CN112304614A - End-to-end rolling bearing intelligent fault diagnosis method adopting multi-attention machine system - Google Patents

End-to-end rolling bearing intelligent fault diagnosis method adopting multi-attention machine system Download PDF

Info

Publication number
CN112304614A
CN112304614A CN202011359124.0A CN202011359124A CN112304614A CN 112304614 A CN112304614 A CN 112304614A CN 202011359124 A CN202011359124 A CN 202011359124A CN 112304614 A CN112304614 A CN 112304614A
Authority
CN
China
Prior art keywords
rolling bearing
fault
data set
attention
signals
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.)
Granted
Application number
CN202011359124.0A
Other languages
Chinese (zh)
Other versions
CN112304614B (en
Inventor
刘永葆
李俊
贺星
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.)
Naval University of Engineering PLA
Original Assignee
Naval University of Engineering PLA
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 Naval University of Engineering PLA filed Critical Naval University of Engineering PLA
Priority to CN202011359124.0A priority Critical patent/CN112304614B/en
Publication of CN112304614A publication Critical patent/CN112304614A/en
Application granted granted Critical
Publication of CN112304614B publication Critical patent/CN112304614B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G01M13/045—Acoustic or vibration analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Acoustics & Sound (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention discloses an end-to-end rolling bearing intelligent fault diagnosis method adopting a multi-attention machine system. The method combines fault feature extraction and fault mode classification, and realizes weighted expression of various fault features through a multi-attention machine mechanism; the intelligent fault diagnosis method for the end-to-end rolling bearing adopting the multi-attention machine system comprises a method for acquiring rolling bearing data and converting vibration signals into images, wherein the method for converting the vibration signals into the images comprises the steps of obtaining corresponding speed and displacement by integrating and calculating vibration acceleration signals of the rolling bearing, and then combining the acceleration signals, the speed signals and the displacement signals to obtain the images with enhanced characteristics. The method overcomes the defects that in the prior art, a deep neural network is easily influenced by non-sensitive characteristics, so that the accuracy of fault diagnosis of the rolling bearing is limited, and meanwhile, a large amount of time is needed for generating training samples and the requirement on professional knowledge is high; the method has the advantages of high diagnosis precision, high diagnosis speed and simple and convenient operation.

Description

End-to-end rolling bearing intelligent fault diagnosis method adopting multi-attention machine system
Technical Field
The invention relates to a fault diagnosis method for a rolling bearing, in particular to an intelligent fault diagnosis method for an end-to-end rolling bearing by adopting a multi-attention machine system.
Background
The rolling bearing is an important equipment basic part and a mechanical universal element which are widely applied, is indispensable in the equipment manufacturing industry, directly determines the performance, quality and reliability of important equipment and host machine products, and is known as an industrial joint. Meanwhile, the rolling bearing is often in a severe working environment and has the characteristics of high running speed, complex structure and easy failure, and the rolling bearing is also one of vulnerable parts of the rotary machine. According to statistics, the faults of the rotating machinery are related to the bearing faults, and once the bearing is in fault, a series of cascading faults are caused, and the operation safety of the whole equipment is directly influenced seriously. Therefore, condition monitoring and fault diagnosis of the rolling bearing have very important significance, and the condition monitoring and fault diagnosis is always one of the important development directions in mechanical fault diagnosis.
With the rapid development of Artificial Intelligence (AI), machine learning methods have been widely used in condition monitoring and fault diagnosis of rolling bearings. However, the conventional intelligent fault diagnosis method has the problems that manual feature selection is required and a large amount of label data is required for training, and in order to solve the problems in fault diagnosis, the emerging deep learning method is gradually applied to the field of fault diagnosis by people. In 2006, Hinton et al proposed to reduce the dimensionality of data using an auto-encoder (autoencoder) and proposed to quickly train a deep belief network in a pre-trained manner to suppress the gradient vanishing problem. Taking this as a marker, deep learning has made a breakthrough in image recognition, speech recognition, natural language processing, and the like as a new method in the emerging pattern recognition field. Meanwhile, due to the deep learning multi-level structure, the deep level relation can be extracted from a large amount of data, and the method is also paid greater attention and applied to the field of bearing fault diagnosis. The existing fault diagnosis of the rolling bearing usually adopts a deep neural network, and the deep neural network is easily influenced by non-sensitive characteristics, so that the accuracy of fault diagnosis of the rolling bearing is influenced.
In processing large amounts of data, different methods are required for the data of different characteristics to correctly display their characteristics. In order to realize fault diagnosis of the rolling bearing, a lot of scholars combine vibration signals and deep learning technology to provide methods such as time-frequency graphs and histograms to convert the vibration signals into images for classification, but generating training samples takes a lot of time and depends on expert experience knowledge to a great extent.
Therefore, there is a need to develop a method for diagnosing rolling bearing failure intelligently and rapidly.
Disclosure of Invention
The invention aims to provide an end-to-end rolling bearing intelligent fault diagnosis method adopting a multi-Attention mechanism, which is a rolling bearing fault diagnosis method based on the combination of CBAM-ResNet.
In order to achieve the purpose, the technical scheme of the invention is as follows: the intelligent fault diagnosis method for the end-to-end rolling bearing by adopting a multi-attention machine system is characterized by comprising the following steps of: combining fault feature extraction and fault mode classification, and realizing weighted expression of various fault features through a multi-attention machine mechanism;
the intelligent fault diagnosis method for the end-to-end rolling bearing adopting the multi-attention machine system comprises a method for acquiring rolling bearing data and converting vibration signals into images, wherein the method for converting the vibration signals into the images comprises the steps of obtaining corresponding speed and displacement by integrating and calculating vibration acceleration signals of the rolling bearing, and then combining the acceleration signals, the speed signals and the displacement signals to obtain the images with enhanced characteristics.
In the technical scheme, the intelligent fault diagnosis method for the end-to-end rolling bearing by adopting the multi-attention machine system comprises the following steps which are executed in sequence,
the method comprises the following steps: acquiring rolling bearing data;
step two: preprocessing rolling bearing data;
the rolling bearing data preprocessing is to convert vibration signals of the rolling bearing into images; the method for converting the vibration signal of the rolling bearing into the image comprises the following steps: obtaining corresponding speed and displacement through integral calculation of vibration acceleration signals of the rolling bearing, and combining the acceleration signals, the speed signals and the displacement signals to obtain an image with enhanced characteristics;
step three: constructing a CBAM-ResNet diagnostic model and analyzing and constructing a diagnostic result;
and (4) sending the image in the step one into a CBAM-ResNet diagnostic model for training, and classifying the test data set by using the trained CBAM-ResNet diagnostic model.
In the technical scheme, the adopted multi-attention machine is a channel attention and space attention structure.
In the above technical solution, in the second step, the method for converting the vibration signal into the image specifically includes the following steps,
intercepting a vibration signal sample according to four categories of outer ring faults, inner ring faults, rolling body faults and normal conditions of the rolling bearing and according to the standard of 50% of an overlapped signal to obtain a vibration acceleration signal data set { AC i1,2, …, M × L, where L represents the total number of samples and M × M represents the pixel size of the image;
obtained by integration, a velocity signal data set VE i1,2, …, mxmxmxl }, displacement signal data set { DIS ═ i | 1,2, …, M × L }, displacement signal data set { DISi|i＝1,2,…,M×M×L}；
Then, the vibration acceleration signal data set, the speed signal data set and the displacement signal data set are respectively substituted into the formula (6) for processing, so that the data range is converted into the range between [0,255],
Figure BDA0002803480530000031
in equation (6), P (m, m), m ═ 1,2 … j represents a pixel point of an image, the function round (·) is a rounding function, and x isiRepresenting the ith sample, x, of the data setminRepresenting the minimum value of a sample, x, in the datasetmaxRepresents a maximum value of samples in the dataset;
in the generated RGB image, the red channel pixel values are filled by the acceleration signal data set, the green channel pixel values are filled by the velocity signal data set, and the blue channel pixel values are filled by the displacement signal data set.
In the above technical solution, in step three, in the original ResNet network structure, a CBAM attention module is added behind each group of residual blocks, and the number of outputs of the output layer is set to 4 to match four health types of an outer ring fault, an inner ring fault, a rolling element fault, and a normal condition of the rolling bearing.
The invention has the following advantages:
(1) the method has the advantages that the problems of overfitting, gradient disappearance or gradient explosion easily occurring in the training process of the deep neural network model are solved by introducing the residual block to construct the deep network; in the original ResNet network structure, a CBAM attention module is added behind each group of residual blocks, and the feature extraction capability of the model is improved by introducing the CBAM attention module;
(2) the invention can be selectively characterized by using an attention mechanism, thereby effectively overcoming the problem that a deep neural network is easily influenced by non-sensitive characteristics, and more fully utilizing the characteristics and the information among the characteristics;
(3) according to the invention, fault feature extraction and fault mode classification are fused together, and the attention mechanism can realize weighted expression of different features, so that the classified features have higher expression capability;
(4) the method utilizes the characteristic that the vibration acceleration signal can obtain corresponding speed and displacement through integral calculation, combines the acceleration signal, the speed signal and the displacement signal to obtain an image with enhanced characteristics, and can quickly obtain an image with enhanced characteristics from original data without presetting parameters or expert experience; and then, the graph is taken as a model and input into a CBAM-ResNet diagnosis model for training and diagnosis, so that the distribution characteristics of a fault mode can be captured, and the diagnosis speed and the diagnosis precision are improved.
(5) The classification precision of the model training set is close to 100%, and the fault diagnosis precision of the rolling bearing reaches 98.33%.
Drawings
Fig. 1 is a diagram of residual units of a deep residual network according to the present invention.
Fig. 2 is a structural diagram of a convolutional neural network incorporating CBAM according to the present invention.
FIG. 3 is a diagnostic flow chart of the present invention.
Fig. 4 is a flowchart of a method for converting a vibration signal into an image according to the present invention.
FIG. 5 is a structural diagram of a CBAM-ResNet model in the present invention.
Fig. 6 is a structural view of a laboratory table used in example 1 of the present invention.
Fig. 7 is a flowchart of data conversion into an image according to embodiment 1 of the present invention.
Fig. 8 is a graph showing the relationship between the accuracy and the loss function value of the CBAM-ResNet model in the training set and the number of iterations in embodiment 1 of the present invention.
FIG. 9 is a confusion matrix chart of the diagnosis results in embodiment 1 of the present invention.
Fig. 10 is a structural view of a laboratory table used in example 2 of the present invention.
Fig. 11 is a time domain diagram of the vibration signal of each health type in embodiment 2 of the present invention.
Fig. 12 is a graph showing the relationship between the accuracy and the loss function value of the CBAM-ResNet model in the training set and the number of iterations in embodiment 2 of the present invention.
FIG. 13 is a confusion matrix chart of the diagnosis results in embodiment 2 of the present invention.
Detailed Description
The embodiments of the present invention will be described in detail with reference to the accompanying drawings, which are not intended to limit the present invention, but are merely exemplary. While the advantages of the invention will be clear and readily understood by the description.
With reference to the accompanying drawings: the intelligent fault diagnosis method for the end-to-end rolling bearing by adopting the multi-attention machine system combines fault feature extraction and fault mode classification, realizes the weighted expression of various fault features by the multi-attention machine system, and can selectively characterize by using the attention machine system, thereby effectively overcoming the problem that a deep neural network is easily influenced by non-sensitive features and more fully utilizing the features and information among the features; because the attention mechanism has the excellent characteristics, in the characteristic extraction stage, the important characteristics are focused, meanwhile, the unnecessary characteristics are inhibited, the characteristic expression capability of the convolutional neural network model is improved, and the accuracy of fault diagnosis of the rolling bearing is further improved on the premise of not obviously increasing the calculated amount and the parameter amount; the end-to-end method is characterized in that the fault mode classification is finished by directly using the acquired vibration signals, and the manual characteristic extraction process is not needed in the middle;
the intelligent fault diagnosis method for the end-to-end rolling bearing adopting the multi-attention machine system comprises the steps of acquiring data of the rolling bearing and converting vibration signals into images, wherein the method for converting the vibration signals into the images is to obtain corresponding speed and displacement through integral calculation of vibration acceleration signals of the rolling bearing, then combine the acceleration signals, the speed signals and the displacement signals to obtain images with enhanced characteristics, and quickly and accurately realize end-to-end rolling bearing fault diagnosis through a simple diagnosis process.
Furthermore, the intelligent fault diagnosis method for the end-to-end rolling bearing adopting the multi-attention machine system comprises the following steps which are executed in sequence,
the method comprises the following steps: acquiring rolling bearing data; wherein, collecting original vibration signal is prior art;
step two: preprocessing rolling bearing data;
the rolling bearing data preprocessing is to convert the vibration signals of the rolling bearing into images (as shown in fig. 4, fig. 4 is sample segmentation and combines acceleration, velocity and displacement signals to form images) as model input; the method for converting the vibration signal of the rolling bearing into the image comprises the following steps: obtaining corresponding speed and displacement through integral calculation of vibration acceleration signals of the rolling bearing, and combining the acceleration signals, the speed signals and the displacement signals to obtain an image with enhanced characteristics; data preprocessing is the first step of deep learning and is also an important step; in the process of processing a large amount of data, the data with different characteristics can correctly display the characteristics of the data by different methods, the method for converting the vibration signal of the rolling bearing into the image can quickly obtain the image with enhanced characteristics from the original data without presetting parameters or expert experience, has the characteristics of quickness and intelligence, and quickly and accurately realizes the end-to-end rolling bearing fault diagnosis through a simple diagnosis process; the defects that in the prior art, in order to realize fault diagnosis of the rolling bearing, vibration signals are converted into images for classification by using methods such as time-frequency graphs and histograms in combination with vibration signals and deep learning technologies, but a large amount of time is needed for generating training samples, and the generation of the training samples depends on expert experience knowledge to a great extent, is time-consuming and has high requirements on professional knowledge are overcome;
step three: constructing a CBAM-ResNet diagnostic model and analyzing and constructing a diagnostic result;
sending the image in the step one into a CBAM-ResNet diagnostic model for training, and classifying a test data set by using the trained CBAM-ResNet diagnostic model (as shown in figure 3), wherein the test data set is the image preprocessed in the step two; step two, forming a data set after preprocessing the images, and then dividing the data set into a training data set and a testing data set; the invention does not need to manually participate in the fault feature extraction and the fault mode classification, and the fault feature extraction and the fault mode classification are both completed by the CBAM-ResNet diagnostic model, thereby having the characteristics of rapidness and intelligence.
Both the CBAM model and the ResNet model are prior art.
Further, the multiple attention mechanism employed is a channel attention and spatial attention structure (as shown in FIG. 2); the CBAM model is a model combining a channel attention structure and a space attention structure; the channel attention and spatial attention structure (CBAM) uses an attention mechanism in the channel dimension and the spatial dimension respectively, emphasizes meaningful features in the two dimensions of the space and the channel, pays attention to important features and suppresses unnecessary features; meanwhile, the channel attention structure and the space attention structure are mutually independent, so that the channel attention structure and the space attention structure can be used as independent modules in the existing convolutional neural network architecture;
the processing of the multiple attention mechanism (CBAM) includes the following two operations:
Figure BDA0002803480530000061
Figure BDA0002803480530000062
in the formulae (4) and (5),
Figure BDA0002803480530000063
representing the characteristics of the input, cxhxb representing the dimensions of each channel,
Figure BDA0002803480530000064
the attention weight in the channel dimension is represented,
Figure BDA0002803480530000065
representing attention weights, signs, in spatial dimensions
Figure BDA0002803480530000066
Representing element-by-element multiplication; r represents a feature space; c × 1 × 1 represents the channel dimension; 1 × H × B represents a spatial dimension; f' represents a weighted value of the input feature multiplied element by the channel attention weight; f "represents a weighted value of F' multiplied element by the spatial attention weight.
Further, in the second step, the method for converting the vibration signal into the image specifically comprises the following steps,
according to four categories of outer ring faults, inner ring faults, rolling body faults and normal conditions of a rolling bearing, intercepting vibration signal samples according to the standard of 50% of overlapping signals (for example, starting from the 1 st point of vibration signal data, intercepting the nth point as the first signal sample, then starting from the (N-m) th point of the next signal sample, intercepting the (2N-m) th point as the second signal sample, and sequentially and analogically realizing the segmented overlapping interception of the signals to obtain N signal samples in total, wherein N is a positive integer greater than or equal to 1, m is less than N, and N is a positive integer), and obtaining a vibration acceleration signal data set { AC (alternating Current)i1,2, …, M × L, where L represents the total number of samples and M × M represents the pixel size of the image; i represents the sequence number of each data in the data set;
obtained by integration, a velocity signal data set VE i1,2, …, mxmxmxl }, displacement signal data set { DIS ═ i | 1,2, …, M × L }, displacement signal data set { DIS i1,2, …, M × L }; wherein i represents the sequence number of each data in the data set; m × M represents a pixel size of an image; l represents the total number of samples;
the method for obtaining the corresponding speed and displacement by carrying out integral calculation on the vibration acceleration signal of the rolling bearing is the prior art;
then, the vibration acceleration signal data set, the speed signal data set and the displacement signal data set are respectively substituted into a formula (6) for processing, so that the data range is converted into a range between [0 and 255] (wherein, data normalization is carried out for converting the data into an RGB image, the value range of each pixel point of the RGB image is required to be [0 and 255]), and an RGB image is generated;
Figure BDA0002803480530000071
wherein, the formula (6) is used for normalizing the value of each pixel point on three channels of the RGB image to [0,255 ];
in formula (6), P (m, m), m is 1,2 … j represents a pixel of an image, and j is a positive integer; p (m, m) represents each pixel point on the image; the function round (·) is a rounding function; x is the number ofiRepresenting the ith sample of the data set; x is the number ofminRepresents the minimum of samples in the dataset; x is the number ofmaxRepresents a maximum value of samples in the dataset;
in the generated RGB image, the red channel pixel values are filled by the acceleration signal data set, the green channel pixel values are filled by the velocity signal data set, and the blue channel pixel values are filled by the displacement signal data set (as shown in fig. 4).
Furthermore, in step three, in the original ResNet network structure, a CBAM attention module is added after each group of residual blocks, and the number of output layers is set to 4 to match four health types (as shown in fig. 5) of the rolling bearing, such as outer ring fault, inner ring fault, rolling body fault and normal condition;
the original ResNet network is composed of a series of residual blocks, the residual block structure is as shown in fig. 1, for a stacking layer structure, when the input is x, the learned characteristics are denoted as h (x), the residual of the structure is f (x) ═ h (x) — x, when the residual f (x) ═ 0, the stacking layer only performs identity mapping, the deep network performance cannot be degraded, and when the residual f (x) ≠ 0, the stacking layer learns new characteristics on the basis of the input characteristics, so that the deep network has better performance;
let xlAnd xl+1Respectively representing the input and output of the l-th layer residual block, F (x)l,Wl) Is a residual function, W, representing the learned residuallIs the weight vector to be learned, a residual block can be represented as:
yl＝h(xl)+F(xl,Wl) (1)
xl+1＝f(yl) (2)
wherein f (-) is an activation function, typically a Relu function;
in formula (1), h (-) represents the learned feature;
in formula (2), f (-) is the Relu activation function; y islRepresenting the output of the residual block before the activation function;
the learning features from the shallow layer L to the deep layer L are obtained based on the equations (1) and (2):
Figure BDA0002803480530000081
wherein x isLRepresents the output of the L-th layer residual block, WiIs the weight vector to be learned;
in the formula (3), xlRepresents the output of the L-th layer residual block; x is the number oflRepresenting an output representing the l-th layer residual block; f (x)i，wi) Representing a residual function; x is the number ofiRepresenting the input of the i-th layer residual block.
Examples
The invention is explained in detail by taking the embodiment of the invention applied to the fault diagnosis of the rolling bearing at the driving end of a certain motor as an example, and has the guiding function of applying the invention to the fault diagnosis of other rolling bearings.
Example 1
In this example, a public rolling bearing dataset from the CWRU (Case Western Reserve University) bearing data center was used.
A rolling bearing failure diagnosis method in the present embodiment includes the steps of,
s1: collecting data;
the experiment table adopted in the embodiment is shown in fig. 6, and the bearing experiment table mainly comprises a motor, a sensor, a rolling bearing and a dynamometer; vibration data of rolling bearings at the driving end and the fan end of the motor are obtained by an acceleration sensor arranged on an induction motor shell, and the sampling frequency is 12 kHz;
besides the normal state of the rolling bearing, three fault states of inner ring fault, outer ring fault and rolling body fault are artificially introduced in an electric spark machining mode; in addition, each fault category had three defect diameters (0.007,0.014 and 0.021 inches) and bearing vibration signal acquisition was performed at three rotational speeds, respectively; in the present embodiment, the classification of the motor drive end rolling bearing failure at 1797rpm (load 0) and a defect diameter of 0.021 inch was studied; therefore, 4 health types including the normal state are included in the sample of the present embodiment, that is, the inner ring failure, the outer ring failure, the rolling element failure, and the normal state;
according to four categories of outer ring faults, inner ring faults, rolling body faults and normal conditions of the rolling bearing, the rolling bearing data acquisition method is adopted to obtain a vibration acceleration signal data set { ACi|i＝1,2,…,M×M×L}；
Obtained by integration, a velocity signal data set VE i1,2, …, mxmxmxl }, displacement signal data set { DIS ═ i | 1,2, …, M × L }, displacement signal data set { DISi|i＝1,2,…,M×M×L}；
S2: converting the data into an image;
by adopting the method for converting the vibration signals of the rolling bearing into the images, the vibration data of the four health types are converted into RGB images, and the result is shown in FIG. 7; wherein, each health type has 600 RGB images with the pixel size of 32 multiplied by 32, and the total number of the RGB images comprises 2400; randomly selecting 95% of data in the samples as training data, and taking the rest 5% of sample data as test samples, namely 2280 samples for training and 120 samples for testing;
as can be seen from the converted image (as shown in fig. 7), the RGB images of the rolling bearings differ from the health type; the image color points with faults of the outer ring are the most dense, compared with the image color points with faults of the inner ring, the image color points with faults of the rolling body are the most sparse, and the normal type image has obvious stripe characteristics;
s3: analyzing the diagnosis result of the CBAM-ResNet model;
after the conversion from data to images, sending the data set into a CBAM-CNN model for training, and then classifying the test data set by using the trained model; the training and the testing of the model are both carried out on the same workstation, namely a GeForce RTX 2080Ti display card (11GB video memory) under the windows10 environment is realized by using Python3.6 programming under a TensorFlow platform;
in the training process, the iteration number is set to be 120, the batch size is set to be 32, network parameters are updated by adopting an Adam optimization algorithm, the initial learning rate is set to be 0.001, and the learning rate is dynamically changed in the training process; the accuracy and loss function value of the CBAM-ResNet model in the training set and the relation graph of the iteration times are shown in FIG. 8;
as can be seen from fig. 8, it can be seen that the model training results are: the classification precision of the training set is close to 100%, which shows that the model has good fitting effect, the loss function is smoothly reduced and tends to be stable, and the loss function is not trapped in local optimization; then, the data of the test set is input into the trained diagnosis model, the results are averaged through 10 random experiments, and the classification precision of the obtained test set reaches 98.33%, which shows that the diagnosis method can capture the distribution characteristics of the fault mode. To further understand the detailed classification of each health type, a confusion matrix of diagnostic results is plotted, as shown in FIG. 9; as can be seen from fig. 9, in the test set, one inner ring fault is misjudged as a rolling element fault, one outer ring fault is misjudged as an inner ring fault, and all normal conditions are correctly classified.
The feature that the vibration acceleration signal is utilized to obtain the corresponding speed and displacement through integral calculation is further explained through experiments of a CWRU bearing fault data set, and the effectiveness and superiority of an RGB image with enhanced characteristics are formed by combining the acceleration signal, the speed signal and the displacement signal.
The acceleration signal, the speed signal and the displacement signal are respectively and independently combined into an RGB image, the acceleration signal and the speed signal are combined into the RGB image, the acceleration signal and the displacement signal are combined into the RGB image, and the speed and the displacement signal are combined into the RGB image, 6 conditions are compared and analyzed with the condition of embodiment 1 (number 7 in table 1) of the invention, the obtained image data sets are respectively sent into a CBAM-ResNet rolling bearing fault diagnosis model for training and diagnosis, and the rolling bearing fault diagnosis precision is shown in table 1.
TABLE 1 Rolling bearing fault diagnosis accuracy under different signal composition image data sets
Figure BDA0002803480530000101
As can be seen from table 1, when an RGB image is formed by using an acceleration signal, a velocity signal, and a displacement signal alone, the diagnostic accuracy of fault diagnosis performed by using an image formed by the acceleration signal in embodiment 1 of the present invention is significantly higher than the diagnostic accuracy of fault diagnosis performed by using an image formed by the velocity signal and the displacement signal; and when the acceleration signal is respectively combined with the other two signals, the diagnosis precision is respectively improved to 94.11 percent and 94.21 percent. But all are lower than the diagnostic accuracy of the method for converting the signal into the image provided by the embodiment 1 of the invention.
Verification method
In order to further illustrate the performance superiority of the CBAM-ResNet model in the present invention, three common classification models, namely, an SVM (Support Vector Machine), a BP (artificial neural network model) and a CNN (convolutional neural network model), are selected to perform comparative analysis with the above example 1. The SVM model uses 8 characteristic parameters including standard deviation, kurtosis, average value, root mean square, wave form factor, peak value factor, margin factor and kurtosis factor of a vibration signal of a rolling bearing (a ball bearing). The BP neural network also adopts the time domain characteristic parameters as input, and comprises 2 hidden layers, and each layer comprises 50 neurons. 95% of data in the samples are randomly selected as training data, the rest 5% of sample data are selected as test samples, several classification models pass 10 random experiments, and the results are averaged. The results of the failure diagnosis are shown in table 2.
TABLE 2 comparison of the results of the experiments with different models
Figure BDA0002803480530000111
As can be seen from table 2: after the BP neural network is trained by using the characteristic samples, the recognition rate is 83.33%, the accumulated total number of incorrectly recognized samples is 20, the recognition rate can reach 91.67% by using an SVM bearing fault diagnosis method and continuously trying, and after a proper parameter is selected by experience, the classification result can reach 95.83% by using a CNN model, but the classification result is still lower than the classification result of the CBAM-ResNet model in the embodiment 1 of the invention.
Example 2
The present embodiment employs a local laboratory bearing failure data set.
The embodiment uses a mechanical fault comprehensive simulation experiment table of SpectraQuest company to carry out a fault diagnosis test on a rolling bearing, and the specific diagnosis method comprises the following steps,
s1: data acquisition and preprocessing;
as shown in fig. 10, the experiment table in the present embodiment mainly includes a rolling bearing, a detachable bearing seat, a motor, a speed-adjusting device, and the like; in the experiment, a point corrosion fault is prefabricated on an inner ring, an outer ring and a rolling body by utilizing electric sparks respectively, a vibration signal of the rolling bearing is picked up by using a piezoelectric acceleration sensor arranged above a bearing seat in the experiment, the sampling frequency is 12kHz, and the signal acquisition is carried out through a data acquisition system after the amplification and the filtering;
in the embodiment, four health types including inner ring faults, outer ring faults and rolling body faults in a normal state of the motor side rolling bearing are classified;
the collecting method of this embodiment is the same as embodiment 1;
after enough data are obtained, the vibration acceleration signal is used for obtaining a time domain diagram of the vibration signal of each health type, namely the corresponding speed and displacement signal through integral calculation, as shown in fig. 11; as can be seen from fig. 11, the amplitudes of the three types of signals for each health type are greatly different;
s2: converting the data into an image;
the method for converting the adopted data into the image is the same as the embodiment 1, and RGB images of four health types are obtained; the size and arrangement of the data set were the same as in example 1;
s3: analyzing the diagnosis result;
the parameter setting of the model training process is the same as that of the embodiment 1; the graph of the relationship between the accuracy and the loss function value of the CBAM-ResNet model in the training set and the iteration times is shown in FIG. 12; as can be seen from fig. 12, by training the diagnostic model, as the number of iterations increases, the classification accuracy of the training set gradually increases and tends to be stable, and at the same time, the loss function smoothly decreases and tends to be stable;
then, inputting the data of the test set into a trained diagnosis model, and averaging the results through 10 random experiments to obtain the classification accuracy of the test set of 97.50%; to further understand the detailed classification of each health type, a confusion matrix of diagnostic results is plotted, as shown in FIG. 13; as can be seen from fig. 13, in the test set, there were 3 rolling element failures that were misclassified and all of the remaining three health types were correctly classified.
The feature that the vibration acceleration signal is utilized to obtain the corresponding speed and displacement through integral calculation in the invention is further explained through experiments of a local laboratory bearing fault data set, and the acceleration, speed and displacement signals are combined to form the RGB image with enhanced characteristics, so that the effectiveness, superiority and generalization performance are improved.
The acceleration signal, the velocity signal and the displacement signal are respectively and independently combined into an RGB image, the acceleration signal and the velocity signal are combined into an RGB image, the acceleration signal and the displacement signal are combined into an RGB image, and the velocity and the displacement signal are combined into an RGB image, 6 conditions are compared and analyzed with the condition of embodiment 2 (number 7 in table 3) of the invention, the obtained image data sets are respectively sent into a CBAM-ResNet rolling bearing fault diagnosis model for training and diagnosis, and the rolling bearing fault diagnosis precision is shown in table 3.
TABLE 3 Fault diagnosis accuracy of rolling bearing under different signal composition image data sets
Figure BDA0002803480530000131
As can be seen from table 3, the diagnosis accuracy of performing fault diagnosis using an image composed of acceleration signals is significantly higher than the diagnosis accuracy of performing fault diagnosis using an image composed of velocity signals and displacement signals; the diagnostic accuracy of the image formed by combining the acceleration signal with the velocity signal and the displacement signal respectively can reach 94.70% and 94.18%, respectively, but both are lower than the diagnostic accuracy of embodiment 2 of the invention.
Verification method
In order to further illustrate the performance superiority of the CBAM-ResNet model in the present invention, the present embodiment selects three common fault diagnosis models of SVM, BP, and CNN to perform comparative analysis with embodiment 2 of the present invention, and the parameter settings are the same as those in embodiment 1; the results are averaged through 10 random experiments; the results of the failure diagnosis are shown in table 4.
TABLE 4 comparison of the results of the experiments with different models
Figure BDA0002803480530000132
Figure BDA0002803480530000141
As can be seen from table 4, the classification accuracy obtained by the CBAM-ResNet diagnostic model in embodiment 2 of the present invention is improved by 15.83% compared with that obtained by the BP neural network, which brings an obvious improvement to the fault diagnosis accuracy of the rolling bearing, and also confirms that the CBAM-ResNet diagnostic model in the present invention can more accurately capture the implicit features of the data set.
The embodiment of the invention is verified and analyzed by two different embodiments, thereby illustrating the effectiveness and generalization capability of the invention; meanwhile, the invention omits the fussy manual feature extraction process in the diagnosis process, reduces the failure rate of the rolling bearing fault and realizes end-to-end fault diagnosis.
The method utilizes the characteristic that the convolutional neural network can enhance the nonlinear characterization capability of the fault diagnosis model, and introduces a channel attention mechanism and a space attention mechanism on the basis to model the nonlinear relation between the characteristics; the experimental results of the two examples show that: (1) the characteristic that the corresponding speed and displacement are obtained by utilizing the rolling bearing fault vibration acceleration signal through integral calculation is utilized, and an image with enhanced characteristics is obtained after the acceleration signal, the speed signal and the displacement signal are combined, so that the method can be used for fault diagnosis of the rolling bearing; (2) the established CBAM-ResNet diagnostic model can automatically extract features and complete end-to-end fault diagnosis of the rolling bearing; (3) compared with the existing data-driven fault diagnosis method, the method provided by the invention has better diagnosis precision and better robustness in the fault diagnosis of the rolling bearing.
Other parts not described belong to the prior art.

Claims (5)

1. The intelligent fault diagnosis method for the end-to-end rolling bearing by adopting a multi-attention machine system is characterized by comprising the following steps of: the fault feature extraction and the fault mode classification are combined, and the weighted expression of various fault features is realized through a multi-attention mechanism.
2. The intelligent fault diagnosis method for the end-to-end rolling bearing adopting the multi-attention mechanism is characterized by comprising the following steps of: consists of the following steps which are executed in sequence,
the method comprises the following steps: acquiring rolling bearing data;
step two: preprocessing rolling bearing data;
the rolling bearing data preprocessing is to convert vibration signals of the rolling bearing into images; the method for converting the vibration signal of the rolling bearing into the image comprises the following steps: obtaining corresponding speed and displacement through integral calculation of vibration acceleration signals of the rolling bearing, and combining the acceleration signals, the speed signals and the displacement signals to obtain an image with enhanced characteristics;
step three: constructing a CBAM-ResNet diagnostic model and analyzing and constructing a diagnostic result;
and sending the image in the step two into a CBAM-ResNet diagnostic model for training, and classifying the test data set by using the trained CBAM-ResNet diagnostic model.
3. The intelligent fault diagnosis method for the end-to-end rolling bearing adopting the multi-attention machine system according to claim 2 is characterized in that: the multiple attention mechanism employed is a channel attention and spatial attention structure.
4. The intelligent fault diagnosis method for the end-to-end rolling bearing adopting the multi-attention machine system according to claim 3 is characterized in that: in the second step, the method for converting the vibration signal into the image specifically comprises the following steps,
intercepting a vibration signal sample according to four categories of outer ring faults, inner ring faults, rolling body faults and normal conditions of the rolling bearing and according to the standard of 50% of an overlapped signal to obtain a vibration acceleration signal data set { ACi1, 2., mxmxmxl }, where L represents the total number of samples and mxm represents the pixel size of the image;
obtained by integration, a velocity signal data set VEi1,2, mxmxmxl, displacement signal data set { DIS |i|i＝1，2，...，M×M×L}；
Then, the vibration acceleration signal data set, the speed signal data set and the displacement signal data set are respectively substituted into the formula (6) for processing, so that the data range is converted into the range between [0,255],
Figure FDA0002803480520000011
in equation (6), P (m, m), m ═ 1, 2.. j denotes a pixel point of an image, the function round (·) is a rounding function, and x isiRepresenting the ith sample, x, of the data setminRepresenting the minimum value of a sample, x, in the datasetmaxRepresents a maximum value of samples in the dataset;
in the generated RGB image, the red channel pixel values are filled by the acceleration signal data set, the green channel pixel values are filled by the velocity signal data set, and the blue channel pixel values are filled by the displacement signal data set.
5. The intelligent fault diagnosis method for the end-to-end rolling bearing adopting the multi-attention machine system according to claim 4 is characterized in that: in the third step, in the original ResNet network structure, a CBAM attention module is added behind each group of residual blocks, and the output number of an output layer is set to be 4 so as to match four health types of the rolling bearing, namely the outer ring fault, the inner ring fault, the rolling body fault and the normal condition.
CN202011359124.0A 2020-11-27 2020-11-27 Intelligent fault diagnosis method for end-to-end rolling bearing by adopting multi-attention mechanism Active CN112304614B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011359124.0A CN112304614B (en) 2020-11-27 2020-11-27 Intelligent fault diagnosis method for end-to-end rolling bearing by adopting multi-attention mechanism

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011359124.0A CN112304614B (en) 2020-11-27 2020-11-27 Intelligent fault diagnosis method for end-to-end rolling bearing by adopting multi-attention mechanism

Publications (2)

Publication Number Publication Date
CN112304614A true CN112304614A (en) 2021-02-02
CN112304614B CN112304614B (en) 2023-04-25

Family

ID=74486967

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011359124.0A Active CN112304614B (en) 2020-11-27 2020-11-27 Intelligent fault diagnosis method for end-to-end rolling bearing by adopting multi-attention mechanism

Country Status (1)

Country Link
CN (1) CN112304614B (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113340598A (en) * 2021-06-01 2021-09-03 西安交通大学 Rolling bearing intelligent fault diagnosis method based on regularization sparse model
CN113449618A (en) * 2021-06-17 2021-09-28 南京航空航天大学 Method for carrying out deep learning rolling bearing fault diagnosis based on feature fusion and mixed enhancement
CN114118140A (en) * 2021-10-29 2022-03-01 新黎明科技股份有限公司 Multi-view intelligent fault diagnosis method and system for explosion-proof motor bearing
CN114992138A (en) * 2022-06-13 2022-09-02 湘潭大学 Method for diagnosing faults of water supply pump of industrial circulating water system based on GAN-CNN-BilSTM
CN114993677A (en) * 2022-05-11 2022-09-02 山东大学 Rolling bearing fault diagnosis method and system based on unbalanced small sample data
WO2023137807A1 (en) * 2022-01-21 2023-07-27 苏州大学 Rolling bearing class imbalance fault diagnosis method and system

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108334936A (en) * 2018-01-30 2018-07-27 华中科技大学 Failure prediction method based on migration convolutional neural networks
CN109765053A (en) * 2019-01-22 2019-05-17 中国人民解放军海军工程大学 Utilize the Fault Diagnosis of Roller Bearings of convolutional neural networks and kurtosis index
CN209085657U (en) * 2017-08-02 2019-07-09 强力物联网投资组合2016有限公司 For data gathering system related or industrial environment with chemical production technology
CN111833307A (en) * 2020-06-15 2020-10-27 湖南科技大学 Method and device for diagnosing cracks of wind turbine blade

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN209085657U (en) * 2017-08-02 2019-07-09 强力物联网投资组合2016有限公司 For data gathering system related or industrial environment with chemical production technology
CN108334936A (en) * 2018-01-30 2018-07-27 华中科技大学 Failure prediction method based on migration convolutional neural networks
CN109765053A (en) * 2019-01-22 2019-05-17 中国人民解放军海军工程大学 Utilize the Fault Diagnosis of Roller Bearings of convolutional neural networks and kurtosis index
CN111833307A (en) * 2020-06-15 2020-10-27 湖南科技大学 Method and device for diagnosing cracks of wind turbine blade

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
王威: "基于多尺度卷积神经网络的电机故障诊断方法研究", 《万方学位论文数据库》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113340598A (en) * 2021-06-01 2021-09-03 西安交通大学 Rolling bearing intelligent fault diagnosis method based on regularization sparse model
CN113340598B (en) * 2021-06-01 2024-05-28 西安交通大学 Rolling bearing intelligent fault diagnosis method based on regularized sparse model
CN113449618A (en) * 2021-06-17 2021-09-28 南京航空航天大学 Method for carrying out deep learning rolling bearing fault diagnosis based on feature fusion and mixed enhancement
CN114118140A (en) * 2021-10-29 2022-03-01 新黎明科技股份有限公司 Multi-view intelligent fault diagnosis method and system for explosion-proof motor bearing
CN114118140B (en) * 2021-10-29 2022-09-23 新黎明科技股份有限公司 Multi-view intelligent fault diagnosis method and system for explosion-proof motor bearing
WO2023137807A1 (en) * 2022-01-21 2023-07-27 苏州大学 Rolling bearing class imbalance fault diagnosis method and system
CN114993677A (en) * 2022-05-11 2022-09-02 山东大学 Rolling bearing fault diagnosis method and system based on unbalanced small sample data
CN114992138A (en) * 2022-06-13 2022-09-02 湘潭大学 Method for diagnosing faults of water supply pump of industrial circulating water system based on GAN-CNN-BilSTM
CN114992138B (en) * 2022-06-13 2024-03-26 湘潭大学 GAN-CNN-BiLSTM-based industrial circulating water system water supply pump fault diagnosis method

Also Published As

Publication number Publication date
CN112304614B (en) 2023-04-25

Similar Documents

Publication Publication Date Title
CN112304614B (en) Intelligent fault diagnosis method for end-to-end rolling bearing by adopting multi-attention mechanism
CN109765053B (en) Rolling bearing fault diagnosis method using convolutional neural network and kurtosis index
CN113191215B (en) Rolling bearing fault diagnosis method integrating attention mechanism and twin network structure
CN112200244B (en) Intelligent detection method for anomaly of aerospace engine based on hierarchical countermeasure training
CN110516305B (en) Intelligent fault diagnosis method under small sample based on attention mechanism meta-learning model
CN112763214B (en) Rolling bearing fault diagnosis method based on multi-label zero-sample learning
CN109858352A (en) A kind of method for diagnosing faults based on compressed sensing and the multiple dimensioned network of improvement
CN109782603A (en) The detection method and monitoring system of rotating machinery coupling fault
CN111046916A (en) Motor fault diagnosis method and system based on void convolution capsule network
CN106447039A (en) Non-supervision feature extraction method based on self-coding neural network
CN110399854B (en) Rolling bearing fault classification method based on mixed feature extraction
CN112257530A (en) Rolling bearing fault diagnosis method based on blind signal separation and support vector machine
CN112633195A (en) Myocardial infarction identification and classification method based on frequency domain features and deep learning
CN113158984B (en) Bearing fault diagnosis method based on complex Morlet wavelet and lightweight convolution network
CN111310719B (en) Unknown radiation source individual identification and detection method
CN114491823B (en) Train bearing fault diagnosis method based on improved generation countermeasure network
CN115905976B (en) Attention mechanism-based fault diagnosis method, system and equipment for Highway Bi-LSTM bearing
CN112257741A (en) Method for detecting generative anti-false picture based on complex neural network
CN116778272A (en) Hydropower unit small sample working condition identification method based on multichannel information fusion
CN112160877A (en) Fan bearing fault diagnosis method based on SFA and CNN
CN107451624A (en) The mechanized equipment health status recognition methods of network is locally connected based on orthogonalization
CN113655778A (en) Underwater propeller fault diagnosis system and method based on time-frequency energy
CN111783941A (en) Mechanical equipment diagnosis and classification method based on probability confidence degree convolutional neural network
CN114354195A (en) Rolling bearing fault diagnosis method of depth domain self-adaptive convolution network
CN114383846B (en) Bearing composite fault diagnosis method based on fault label information vector

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