02 2015-08-19 安徽工业大学 Rolling bearing fault diagnosis method based on composite multi-scale permutation entropies
CN112651167A (en) * 2020-12-02 2021-04-13 杭州电子科技大学 Semi-supervised rolling bearing fault diagnosis method based on graph neural network
CN112985809A (en) * 2021-02-09 2021-06-18 上海电机学院 Rolling bearing fault diagnosis method based on signal multi-dimensional fine image
CN113532848A (en) * 2021-06-24 2021-10-22 南京工业大学 Fault diagnosis system for planetary gear box
CN114137358A (en) * 2021-11-17 2022-03-04 国网天津市电力公司信息通信公司 Power transmission line fault diagnosis method based on graph convolution neural network
CN114627657A (en) * 2022-03-09 2022-06-14 哈尔滨理工大学 Adaptive traffic signal control method based on deep graph reinforcement learning
CN115017955A (en) * 2022-06-21 2022-09-06 哈尔滨理工大学 Planetary gearbox fault diagnosis method based on semi-supervised graph convolution
CN115112372A (en) * 2022-06-22 2022-09-27 哈尔滨工业大学(深圳) Bearing fault diagnosis method and device, electronic equipment and storage medium
CN115563553A (en) * 2022-09-29 2023-01-03 南京理工大学 Aviation bearing fault diagnosis method of deep migration map convolution network under fluctuation working condition
CN115600088A (en) * 2022-10-15 2023-01-13 福州大学(Cn) Distribution transformer fault diagnosis method based on vibration signals
CN115734274A (en) * 2022-11-30 2023-03-03 南京邮电大学 Cellular network fault diagnosis method based on deep learning and knowledge graph

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104849050A (en) * 2015-06-02 2015-08-19 安徽工业大学 Rolling bearing fault diagnosis method based on composite multi-scale permutation entropies
CN112651167A (en) * 2020-12-02 2021-04-13 杭州电子科技大学 Semi-supervised rolling bearing fault diagnosis method based on graph neural network
CN112985809A (en) * 2021-02-09 2021-06-18 上海电机学院 Rolling bearing fault diagnosis method based on signal multi-dimensional fine image
CN113532848A (en) * 2021-06-24 2021-10-22 南京工业大学 Fault diagnosis system for planetary gear box
CN114137358A (en) * 2021-11-17 2022-03-04 国网天津市电力公司信息通信公司 Power transmission line fault diagnosis method based on graph convolution neural network
CN114627657A (en) * 2022-03-09 2022-06-14 哈尔滨理工大学 Adaptive traffic signal control method based on deep graph reinforcement learning
CN115017955A (en) * 2022-06-21 2022-09-06 哈尔滨理工大学 Planetary gearbox fault diagnosis method based on semi-supervised graph convolution
CN115112372A (en) * 2022-06-22 2022-09-27 哈尔滨工业大学(深圳) Bearing fault diagnosis method and device, electronic equipment and storage medium
CN115563553A (en) * 2022-09-29 2023-01-03 南京理工大学 Aviation bearing fault diagnosis method of deep migration map convolution network under fluctuation working condition
CN115600088A (en) * 2022-10-15 2023-01-13 福州大学(Cn) Distribution transformer fault diagnosis method based on vibration signals
CN115734274A (en) * 2022-11-30 2023-03-03 南京邮电大学 Cellular network fault diagnosis method based on deep learning and knowledge graph

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于精细复合多尺度熵和自编码的滚动轴承故障诊断方法;郑近德等;《噪声与振动控制》;第39卷(第2期);第175-180, 193页 *

Also Published As

Publication number Publication date
CN116358871A (en) 2023-06-30

Similar Documents

Publication Publication Date Title
CN116358871B (en) Rolling bearing weak signal composite fault diagnosis method based on graph rolling network
Saufi et al. Challenges and opportunities of deep learning models for machinery fault detection and diagnosis: A review
Yang et al. Fault diagnosis of gearbox based on RBF-PF and particle swarm optimization wavelet neural network
Jiao et al. Deep coupled dense convolutional network with complementary data for intelligent fault diagnosis
CN110849626B (en) Self-adaptive sparse compression self-coding rolling bearing fault diagnosis system
CN112906644B (en) Mechanical fault intelligent diagnosis method based on deep migration learning
Li et al. Joint attention feature transfer network for gearbox fault diagnosis with imbalanced data
CN113188794B (en) Gearbox fault diagnosis method and device based on improved PSO-BP neural network
CN113505664B (en) Fault diagnosis method for planetary gear box of wind turbine generator
Li et al. Multi-scale CNN for multi-sensor feature fusion in helical gear fault detection
Ghulanavar et al. Diagnosis of faulty gears by modified AlexNet and improved grasshopper optimization algorithm (IGOA)
CN113112000A (en) Construction method and application of multilayer convolutional neural network model
Han et al. A novel intelligent fault diagnosis method based on dual convolutional neural network with multi-level information fusion
Zhang et al. Compound Fault Diagnosis for Gearbox Based Using of Euclidean Matrix Sample Entropy and One‐Dimensional Convolutional Neural Network
CN117315328B (en) Gear fault diagnosis method for improving convolutional neural network under attention mechanism
CN116821762A (en) Mechanical fault diagnosis method based on multi-scale graph attention fusion network
He et al. Fault identification of rotating machinery based on dynamic feature reconstruction signal graph
CN117191396A (en) Gear box fault diagnosis method based on two-stage migration
CN116993322A (en) Fault diagnosis method and system based on convolutional network
CN115127813B (en) Rolling bearing multichannel fusion diagnosis method based on tensor features and tensor supporting machine
CN114354185B (en) Gear box fault diagnosis method based on decoupling network
Chen et al. A compound fault diagnosis method based on Multi-task Learning with Multi-gate Mixture-of-Experts
CN112161805A (en) Bearing tiny fault diagnosis method based on time series scale analysis and CNN
CN117992863B (en) Rotary machine fault diagnosis method based on interpretable stationary wavelet packet convolution network
CN117516925B (en) MSDC-Swin-T-based gearbox fault diagnosis method

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