CN112364706A - Small sample bearing fault diagnosis method based on class imbalance - Google Patents
Small sample bearing fault diagnosis method based on class imbalance Download PDFInfo
- Publication number
- CN112364706A CN112364706A CN202011116572.8A CN202011116572A CN112364706A CN 112364706 A CN112364706 A CN 112364706A CN 202011116572 A CN202011116572 A CN 202011116572A CN 112364706 A CN112364706 A CN 112364706A
- Authority
- CN
- China
- Prior art keywords
- sample
- class
- fault diagnosis
- samples
- fault
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000005096 rolling process Methods 0.000 claims abstract description 16
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 12
- 238000011160 research Methods 0.000 claims abstract description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 20
- 238000005070 sampling Methods 0.000 claims description 11
- 230000000694 effects Effects 0.000 claims description 10
- 230000011218 segmentation Effects 0.000 claims description 10
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 8
- 238000007635 classification algorithm Methods 0.000 claims description 6
- 230000001133 acceleration Effects 0.000 claims description 5
- 230000003321 amplification Effects 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 230000001174 ascending effect Effects 0.000 claims description 4
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 abstract description 3
- 238000004458 analytical method Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000003672 processing method Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000002950 deficient Effects 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- 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
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/259—Fusion by voting
-
- 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)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a small sample bearing fault diagnosis method based on class unbalance, which belongs to the field of vibration signal analysis and comprises the following steps: constructing a data set with class imbalance, and researching the influence of the class imbalance degree on the fault diagnosis performance; the one-dimensional original vibration signal is segmented through a sliding window, one sample is enhanced into a plurality of small samples with similar characteristics, and the utilization rate of sample points is improved; taking the plurality of enhanced samples as input, and extracting signal features by utilizing a deep convolutional neural network; classifying the segmented new samples by a voting method in ensemble learning, and setting the category with the largest number of labels as a final label corresponding to the original sample; and a plurality of indexes are adopted to evaluate the diagnosis result, so that the diagnosis is more real and reliable. The method aims to research the influence of class imbalance on the diagnosis performance, improve the feature extraction capability through data enhancement, improve the fault diagnosis capability through an integrated learning classification method and provide a basis for fault diagnosis of the rolling bearing.
Description
Technical Field
The invention relates to the technical field of vibration signal analysis, in particular to a small sample bearing fault diagnosis method based on class imbalance.
Background
Rolling bearings are one of the key components in rotating mechanical equipment with wide application, and the health condition of the rolling bearings can affect the safe and stable operation of the equipment. Due to the complex working condition, the severe environment and other reasons, parts are easy to damage, equipment is stopped and production is stopped, and even casualties are caused; meanwhile, the fault characteristics are easily submerged by noise, the characteristic extraction is difficult, and the fault diagnosis performance is poor; therefore, it becomes a precondition for realizing fault diagnosis by effectively extracting important information in the bearing signal.
Due to the characteristics of large equipment quantity, wide range, more measuring points of each equipment, high data sampling frequency and the like, the field of fault diagnosis is promoted to enter a big data era. The intelligent fault diagnosis is an important means for guaranteeing stable and reliable operation of the rotary machine under the drive of big data, and in order to accurately identify the health state of equipment, the intelligent diagnosis needs to rely on sufficient available monitoring data to train an intelligent fault diagnosis model. Under actual working conditions, the rotary machine is in a normal service state for a long time, and all fault types on the site are difficult to obtain at one time. The normal samples are rich, and the fault samples are deficient, so that the bearing sample data are in an unbalance-like state, namely the fault samples are far less than the normal samples.
At present, strategies for solving the class imbalance classification problem can be divided into two categories, namely a data level and an algorithm level, wherein the data level is used for changing the sample distribution of a training set and reducing the imbalance degree; the algorithm level considers the difference of the costs of different misclassification conditions and optimizes the algorithm, so that the algorithm can have a better diagnosis effect under the class imbalance data. However, the cost sensitive weight of various faults in the aspect of the algorithm is not easy to determine, and the subsequent fault diagnosis performance is influenced.
Disclosure of Invention
The invention provides a small sample bearing fault diagnosis method based on class unbalance, which is used for researching the influence of class unbalance degree on fault diagnosis performance, and adopts a sliding window to perform data segmentation, so that the sample size is expanded, and the utilization rate of sample points is improved; the deep convolutional neural network is used for extracting signal characteristics, fault classification is realized by combining with an integrated learning classification algorithm, and the fault diagnosis performance is improved.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a small sample bearing fault diagnosis method based on class unbalance comprises the following steps:
step 1, sampling original vibration data of a rolling bearing to obtain an original vibration signal, setting different class unbalance degrees, constructing data sets under different class unbalance ratios, and researching the influence of the unbalance ratios on fault diagnosis performance;
step 2, performing data segmentation on the obtained original vibration signal by using a sliding window, segmenting an original vibration signal sample into a plurality of small samples with similar characteristics, improving the utilization rate of original vibration signal sample points, and realizing data enhancement processing;
step 3, taking the new sample after data enhancement processing as input, and extracting signal characteristics by using a deep convolutional neural network; gradient reduction is realized by using an Adam algorithm, and the problems of sparse gradient and noise are solved;
step 4, classifying a plurality of samples obtained by segmenting original vibration signal data by using a voting classification method in ensemble learning, and voting out the class with the largest number of labels as the label of the original sample, so that the accuracy of fault classification is improved; and simultaneously evaluating the fault diagnosis effect by using a plurality of indexes.
The technical scheme of the invention is further improved as follows: in the step 1, the original vibration signals of the rolling bearing comprise normal signals, inner ring fault signals, outer ring fault signals and rolling body fault signals.
The technical scheme of the invention is further improved as follows: in step 1, the data set under the class imbalance ratio comprises normal samples of a majority class and fault samples of a minority class; the data set at class imbalance ratio was constructed as follows:
for the similar unbalance fault diagnosis research, constructing a data set under different similar unbalance rates R through experiments;
R＝Nd/Nn (1)
wherein N isdRepresenting the number of fault samples, NnThe number of samples representing a normal condition;
and inputting the data set under the class imbalance ratio into a deep convolutional neural network, performing fault diagnosis and classification through a softmax classification layer, and analyzing the influence of the class imbalance ratio on the fault diagnosis performance.
The technical scheme of the invention is further improved as follows: in step 2, the data enhancement processing specifically includes:
according to the sampling frequency, the motor rotating speed and the single sample length requirement, a sliding window and a sliding step length with proper sizes are selected to segment the original vibration signal; reconstructing an original vibration signal into a sample with a specified length so as to facilitate the expansion of the sample and the construction of a model; sliding a window along the one-dimensional time signal by utilizing the periodicity of the bearing fault signal; the original vibration signals refer to vibration acceleration signals acquired by an accelerometer, and comprise healthy bearings and bearing signals in various fault states; the number of samples after segmentation is calculated as follows:
where N represents the single sample length, N represents the window length, m represents the sliding step, and X represents the number of new samples after slicing.
The technical scheme of the invention is further improved as follows: in step 3, the deep convolutional neural network comprises an input layer, a plurality of convolutional layers, a pooling layer and a global average pooling layer.
The technical scheme of the invention is further improved as follows: in step 3, the extracting the signal features specifically includes:
constructing a deep convolution neural network by utilizing deep learning knowledge, and extracting signal characteristics through multilayer convolution and pooling; calculating first moment estimation and second moment estimation of the gradient by adopting an Adam algorithm, and designing independent adaptive learning rates for different parameters;
the first moment estimation and the second moment estimation are calculated according to the following formula:
mt＝β1mt-1+(1-β1)gt (3)
vt＝β2vt-1+(1-β2)gt 2 (4)
adam's update rule is:
where eta represents the step size, thetatAnd thetat+1Represents the weight of the t step and the t +1 step; the algorithm combines the optimal performance of AdaGrad and RMSProp, and can provide an optimization method for solving the problems of sparse gradient and noise.
The technical scheme of the invention is further improved as follows: in step 4, classifying the samples by an ensemble learning voting method, and identifying the fault type, and the specific steps are as follows:
selecting the label with the most output by adopting a hard voting classification algorithm, and if the number of the labels is equal, selecting the labels according to an ascending order, wherein the formula is as follows:
wherein i represents the i-th sample, X represents the amount of the sample after amplification, and yiIndicates the label corresponding to the ith sample, Y indicates the label type, I (h)i(x)＝yi) Representing the number of samples corresponding to each type of label; obtaining category labels corresponding to the X samples through an algorithm, counting the number of each category label, and setting the category with the largest number of labels as the label corresponding to the original sample; meanwhile, various indexes including F1 scores, recall rate, accuracy and the like are used for evaluation, and the classification effect is visualized through T-SNE.
Due to the adoption of the technical scheme, the invention has the technical progress that:
1. the invention directly extracts and identifies the characteristics of the one-dimensional original signal, thereby avoiding the problem of information loss when the original data is converted into frequency domain data.
2. The invention researches the influence of the class unbalance degree on the fault diagnosis performance, so that the research is more appropriate for practical application.
3. The data enhancement processing method adopted by the invention can enlarge the sample size, improve the utilization rate of sample points, further improve the characteristic extraction capability of the model to signals, and is suitable for small sample situations.
4. The integrated learning classification algorithm adopted by the invention is based on the majority theorem, and the obtained result is higher than the accuracy of any optimal model in the prediction model sequence.
Drawings
FIG. 1 is a flow chart of a diagnostic method of the present invention;
FIG. 2 is a time domain waveform diagram of a vibration signal obtained by an acceleration sensor in the present invention;
FIG. 3 is a schematic illustration of the accuracy of the class imbalance level versus the diagnostic performance study of a fault balance data set in accordance with the present invention;
FIG. 4 is a schematic illustration of the degree of class imbalance versus the accuracy of the imbalance data set for the results of the fault diagnosis performance study in accordance with the present invention;
FIG. 5 is a graph illustrating the effect of the sliding window slicing data of the present invention;
FIG. 6 is a graph of the fault diagnosis accuracy of an imbalance-like data set after the data enhancement and ensemble learning voting classification method of the present invention is employed;
FIG. 7 is a depiction of the t-SNE before treatment according to the invention;
FIG. 8 is a depiction of the t-SNE after treatment according to the invention.
Detailed Description
The invention provides a small sample bearing fault diagnosis method based on class unbalance, which is developed aiming at the problems that the cost sensitive weight of various faults is not easy to determine in the aspect of algorithms in the prior art, the subsequent fault diagnosis performance is influenced and the like.
The invention is described in further detail below with reference to the accompanying drawings:
a small sample bearing fault diagnosis method based on class imbalance is characterized in that: the method comprises the following steps:
step 1, sampling original vibration data of a rolling bearing to obtain an original vibration signal, setting different class unbalance degrees, constructing data sets under different class unbalance ratios, and researching the influence of the unbalance ratios on fault diagnosis performance;
the original vibration signals of the rolling bearing comprise normal signals, inner ring fault signals, outer ring fault signals and rolling body fault signals.
The data set at the class imbalance ratio comprises normal samples of a majority class and fault samples of a minority class; the data set at class imbalance ratio was constructed as follows:
for the similar unbalance fault diagnosis research, constructing a data set under different similar unbalance rates R through experiments;
R＝Nd/Nn (1)
wherein N isdRepresenting the number of fault samples, NnThe number of samples representing a normal condition;
and inputting the data set under the class imbalance ratio into a deep convolutional neural network, performing fault diagnosis and classification through a softmax classification layer, and analyzing the influence of the class imbalance ratio on the fault diagnosis performance.
Step 2, performing data segmentation on the obtained original vibration signal by using a sliding window, segmenting an original vibration signal sample into a plurality of small samples with similar characteristics, improving the utilization rate of original vibration signal sample points, and realizing data enhancement processing;
the data enhancement processing specifically includes:
according to the sampling frequency, the motor rotating speed and the single sample length requirement, a sliding window and a sliding step length with proper sizes are selected to segment the original vibration signal; reconstructing an original vibration signal into a sample with a specified length so as to facilitate the expansion of the sample and the construction of a model; sliding a window along the one-dimensional time signal by utilizing the periodicity of the bearing fault signal; the original vibration signals refer to vibration acceleration signals acquired by an accelerometer, and comprise healthy bearings and bearing signals in various fault states; the number of samples after segmentation is calculated as follows:
where N represents the single sample length, N represents the window length, m represents the sliding step, and X represents the number of new samples after slicing.
Step 3, taking the new sample after data enhancement processing as input, and extracting signal characteristics by using a deep convolutional neural network; gradient reduction is realized by using an Adam algorithm, and the problems of sparse gradient and noise are solved;
the extracting the signal features specifically includes:
constructing a deep convolution neural network by utilizing deep learning knowledge, and extracting signal characteristics through multilayer convolution and pooling; calculating first moment estimation and second moment estimation of the gradient by adopting an Adam algorithm, and designing independent adaptive learning rates for different parameters;
the first moment estimation and the second moment estimation are calculated according to the following formula:
mt＝β1mt-1+(1-β1)gt (3)
vt＝β2vt-1+(1-β2)gt 2 (4)
adam's update rule is:
where eta represents the step size, thetatAnd thetat+1Represents the weight of the t step and the t +1 step; the algorithm combines the optimal performance of AdaGrad and RMSProp, and can provide an optimization method for solving the problems of sparse gradient and noise.
Step 4, classifying a plurality of samples obtained by segmenting original vibration signal data by using a voting classification method in ensemble learning, and voting out the class with the largest number of labels as the label of the original sample, so that the accuracy of fault classification is improved; and simultaneously evaluating the fault diagnosis effect by using a plurality of indexes.
Classifying the samples by an ensemble learning voting method, and identifying fault types, wherein the method specifically comprises the following steps:
selecting the label with the most output by adopting a hard voting classification algorithm, and if the number of the labels is equal, selecting the labels according to an ascending order, wherein the formula is as follows:
wherein i represents the i-th sample, X represents the amount of the sample after amplification, and yiIndicates the label corresponding to the ith sample, Y indicates the label type, I (h)i(x)＝yi) Representing the number of samples corresponding to each type of label; obtaining category labels corresponding to the X samples through an algorithm, counting the number of each category label, and setting the category with the largest number of labels as the label corresponding to the original sample; meanwhile, various indexes including F1 scores, recall rate, accuracy and the like are used for evaluation, and the classification effect is visualized through T-SNE.
Specifically, the method comprises the following steps:
as shown in fig. 1, a method for diagnosing a fault of a small sample bearing based on class imbalance includes the following steps:
(1) sampling a fault signal of the rolling bearing, setting different class unbalance degrees, and constructing data sets under different class unbalance ratios, wherein the class unbalance data sets are composed of a plurality of normal samples and a few fault samples. Researching the influence of the class unbalance degree on the fault diagnosis performance according to the class unbalance data set; and inputting the unbalanced data set into a deep convolutional neural network, and judging the fault diagnosis performance through softmax classification.
(2) And carrying out data segmentation on the original signal by utilizing a sliding window to realize data enhancement.
(3) And taking the new sample after data enhancement as input, and extracting signal features by utilizing a deep convolution neural network.
(4) Classifying a plurality of samples obtained by segmenting a piece of data originally by using a voting method in ensemble learning, selecting and outputting the most labels, and if the number of the labels is equal, selecting according to an ascending order, wherein the formula is as follows:
wherein i represents the i-th sample, X represents the amount of the sample after amplification, and yiIndicates the label corresponding to the ith sample, Y indicates the label type, I (h)i(x)＝yi) Indicating the number of samples corresponding to each type of label. And obtaining class labels corresponding to the X samples through an algorithm, counting the number of each class of labels, and setting the class with the largest number of labels as the label corresponding to the original sample. Meanwhile, the fault diagnosis effect is evaluated by using a plurality of indexes.
Fig. 2 is a time-domain waveform diagram of the vibration signal obtained by the present invention. Multiple groups of vibration signals of the rolling bearing in different running states are obtained through the acceleration sensor, wherein the vibration signals comprise a normal running state, an inner ring fault running state, an outer ring fault running state and a rolling body fault running state. The data set is a vibration signal collected at a sampling frequency of 12.8kHz and a motor speed of 1920 r/min.
As shown in fig. 3 and 4, the influence of the class imbalance degree on the fault diagnosis performance is studied by the fault diagnosis accuracy of the balanced data set and the unbalanced data set. The balance data set is obtained by randomly intercepting 100 groups of sampling data from each type of vibration signals as sample data, wherein the length of each group of signals is 6000 sampling points, and 400 groups of vibration signals of the rolling bearing in different running states are obtained. The unbalanced data set is set with different class unbalanced ratios according to experimental requirements, and a corresponding class unbalanced data set is constructed.
As shown in fig. 5, a sliding window segmentation method is adopted to segment an original sample into a plurality of new samples with the same size and similar characteristics, so as to realize data enhancement. The number of samples after segmentation is:
wherein, N represents the length of a single sample, N represents the length of a window, m represents the sliding step length, and X represents the number of new samples after slicing.
Fig. 6 shows the fault diagnosis accuracy of the imbalance-like data set after the data enhancement and ensemble learning voting classification method of the present invention is adopted. The class imbalance ratios were 0.2, 0.4, 0.6, 0.8, 1, respectively. As can be seen, the method provided by the invention greatly improves the fault diagnosis performance of the imbalance-like data set.
As shown in fig. 7 and 8, when the imbalance ratio R is 0.2, the extracted fault features of the global mean pooling layer are reduced to a two-dimensional plane by using a t-SNE algorithm, and a visual graph is presented by using a scatter diagram, and the distance change between the inter-class distribution and the intra-class distribution of the imbalance-like data sets before and after using the method of the present invention can be observed through the t-SNE visual graph, so as to compare the effects of extracting and classifying the fault features before and after using the method of the present invention.
In conclusion, the method directly extracts and identifies the features of the one-dimensional original signals, researches the influence of the class imbalance degree on the fault diagnosis performance, enlarges the sample size and improves the utilization rate of sample points by adopting a data enhancement processing method, extracts the signal features by utilizing a deep convolutional neural network, realizes fault classification by combining an ensemble learning classification algorithm and improves the fault diagnosis performance.
Claims (7)
1. A small sample bearing fault diagnosis method based on class imbalance is characterized in that: the method comprises the following steps:
step 1, sampling original vibration data of a rolling bearing to obtain an original vibration signal, setting different class unbalance degrees, constructing data sets under different class unbalance ratios, and researching the influence of the unbalance ratios on fault diagnosis performance;
step 2, performing data segmentation on the obtained original vibration signal by using a sliding window, segmenting an original vibration signal sample into a plurality of small samples with similar characteristics, improving the utilization rate of original vibration signal sample points, and realizing data enhancement processing;
step 3, taking the new sample after data enhancement processing as input, and extracting signal characteristics by using a deep convolutional neural network; gradient reduction is realized by using an Adam algorithm, and the problems of sparse gradient and noise are solved;
step 4, classifying a plurality of samples obtained by segmenting original vibration signal data by using a voting classification method in ensemble learning, and voting out the class with the largest number of labels as the label of the original sample, so that the accuracy of fault classification is improved; and simultaneously evaluating the fault diagnosis effect by using a plurality of indexes.
2. The small sample bearing fault diagnosis method based on the class unbalance according to claim 1, characterized in that: in the step 1, the original vibration signals of the rolling bearing comprise normal signals, inner ring fault signals, outer ring fault signals and rolling body fault signals.
3. The small sample bearing fault diagnosis method based on the class unbalance according to claim 1, characterized in that: in step 1, the data set under the class imbalance ratio comprises normal samples of a majority class and fault samples of a minority class; the data set at class imbalance ratio was constructed as follows:
for the similar unbalance fault diagnosis research, constructing a data set under different similar unbalance rates R through experiments;
R＝Nd/Nn (1)
wherein N isdRepresenting the number of fault samples, NnThe number of samples representing a normal condition;
and inputting the data set under the class imbalance ratio into a deep convolutional neural network, performing fault diagnosis and classification through a softmax classification layer, and analyzing the influence of the class imbalance ratio on the fault diagnosis performance.
4. The small sample bearing fault diagnosis method based on the class unbalance according to claim 1, characterized in that: in step 2, the data enhancement processing specifically includes:
according to the sampling frequency, the motor rotating speed and the single sample length requirement, a sliding window and a sliding step length with proper sizes are selected to segment the original vibration signal; reconstructing an original vibration signal into a sample with a specified length so as to facilitate the expansion of the sample and the construction of a model; sliding a window along the one-dimensional time signal by utilizing the periodicity of the bearing fault signal; the original vibration signals refer to vibration acceleration signals acquired by an accelerometer, and comprise healthy bearings and bearing signals in various fault states; the number of samples after segmentation is calculated as follows:
where N represents the single sample length, N represents the window length, m represents the sliding step, and X represents the number of new samples after slicing.
5. The small sample bearing fault diagnosis method based on the class unbalance according to claim 1, characterized in that: in step 3, the deep convolutional neural network comprises an input layer, a plurality of convolutional layers, a pooling layer and a global average pooling layer.
6. The small sample bearing fault diagnosis method based on the class unbalance according to claim 1, characterized in that: in step 3, the extracting the signal features specifically includes:
constructing a deep convolution neural network by utilizing deep learning knowledge, and extracting signal characteristics through multilayer convolution and pooling; calculating first moment estimation and second moment estimation of the gradient by adopting an Adam algorithm, and designing independent adaptive learning rates for different parameters;
the first moment estimation and the second moment estimation are calculated according to the following formula:
mt＝β1mt-1+(1-β1)gt (3)
vt＝β2vt-1+(1-β2)gt 2 (4)
adam's update rule is:
where eta represents the step size, thetatAnd thetat+1Represents the weight of the t step and the t +1 step; the algorithm combines the optimal performance of AdaGrad and RMSProp, and can provide an optimization method for solving the problems of sparse gradient and noise.
7. The small sample bearing fault diagnosis method based on the class unbalance according to claim 1, characterized in that: in step 4, classifying the samples by an ensemble learning voting method, and identifying the fault type, and the specific steps are as follows:
selecting the label with the most output by adopting a hard voting classification algorithm, and if the number of the labels is equal, selecting the labels according to an ascending order, wherein the formula is as follows:
wherein i represents the i-th sample, X represents the amount of the sample after amplification, and yiIndicates the label corresponding to the ith sample, Y indicates the label type, I (h)i(x)＝yi) Representing the number of samples corresponding to each type of label; obtaining category labels corresponding to the X samples through an algorithm, counting the number of each category label, and setting the category with the largest number of labels as the label corresponding to the original sample; meanwhile, various indexes including F1 scores, recall rate, accuracy and the like are used for evaluation, and the classification effect is visualized through T-SNE.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011116572.8A CN112364706A (en) | 2020-10-19 | 2020-10-19 | Small sample bearing fault diagnosis method based on class imbalance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011116572.8A CN112364706A (en) | 2020-10-19 | 2020-10-19 | Small sample bearing fault diagnosis method based on class imbalance |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112364706A true CN112364706A (en) | 2021-02-12 |
Family
ID=74506861
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011116572.8A Pending CN112364706A (en) | 2020-10-19 | 2020-10-19 | Small sample bearing fault diagnosis method based on class imbalance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112364706A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113537460A (en) * | 2021-06-29 | 2021-10-22 | 哈尔滨工业大学 | Method for constructing multithreading neural network model suitable for flare prediction problem |
| CN113902745A (en) * | 2021-12-10 | 2022-01-07 | 山东捷瑞数字科技股份有限公司 | Method and device for identifying accurate fault of gearbox of commercial vehicle based on image processing |
| CN114418117A (en) * | 2022-01-20 | 2022-04-29 | 北京航空航天大学 | Meta-learning method with adaptive learning rate for few-sample fault diagnosis |
| 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 |
| CN116680550A (en) * | 2023-05-23 | 2023-09-01 | 南京航空航天大学 | Rolling bearing fault diagnosis method based on active learning under sample imbalance |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108646725A (en) * | 2018-07-31 | 2018-10-12 | 河北工业大学 | Dual model method for diagnosing faults based on dynamic weighting |
| CN109086412A (en) * | 2018-08-03 | 2018-12-25 | 北京邮电大学 | A kind of unbalanced data classification method based on adaptive weighted Bagging-GBDT |
| CN110307983A (en) * | 2019-06-19 | 2019-10-08 | 电子科技大学 | Unmanned plane Method for Bearing Fault Diagnosis based on CNN-Bagging |
| CN110333076A (en) * | 2019-06-19 | 2019-10-15 | 电子科技大学 | Method for Bearing Fault Diagnosis based on CNN-Stacking |
| CN110567720A (en) * | 2019-08-07 | 2019-12-13 | 东北电力大学 | method for diagnosing depth confrontation of fault of fan bearing under unbalanced small sample scene |
| CN110647871A (en) * | 2019-11-05 | 2020-01-03 | 燕山大学 | Rolling bearing fault diagnosis method and system based on time domain specific quantity enhancement |
| CN110704662A (en) * | 2019-10-17 | 2020-01-17 | 广东工业大学 | Image classification method and system |
| CN111091201A (en) * | 2019-12-23 | 2020-05-01 | 北京邮电大学 | Data partition mixed sampling-based unbalanced integrated classification method |
| CN111238815A (en) * | 2020-02-26 | 2020-06-05 | 中国民航大学 | Bearing fault identification method based on data enhancement under sample imbalance |
| CN111275024A (en) * | 2020-03-22 | 2020-06-12 | 东北电力大学 | Fault diagnosis method for rolling bearing of variable-working-condition fan driven by limited data containing noise label |
| CN111665066A (en) * | 2020-05-18 | 2020-09-15 | 东华大学 | Equipment fault self-adaptive upper and lower early warning boundary generation method based on convolutional neural network |
-
2020
- 2020-10-19 CN CN202011116572.8A patent/CN112364706A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108646725A (en) * | 2018-07-31 | 2018-10-12 | 河北工业大学 | Dual model method for diagnosing faults based on dynamic weighting |
| CN109086412A (en) * | 2018-08-03 | 2018-12-25 | 北京邮电大学 | A kind of unbalanced data classification method based on adaptive weighted Bagging-GBDT |
| CN110307983A (en) * | 2019-06-19 | 2019-10-08 | 电子科技大学 | Unmanned plane Method for Bearing Fault Diagnosis based on CNN-Bagging |
| CN110333076A (en) * | 2019-06-19 | 2019-10-15 | 电子科技大学 | Method for Bearing Fault Diagnosis based on CNN-Stacking |
| CN110567720A (en) * | 2019-08-07 | 2019-12-13 | 东北电力大学 | method for diagnosing depth confrontation of fault of fan bearing under unbalanced small sample scene |
| CN110704662A (en) * | 2019-10-17 | 2020-01-17 | 广东工业大学 | Image classification method and system |
| CN110647871A (en) * | 2019-11-05 | 2020-01-03 | 燕山大学 | Rolling bearing fault diagnosis method and system based on time domain specific quantity enhancement |
| CN111091201A (en) * | 2019-12-23 | 2020-05-01 | 北京邮电大学 | Data partition mixed sampling-based unbalanced integrated classification method |
| CN111238815A (en) * | 2020-02-26 | 2020-06-05 | 中国民航大学 | Bearing fault identification method based on data enhancement under sample imbalance |
| CN111275024A (en) * | 2020-03-22 | 2020-06-12 | 东北电力大学 | Fault diagnosis method for rolling bearing of variable-working-condition fan driven by limited data containing noise label |
| CN111665066A (en) * | 2020-05-18 | 2020-09-15 | 东华大学 | Equipment fault self-adaptive upper and lower early warning boundary generation method based on convolutional neural network |
Non-Patent Citations (2)
| Title |
|---|
| 曲建岭等: "基于一维卷积神经网络的滚动轴承自适应故障诊断算法", 《仪器仪表学报》 * |
| 金余丰: "基于深度学习的滚动轴承故障诊断方法研究", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113537460A (en) * | 2021-06-29 | 2021-10-22 | 哈尔滨工业大学 | Method for constructing multithreading neural network model suitable for flare prediction problem |
| CN113902745A (en) * | 2021-12-10 | 2022-01-07 | 山东捷瑞数字科技股份有限公司 | Method and device for identifying accurate fault of gearbox of commercial vehicle based on image processing |
| CN114418117A (en) * | 2022-01-20 | 2022-04-29 | 北京航空航天大学 | Meta-learning method with adaptive learning rate for few-sample fault diagnosis |
| CN114418117B (en) * | 2022-01-20 | 2024-05-28 | 北京航空航天大学 | Meta-learning method with adaptive learning rate for few-sample fault diagnosis |
| 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 |
| CN116680550A (en) * | 2023-05-23 | 2023-09-01 | 南京航空航天大学 | Rolling bearing fault diagnosis method based on active learning under sample imbalance |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112364706A (en) | Small sample bearing fault diagnosis method based on class imbalance | |
| WO2021135630A1 (en) | Rolling bearing fault diagnosis method based on grcmse and manifold learning | |
| Yang et al. | Bearing fault automatic classification based on deep learning | |
| CN112418013B (en) | Complex working condition bearing fault diagnosis method based on meta-learning under small sample | |
| CN107144430B (en) | A kind of Method for Bearing Fault Diagnosis based on incremental learning | |
| CN110070060B (en) | Fault diagnosis method for bearing equipment | |
| CN113375941B (en) | Open-set fault diagnosis method for bearings of high-speed motor train unit | |
| CN108830127A (en) | A kind of rotating machinery fault feature intelligent diagnostic method based on depth convolutional neural networks structure | |
| CN111562108A (en) | Rolling bearing intelligent fault diagnosis method based on CNN and FCMC | |
| CN107560849A (en) | A kind of Wind turbines Method for Bearing Fault Diagnosis of multichannel depth convolutional neural networks | |
| CN106874957A (en) | A kind of Fault Diagnosis of Roller Bearings | |
| CN112629851B (en) | Offshore wind turbine gearbox fault diagnosis method based on data enhancement method and image recognition | |
| Hu et al. | Bearing performance degradation assessment based on optimized EWT and CNN | |
| CN109186964A (en) | Rotary machinery fault diagnosis method based on angle resampling and ROC-SVM | |
| CN109932179A (en) | A kind of rolling bearing fault testing method based on the reconstruct of DS Adaptive spectra | |
| Chen et al. | A visualized classification method via t-distributed stochastic neighbor embedding and various diagnostic parameters for planetary gearbox fault identification from raw mechanical data | |
| Zhao et al. | A novel deep fuzzy clustering neural network model and its application in rolling bearing fault recognition | |
| CN114755017B (en) | Variable-speed bearing fault diagnosis method of cross-domain data driving unsupervised field shared network | |
| CN114112398A (en) | Fault diagnosis method for rolling bearing under variable speed working condition | |
| CN111898644A (en) | Intelligent identification method for health state of aerospace liquid engine under fault-free sample | |
| CN111076934A (en) | Method for diagnosing potential fault of bearing based on S transformation | |
| CN114462446A (en) | Rolling bearing fault diagnosis method based on vibration signal and electronic equipment | |
| CN113537152A (en) | Flow field state fault detection method based on deep neural network | |
| CN115221973A (en) | Aviation bearing fault diagnosis method based on enhanced weighted heterogeneous ensemble learning | |
| Li et al. | Transformer-based meta learning method for bearing fault identification under multiple small sample conditions |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210212 |
|
| RJ01 | Rejection of invention patent application after publication |