CN110702411B - Residual error network rolling bearing fault diagnosis method based on time-frequency analysis - Google Patents
Residual error network rolling bearing fault diagnosis method based on time-frequency analysis Download PDFInfo
- Publication number
- CN110702411B CN110702411B CN201910899012.5A CN201910899012A CN110702411B CN 110702411 B CN110702411 B CN 110702411B CN 201910899012 A CN201910899012 A CN 201910899012A CN 110702411 B CN110702411 B CN 110702411B
- Authority
- CN
- China
- Prior art keywords
- time
- frequency
- network
- residual error
- error network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
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
Landscapes
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- General Physics & Mathematics (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a residual error network rolling bearing fault diagnosis method based on time-frequency analysis, which comprises the following steps of: s1, collecting vibration signal data, converting vibration time domain signals of the rolling bearing into a time-frequency graph by using short-time Fourier transform, and converting the time-frequency graph into a two-dimensional gray-scale time-frequency graph; s2, extracting the characteristics of the signals by using a residual error network, and diagnosing the fault type of the bearing; the input of the residual error network is the gray-scale time-frequency diagram generated in step S1, and the output is the result of fault diagnosis. The invention adopts short-time Fourier transform to convert the bearing vibration data into a time-frequency diagram, can clearly reflect the time domain and frequency domain characteristics of the fault bearing during vibration, and is convenient for the network to accurately diagnose different fault types. Because the time-frequency signal contains the time-domain and frequency-domain information of the bearing at the same time, and the deepening of the network layer of the residual network does not cause the problem of gradient disappearance or gradient explosion, the method can obtain higher accuracy when the bearing is subjected to fault diagnosis.
Description
Technical Field
The invention relates to the field of bearing fault diagnosis, in particular to a residual error network rolling bearing fault diagnosis method based on time-frequency analysis.
Background
Rolling bearings are important components of mechanical parts, and a large number of rolling bearings are operated at any time on large-scale equipment or production lines. Once a rolling bearing has a serious failure, the precision of the product is difficult to control, and mechanical equipment or a production line is even stopped. Therefore, it is very important to perform failure diagnosis of the rolling bearing. When the bearing is in fault, the bearing can be found and maintained in time, and the operation reliability of the equipment is greatly improved.
The fault diagnosis of the rolling bearing can adopt the traditional method to carry out the means of feature extraction, fault classification and the like on the vibration signal. However, this method requires the relevant personnel to have rich prior knowledge, and when the vibration signal is mixed with the noise signal, the difficulty of extracting the signal feature is increased correspondingly. Since the fault diagnosis is to identify different fault types, and the deep learning achieves good achievement in the aspect of image identification, the deep learning method can be applied to the fault identification of the rolling bearing. The residual error network solves the problem that deep learning is degraded due to deepening of a network layer, so that the residual error network is used for fault diagnosis of the rolling bearing, high-dimensional characteristics of signals can be extracted by deepening the network layer, and the identification accuracy is improved.
However, the vibration signal for the rolling bearing is a signal with respect to time, i.e., the vibration signal for the rolling bearing contains only time domain information and no frequency domain information. If the vibration signal of the rolling bearing is directly input into the residual error network, the signal characteristics are lost, and the diagnosis accuracy is reduced. Therefore, the short-time Fourier transform can be adopted to convert the time domain signal of the bearing into a time frequency signal, the signal is a two-dimensional signal, the abscissa represents the time domain, and the ordinate represents the frequency domain. So that the input data of the residual network will contain both time domain information and frequency domain information. The network extracts the fault characteristics more comprehensively, which is beneficial to increasing the accuracy of fault diagnosis.
The application number 201710747694.9 is named as a rolling bearing fault diagnosis method based on a convolutional neural network, a short-time Fourier transform method is used for transforming time domain information into frequency domain information of a rolling bearing, the method can extract frequency domain characteristics of signals, but a common convolutional neural network is used in subsequent processing, and the problem that gradient disappears or gradient explosion occurs when the number of network layers is deepened in the network. Therefore, the number of network layers is limited, the high-dimensional characteristics of the bearing signals cannot be extracted, and the diagnosis accuracy is also limited.
The application number 201810339956.2 is named as a method for establishing an intelligent diagnosis model of a rolling bearing based on a convolutional neural network, although one-dimensional signals of the bearing are converted into two-dimensional signals, the conversion method is arranged in sequence, and the two-dimensional signals stacked in the method have no actual physical significance and cannot reflect time domain and frequency domain information of bearing faults. Therefore, the accuracy of bearing fault diagnosis cannot be improved.
In the university of Zhejiang Huangchangcheng Master academic paper, ResNet18 is adopted to diagnose the fault of the time-frequency signal of the bearing in the combined frequency analysis and convolutional neural network fault diagnosis optimization method study, the method directly identifies in a residual error network, does not modify the network structure aiming at the characteristics of the bearing, and does not visually analyze the network training process.
Disclosure of Invention
The invention aims to solve the technical problem of providing a residual error network rolling bearing fault diagnosis method based on time-frequency analysis, and the method can obtain higher accuracy rate when the method is used for carrying out fault diagnosis on a bearing.
The technical scheme adopted by the invention for solving the technical problems is as follows: a time-frequency analysis-based residual error network rolling bearing fault diagnosis method is constructed, and comprises the following steps:
s1, collecting vibration signal data, converting vibration time domain signals of the rolling bearing into a time-frequency graph by using short-time Fourier transform, and converting the time-frequency graph into a two-dimensional gray-scale time-frequency graph;
s2, extracting the characteristics of the signals by using a residual error network, and diagnosing the fault type of the bearing; the input of the residual error network is the gray-scale time-frequency diagram generated in step S1, and the output is the result of fault diagnosis.
In the above scheme, in step S1, each gray-scale time-frequency graph is labeled and divided into a training set and a test set.
In the above scheme, the method further includes step S3: inputting the data of the training set in the step S1 into a residual error network for training, wherein a cross entropy loss function is adopted as a loss function, and an optimization method is an Adam algorithm; and after training is finished, drawing characteristic diagrams of different network layers, and simultaneously performing dimension reduction visualization on the outputs of the different network layers by using a t-SNE algorithm to observe the relationship among the different network layers.
In the above scheme, the method further includes step S4: inputting the test set in the step S1 into the residual error network, and testing the accuracy of the residual error network.
In the above scheme, the short-time fourier transform of the vibration time-domain signal z (t) is:
where t is time, f is frequency, γ (t) is a window function, t' -t represents a sliding window, x represents a complex conjugate, and z (t) is a signal.
In the above scheme, the residual error network is a residual error network with 20 layers, the convolution kernel of the first layer of the residual error network is a convolution kernel of 5 × 5, and the convolution kernel of the later network layer is a convolution kernel of 3 × 3.
The residual error network rolling bearing fault diagnosis method based on time-frequency analysis has the following beneficial effects:
1. the invention adopts short-time Fourier transform to convert the bearing vibration data into a time-frequency diagram, can clearly reflect the time domain and frequency domain characteristics of the fault bearing during vibration, and is convenient for the network to accurately diagnose different fault types. Because the time-frequency signal contains the time-domain and frequency-domain information of the bearing at the same time, and the deepening of the network layer of the residual network does not cause the problem of gradient disappearance or gradient explosion, the method can obtain higher accuracy when the bearing is subjected to fault diagnosis.
2. According to the method, the characteristic graphs of different network layers are output, and a t-SNE visualization algorithm is adopted, so that the change condition of the network characteristics of different network layers during the diagnosis of the fault bearing can be clearly shown, and the adjustment of network parameters is facilitated.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a time-frequency diagram of different fault types;
FIG. 2 is a graph of accuracy and loss function values on a training set;
FIG. 3 is a diagram of input network characteristics;
FIG. 4 is a graph of first stage output characteristics;
FIG. 5 is a second stage output signature graph;
FIG. 6 is a third stage output signature diagram;
FIG. 7 is a fourth stage output signature;
FIG. 8 is a schematic view of a dimension reduction visualization of raw data;
FIG. 9 is a schematic diagram of a first stage output parameter dimension reduction visualization;
FIG. 10 is a schematic diagram of a second stage output parameter dimension reduction visualization;
FIG. 11 is a schematic diagram of a third stage output parameter dimension reduction visualization;
FIG. 12 is a schematic diagram of a fourth stage output parameter dimension reduction visualization.
Detailed Description
For a more clear understanding of the technical features, objects and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
The invention relates to a residual error network rolling bearing fault diagnosis method based on time-frequency analysis, which comprises the following steps of:
s1, acquiring and processing vibration signal data;
the data set used in this case is a bearing data set of the university of western reservoirs, and 10 types of failure are set in this example, as shown in table 1. The data for each fault type is then divided into 2048 x 1 samples, and finally each sample is converted to a 64 x 64 time-frequency grayscale map using a short-time fourier transform.
Given a window function γ (t) of short temporal width, sliding the window, the short-time fourier transform of the signal z (t) is:
where t is time, f is frequency, γ (t) is a window function, t' -t represents a sliding window, x represents a complex conjugate, and z (t) is a signal.
And converting the vibration time domain signal of the rolling bearing into a time frequency diagram by using short-time Fourier transform. Then, the time-frequency graph is converted into a two-dimensional time-frequency gray-scale graph shown in fig. 1, wherein the brighter the color in the time-frequency gray-scale graph, the larger the representative amplitude value, and the darker the color, the smaller the representative amplitude value. In the time-frequency gray-scale image, not only the time domain information of the vibration signal but also the frequency domain information of the vibration signal at different moments can be seen. And finally, labeling each time-frequency gray-scale image, and dividing the time-frequency gray-scale image into a training set and a testing set.
There are 1000 samples for each fault type, then 900 of them are randomly selected as training set, and the remaining 100 are test set. The time-frequency gray-scale images contain time-domain signals and frequency-domain signals when the bearing vibrates, so that the bearing vibration characteristics are more obvious, and the network can conveniently identify the fault type.
TABLE 1 bearing failure types
S2, constructing a residual error network;
the residual error network is a convolutional neural network which directly connects parameters with a network layer behind, and the network can effectively avoid the problem of network degradation caused by network deepening. The residual network can employ deeper network layers than a general convolutional neural network.
And constructing a 20-layer residual error network according to the residual error blocks of the residual error network, wherein in order to extract the low-frequency information of the bearing vibration, the convolution kernel of the first layer of the network is a large convolution kernel of 5 multiplied by 5, and the convolution kernel of the later layer of the network is a convolution kernel of 3 multiplied by 3. The input of the network is the gray-scale time-frequency diagram generated in S1, and the output is the result of fault diagnosis.
The residual network structure is as in table 2:
TABLE 2 residual network architecture
S3, training a residual error network;
and inputting the data of the training set in the S1 into a residual error network for training. The training steps are set to be 500 times, the trained Batch-Size is 50, the learning rate is 0.01, and the Adam algorithm is adopted as the network training algorithm. The training process of the residual error network is as shown in fig. 2, the network rapidly reaches a stable state along with the iteration, the accuracy on the training set also reaches 100%, and the loss function is gradually reduced.
The weight parameter of each layer will change after the network training, fig. 3 is a time-frequency diagram input into the network for diagnosis, and fig. 4 to 7 are feature diagrams output from the first stage to the fourth stage of the time-frequency diagram. Where each box represents an output channel. The contour of the time-frequency diagram can also be roughly seen when it passes through the first stage. However, as the network layer increases, the network starts to extract high-dimensional features, and it can be seen from the figure that the number of output channels of the network increases as the network deepens, and the features of the deep network are more abstract. This feature is mainly used for computer identification of different fault types.
Fig. 8 is an image obtained by performing dimension reduction visualization on raw data by using a t-SNE algorithm, and fig. 9 to 12 are images obtained by performing dimension reduction visualization on output features from a first stage to a fourth stage. Each color in the graph represents a fault type, and the farther the separation between each color represents the better the fault classification. It can be seen that the raw data of fig. 8 has intersections and adjacent portions between the various failure types, so the raw data cannot separate the failure types. In fig. 9 to 12, as the network layer deepens, various fault types start to be separated slowly, and by the output of the last stage, the fault types can be completely separated, and the separation between each fault type is far, so that interference is not easy to generate. The network classification effect is good.
S4, testing a residual error network;
after the residual error network training is finished, the test set in the S1 is input into the network for testing, and the accuracy rate of classification reaches 100%. In order to enable the network to have stable diagnostic effect even in a noisy environment, in this example, gaussian noise of-4 dB, -2dB, 0dB, 2dB, and 4dB is added to the data set for testing, and the test results are shown in table 3:
TABLE 3 test results in different noise environments
It can be seen from table 3 that the network also has a higher diagnostic accuracy in a high noise environment (-4dB), and the test accuracy gradually increases and the loss function value gradually decreases with the decrease of noise, and the accuracy can reach 100% when the noise is 4dB, which indicates that the noise environment has negligible characteristic interference to various fault types, and does not affect the diagnostic result of the network on the bearing.
S5, the residual network output is then used for fault diagnosis.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (2)
1. A residual error network rolling bearing fault diagnosis method based on time-frequency analysis is characterized by comprising the following steps:
s1, collecting vibration signal data, converting vibration time domain signals of the rolling bearing into a time-frequency graph by using short-time Fourier transform, and converting the time-frequency graph into a two-dimensional gray-scale time-frequency graph;
s2, extracting the characteristics of the signals by using a residual error network, and diagnosing the fault type of the bearing; inputting the gray-scale time-frequency diagram generated in the step S1 into the residual error network, and outputting the result of fault diagnosis;
in the step S1, labeling each gray-scale time-frequency graph, and dividing the graph into a training set and a test set;
further comprising step S3: inputting the data of the training set in the step S1 into a residual error network for training, wherein a cross entropy loss function is adopted as a loss function, and an optimization method is an Adam algorithm; drawing characteristic graphs of different network layers after training is completed, and simultaneously performing dimension reduction visualization on the output of the different network layers by using a t-SNE algorithm to observe the relationship among the different network layers;
a short-time Fourier transform of the vibro-time domain signal z (t) is:
wherein t is time, f is frequency, γ (t) is a window function, t' -t represents a sliding window, x represents a complex conjugate, and z (t) is a signal;
the residual error network is a residual error network with 20 layers, the convolution kernel of the first layer of the residual error network is a convolution kernel of 5 multiplied by 5, and the convolution kernel of the later network layer is a convolution kernel of 3 multiplied by 3.
2. The time-frequency analysis based residual error network rolling bearing fault diagnosis method according to claim 1, further comprising the step S4: inputting the test set in the step S1 into the residual error network, and testing the accuracy of the residual error network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910899012.5A CN110702411B (en) | 2019-09-23 | 2019-09-23 | Residual error network rolling bearing fault diagnosis method based on time-frequency analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910899012.5A CN110702411B (en) | 2019-09-23 | 2019-09-23 | Residual error network rolling bearing fault diagnosis method based on time-frequency analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110702411A CN110702411A (en) | 2020-01-17 |
| CN110702411B true CN110702411B (en) | 2020-11-10 |
Family
ID=69195840
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910899012.5A Active CN110702411B (en) | 2019-09-23 | 2019-09-23 | Residual error network rolling bearing fault diagnosis method based on time-frequency analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110702411B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738068B (en) * | 2020-05-13 | 2024-04-02 | 南京航空航天大学 | Transmission shaft fault diagnosis method and system under rotating speed fluctuation working condition |
| CN112577744B (en) * | 2020-11-02 | 2021-12-03 | 西南交通大学 | Rolling bearing fault identification method based on combination of SPA-map and ResNet |
| CN112668417B (en) * | 2020-12-17 | 2023-02-03 | 武汉理工大学 | Rolling bearing intelligent fault diagnosis method based on vibration signals |
| CN112924177B (en) * | 2021-04-02 | 2022-07-19 | 哈尔滨理工大学 | Rolling bearing fault diagnosis method for improved deep Q network |
| CN113405799B (en) * | 2021-05-20 | 2022-06-28 | 新疆大学 | Bearing early fault detection method based on health state index construction and fault early warning limit self-learning |
| CN113537044B (en) * | 2021-07-14 | 2022-08-26 | 哈尔滨理工大学 | Aircraft engine fault diagnosis method based on STFT and improved DenseNet |
| CN113705407B (en) * | 2021-08-20 | 2024-05-28 | 南京航空航天大学 | Fault diagnosis method for variable rotation speed lower shaft crack based on time-frequency residual error neural network |
| CN114091549A (en) * | 2021-09-28 | 2022-02-25 | 国网江苏省电力有限公司苏州供电分公司 | Equipment fault diagnosis method based on deep residual error network |
| CN113916535B (en) * | 2021-10-27 | 2022-08-09 | 河北建投能源投资股份有限公司 | Bearing diagnosis method, system, equipment and medium based on time frequency and CNN |
| CN113869286B (en) * | 2021-12-01 | 2022-02-25 | 中国工程物理研究院电子工程研究所 | Self-adaptive multi-task intelligent fault diagnosis system and fault diagnosis method |
| US11859626B2 (en) * | 2022-01-24 | 2024-01-02 | Hamilton Sundstrand Corporation | Incipient compressor surge detection using artificial intelligence |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108334948A (en) * | 2018-02-09 | 2018-07-27 | 武汉理工大学 | A kind of mechanical bearing fault diagnosis technology based on wide residual error network learning model |
| CN108830127A (en) * | 2018-03-22 | 2018-11-16 | 南京航空航天大学 | A kind of rotating machinery fault feature intelligent diagnostic method based on depth convolutional neural networks structure |
| CN109214356A (en) * | 2018-09-29 | 2019-01-15 | 南京东振测控技术有限公司 | A kind of fan transmission system intelligent fault diagnosis method based on DCNN model |
| CN109299705A (en) * | 2018-10-24 | 2019-02-01 | 电子科技大学 | Rotary machinery fault diagnosis method based on one-dimensional depth residual error convolutional neural networks |
| CN109946080A (en) * | 2019-04-08 | 2019-06-28 | 西安交通大学 | A kind of mechanical equipment health status recognition methods based on embedded recirculating network |
| CN110031226A (en) * | 2019-04-12 | 2019-07-19 | 佛山科学技术学院 | A kind of diagnostic method and device of bearing fault |
-
2019
- 2019-09-23 CN CN201910899012.5A patent/CN110702411B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108334948A (en) * | 2018-02-09 | 2018-07-27 | 武汉理工大学 | A kind of mechanical bearing fault diagnosis technology based on wide residual error network learning model |
| CN108830127A (en) * | 2018-03-22 | 2018-11-16 | 南京航空航天大学 | A kind of rotating machinery fault feature intelligent diagnostic method based on depth convolutional neural networks structure |
| CN109214356A (en) * | 2018-09-29 | 2019-01-15 | 南京东振测控技术有限公司 | A kind of fan transmission system intelligent fault diagnosis method based on DCNN model |
| CN109299705A (en) * | 2018-10-24 | 2019-02-01 | 电子科技大学 | Rotary machinery fault diagnosis method based on one-dimensional depth residual error convolutional neural networks |
| CN109946080A (en) * | 2019-04-08 | 2019-06-28 | 西安交通大学 | A kind of mechanical equipment health status recognition methods based on embedded recirculating network |
| CN110031226A (en) * | 2019-04-12 | 2019-07-19 | 佛山科学技术学院 | A kind of diagnostic method and device of bearing fault |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110702411A (en) | 2020-01-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110702411B (en) | Residual error network rolling bearing fault diagnosis method based on time-frequency analysis | |
| CN104819846B (en) | Rolling bearing sound signal fault diagnosis method based on short-time Fourier transform and sparse laminated automatic encoder | |
| CN109765053B (en) | Rolling bearing fault diagnosis method using convolutional neural network and kurtosis index | |
| Jiang et al. | Bearing fault classification based on convolutional neural network in noise environment | |
| CN109632309A (en) | Based on the rolling bearing fault intelligent diagnosing method for improving S-transformation and deep learning | |
| CN111160167A (en) | Spindle fault classification and identification method based on S-transform deep convolutional neural network | |
| CN108256556A (en) | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network | |
| CN112304614B (en) | Intelligent fault diagnosis method for end-to-end rolling bearing by adopting multi-attention mechanism | |
| Yi et al. | Mechanical compound faults extraction based on improved frequency domain blind deconvolution algorithm | |
| CN112378660A (en) | Intelligent fault diagnosis method for aero-engine bearing based on data driving | |
| CN110991295A (en) | Self-adaptive fault diagnosis method based on one-dimensional convolutional neural network | |
| CN112329914B (en) | Fault diagnosis method and device for buried transformer substation and electronic equipment | |
| CN111275108A (en) | Method for performing sample expansion on partial discharge data based on generation countermeasure network | |
| CN113472390B (en) | Frequency hopping signal parameter estimation method based on deep learning | |
| Wang et al. | Construction of the efficient attention prototypical net based on the time–frequency characterization of vibration signals under noisy small sample | |
| CN117030263A (en) | Bearing fault diagnosis method based on improved residual error network under multi-sensor signal fusion | |
| CN111291918A (en) | Rotating machine degradation trend prediction method based on stationary subspace exogenous vector autoregression | |
| CN114707534A (en) | Rolling bearing fault diagnosis method under small sample data | |
| CN117592543A (en) | Aeroengine rolling bearing fault diagnosis method based on self-supervision learning | |
| CN113657244A (en) | Fan gearbox fault diagnosis method and system based on improved EEMD and speech spectrum analysis | |
| CN110044619B (en) | Multi-fault feature identification method based on sparse multi-cycle group lasso | |
| Martin-del-Campo et al. | Towards zero-configuration condition monitoring based on dictionary learning | |
| CN110347579B (en) | Deep learning test case selection method based on neuron output behavior pattern | |
| CN110160778A (en) | Gearbox fault state identification method based on sequential hypothesis testing | |
| CN116484176A (en) | Bearing fault diagnosis method, system and storage medium based on ultra-wavelet |
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 |