CN114548153A - Planetary gearbox fault diagnosis method based on residual error-capsule network - Google Patents
Planetary gearbox fault diagnosis method based on residual error-capsule network Download PDFInfo
- Publication number
- CN114548153A CN114548153A CN202210072530.1A CN202210072530A CN114548153A CN 114548153 A CN114548153 A CN 114548153A CN 202210072530 A CN202210072530 A CN 202210072530A CN 114548153 A CN114548153 A CN 114548153A
- Authority
- CN
- China
- Prior art keywords
- convolution
- layer
- capsule
- time
- module
- 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
Links
- 239000002775 capsule Substances 0.000 title claims abstract description 105
- 238000003745 diagnosis Methods 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 10
- 238000010586 diagram Methods 0.000 claims description 36
- 238000012545 processing Methods 0.000 claims description 13
- 239000013598 vector Substances 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 6
- 230000007246 mechanism Effects 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 230000000903 blocking effect Effects 0.000 claims description 2
- 230000009466 transformation Effects 0.000 abstract description 2
- 230000001131 transforming effect Effects 0.000 abstract description 2
- 238000010801 machine learning Methods 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000036039 immunity Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000013508 migration Methods 0.000 description 2
- 230000005012 migration Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000005299 abrasion Methods 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012733 comparative method Methods 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000004901 spalling Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
Images
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4084—Scaling of whole images or parts thereof, e.g. expanding or contracting in the transform domain, e.g. fast Fourier transform [FFT] domain scaling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
- Y04S10/52—Outage or fault management, e.g. fault detection or location
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a fault diagnosis method for a planetary gear box based on a residual error-capsule network, which comprises the steps of firstly collecting time-domain vibration signals in a normal state and a fault state of the planetary gear box, carrying out zooming to obtain a time-frequency image after obtaining an original time-frequency image through transformation, and setting a label of the time-frequency image to obtain a training sample; constructing a residual error-capsule network comprising a convolution module, a residual error module, an attention module and a capsule module, training the residual error-capsule network by adopting a training sample, acquiring a time domain vibration signal and transforming to obtain a time-frequency image when fault diagnosis is required to be performed on the planetary gear box, and inputting the time-frequency image into the trained residual error-capsule network to obtain a fault diagnosis result. The invention can improve the fault diagnosis performance of the planetary gearbox.
Description
Technical Field
The invention belongs to the technical field of planetary gearbox fault diagnosis, and particularly relates to a planetary gearbox fault diagnosis method based on a residual error-capsule network.
Background
The planetary gear box is a common device in a rotary mechanical system, and is often used as a core power node in large-scale mechanical equipment due to the characteristics of small and exquisite design, high transmission ratio, stable working process and the like. Gearboxes are generally composed of bearing components, gear components and transmission components, among which the proportion of gear failure accounts for more than 60% of the gearbox component failure ratio according to prior research. This is because gearboxes operate in a harsh environment of high temperature, high corrosion throughout the year, resulting in gears that are highly susceptible to pitting, cracking, and spalling. If the fault is not timely discovered and processed during the operation of the equipment, the fault is further expanded, the whole equipment stops working if the fault is light, and heavy property loss and personnel life safety are caused if the fault is serious.
At present, most of the traditional fault diagnosis methods of the planetary gearbox start from vibration signals, and the vibration signals are one of the simplest and most intuitive reflections of the state of the gearbox. When the vibration signal is abnormal, the gear box is bound to operate in an unstable state, and the vibration signal is a necessary condition for accurately detecting the fault of the gear box. The initial planetary gear fault diagnosis method mainly analyzes various signal index statistics such as time domain, frequency domain, time-frequency domain and the like of vibration signals, and diagnoses faults from the value difference by extracting characteristic values of different fault signals. However, in practical conditions, there are situations of complex signals and excessive interference, and experts and scholars face the problem that the signals cannot be analyzed and processed.
After the twenty-first century, machine learning methods were applied to the field of fault diagnosis due to the rapid development of computer technology. Compared with the traditional numerical analysis method, the method has the advantages that the fault diagnosis by using a machine learning means is more intelligent, and the diagnosis result is relatively more reliable, such as a K nearest neighbor method, a Bayesian theory, a support vector machine and the like. There are also problems with machine learning methods. Firstly, the machine learning technology is unfriendly to support big data, and because the traditional machine learning technology needs to completely load data into a memory, when the data size is large, special processing must be carried out on the data; secondly, for the vibration signals, the machine learning method requires feature extraction on the data, and besides the traditional time-frequency extraction, the extraction method determines the diagnosis result to a great extent; besides, the adaptability of machine learning is poor, model migration is difficult, migration from a known field to an unknown field is difficult, and slight changes of signals are easy to generate misjudgment.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a fault diagnosis method of a planetary gear box based on a residual error-capsule network, and constructs the residual error-capsule network by combining the residual error network and the capsule network, thereby improving the fault diagnosis performance of the planetary gear box.
In order to achieve the above object, the method for diagnosing the fault of the planetary gearbox based on the residual error-capsule network comprises the following steps:
s1: setting K different fault states of the planetary gearbox according to needs, respectively acquiring a plurality of time domain vibration signals with preset time length T in a normal state and each fault state, then obtaining two-dimensional time frequency data for each time domain vibration signal through short-time Fourier transform to generate an original time frequency image, and then zooming the obtained original time frequency image to a preset size to obtain a time frequency image; setting the state label of each time-frequency image as a corresponding state code K, wherein K is 0 to represent a normal state, K is 1,2, …, and K represents a fault state serial number;
s2: constructing a residual error-capsule network, comprising a convolution module, a residual error module, an attention module and a capsule module, wherein:
the convolution module is used for performing convolution operation on the input time-frequency image, realizing feature extraction and data dimension reduction, and outputting the obtained feature image to the residual error module;
the residual error module is used for processing the received characteristic diagram and outputting the obtained characteristic diagram to the attention module;
the attention module is used for acquiring a channel characteristic weight value from the received characteristic diagram by adopting a channel attention mechanism, then performing characteristic redistribution on the input characteristic diagram according to the acquired channel characteristic weight value and sending the generated characteristic diagram to the capsule module;
the capsule module is used for estimating a fault state corresponding to the input time-frequency image according to the received characteristic diagram; the capsule module includes main capsule layer, digital capsule layer and output layer, wherein:
the main capsule layer is used for extracting the characteristics of the characteristic diagram from the attention module and outputting the obtained characteristic diagram to the digital capsule module;
the digital capsule layer comprises K +1 digital capsules, each digital capsule carries out feature extraction on the received feature map to obtain a one-dimensional vector, and the obtained K +1 one-dimensional vectors are output to the output layer;
the output layer is used for compressing the received K +1 one-dimensional vectors, namely solving the module length of the corresponding digital capsule to obtain the probability value of the input time-frequency image belonging to K +1 states;
s3: training the residual error-capsule network constructed in the step S2 by taking each time-frequency image obtained in the step S1 as input and the corresponding fault state label as expected output to obtain a trained residual error-capsule network;
s4: when fault diagnosis needs to be carried out on the planetary gearbox, a time domain vibration signal with the time length of T is collected, time-frequency image conversion and scaling are carried out by adopting the same method in the step S1, and the obtained time-frequency image is input into the trained residual error-capsule network to obtain a fault diagnosis result.
The invention relates to a fault diagnosis method of a planetary gear box based on a residual error-capsule network, which comprises the steps of firstly collecting time domain vibration signals in a normal state and a fault state of the planetary gear box, zooming to obtain a time-frequency image after obtaining an original time-frequency image through transformation, and setting a label of the time-frequency image to obtain a training sample; constructing a residual error-capsule network comprising a convolution module, a residual error module, an attention module and a capsule module, training the residual error-capsule network by adopting a training sample, acquiring a time domain vibration signal and transforming to obtain a time-frequency image when fault diagnosis is required to be performed on the planetary gear box, and inputting the time-frequency image into the trained residual error-capsule network to obtain a fault diagnosis result.
The invention has the following beneficial effects:
1) according to the invention, the fault time-frequency image is manufactured in a time-frequency analysis method and a scaling mode, compared with the traditional image manufacturing means, manual cutting is avoided, and full-automatic operation from collection to manufacturing of the data set is realized;
2) aiming at the defects of the traditional convolutional neural network, the capsule network is extracted and used for fault diagnosis and classification, and vector neurons in the capsule network can learn not only the size information of the features but also the position information among the features;
3) the invention further provides an improved residual error module, which uses two small convolution kernels to replace a large convolution kernel, increases the discrimination capability of the network while reducing the network operation amount by utilizing the advantage of asymmetric convolution, and can still achieve better diagnosis effect under the noise environment by combining with the capsule network.
Drawings
FIG. 1 is a flow diagram of an embodiment of a method for diagnosing planetary gearbox faults based on a residual error-capsule network;
FIG. 2 is an exemplary diagram of a time-frequency image before and after zooming in the present embodiment;
FIG. 3 is a block diagram of a residual-capsule network of the present invention;
FIG. 4 is a structural diagram of a convolution module in the present embodiment;
FIG. 5 is a structural diagram of a residual module in the present embodiment;
fig. 6 is an exemplary diagram of the normalized time-domain vibration signal in the present embodiment.
Detailed Description
The following description of the embodiments of the present invention is provided in order to better understand the present invention for those skilled in the art with reference to the accompanying drawings. It is to be expressly noted that in the following description, a detailed description of known functions and designs will be omitted when it may obscure the subject matter of the present invention.
Examples
FIG. 1 is a flow chart of an embodiment of a method for diagnosing faults of a planetary gearbox based on a residual error-capsule network. As shown in FIG. 1, the method for diagnosing the fault of the planetary gearbox based on the residual error-capsule network comprises the following specific steps:
s101: acquiring a time-frequency image sample:
setting K different fault states of the planetary gearbox according to needs, respectively acquiring a plurality of time domain vibration signals with preset duration T under a normal state and each fault state, then obtaining two-dimensional time frequency data for each time domain vibration signal through short-time Fourier transform, generating an original time frequency image, and then zooming the obtained original time frequency image to a preset size to obtain the time frequency image. The label of each time-frequency image sample is set as the corresponding state code K, wherein K is 0 to represent the normal state, K is 1,2, …, and K represents the fault state serial number.
In order to make the subsequent feature extraction more accurate, the time domain vibration signal may be subjected to a normalization preprocessing, such as Min-max normalization, z-score normalization, etc.
Fig. 2 is an exemplary diagram of a time-frequency image before and after zooming in the embodiment. As shown in fig. 2, in this embodiment, an original time-frequency image is scaled by using a bilinear interpolation algorithm, and the scaled image size is 100 × 100.
S102: constructing a residual error-capsule network:
in order to improve the accuracy of fault diagnosis of the planetary gearbox, the invention combines a residual error network and a capsule network to construct a residual error-capsule network. Fig. 3 is a block diagram of the residual-capsule network of the present invention. As shown in fig. 3, the residual-capsule network includes a convolution module, a residual module, an attention module, and a capsule module, each of which is described in detail below.
The convolution module is used for performing convolution operation on the input time-frequency image, realizing feature extraction and data dimension reduction, and outputting the obtained feature map to the residual error module. Fig. 4 is a structural diagram of a convolution module in the present embodiment. As shown in fig. 4, the convolution module in this embodiment includes a first convolution layer, a pooling layer, and a second convolution layer, where the first convolution layer performs convolution on an input time-frequency image using a convolution kernel with a size of 7 × 7, a convolution step is 2, and outputs an obtained feature map to the pooling layer; the pooling layer performs pooling on the input feature map by adopting a maximum pooling mode with the size of 3 × 3, the pooling step length is 2, and the obtained feature map is output to the second convolution layer; and the second convolution layer is convolved by adopting a convolution kernel with the size of 1 x 1, the convolution step is 1, and the obtained characteristic diagram is taken as the characteristic diagram of the convolution module to be output.
The residual error module is used for processing the received characteristic diagram and outputting the obtained characteristic diagram to the attention module. In order to improve the noise immunity of the network and ensure that the network still has excellent diagnosis effect under a noisy environment, the invention further improves the structure of the residual error module. Fig. 5 is a structural diagram of the residual block in the present embodiment. As shown in fig. 5, the residual error module in this embodiment includes a first convolution layer, a block convolution layer, a connection layer, and a second convolution layer, where:
the first convolution layer is used for convolving the received feature map by a convolution kernel with the size of 1 x 1, and the obtained feature map is respectively output to the blocking convolution layer and the second convolution layer.
The block convolution layer is used for carrying out 4-path splitting on the received feature graph according to the channels, and respectively recording the obtained 4-path features as X1、X2、X3、X4Is processed by block convolution layerObtain 4-way feature Y1、Y2、Y3、Y4The specific method comprises the following steps:
characteristic Y1Is equal to X1;
Characteristic Y2From the feature X2Sequentially carrying out convolution processing on two convolution kernels with the sizes of 1 x 3 and 3 x 1 to obtain the convolution kernel;
characteristic Y3From the feature X3And feature Y2Adding the two convolution kernels with the sizes of 1 × 3 and 3 × 1 in sequence, and performing convolution processing to obtain the product;
characteristic Y4From the feature X4And feature Y3Adding the two convolution kernels with the sizes of 1 × 3 and 3 × 1 in sequence, and performing convolution processing to obtain the product;
will 4 way feature Y1、Y2、Y3、Y4And outputting to the connecting layer.
And the connecting layer is used for splicing the received 4 paths of characteristics to obtain a characteristic diagram and sending the characteristic diagram to the second convolution layer.
And the second convolution layer adds the two received feature maps and then carries out activation processing, and the obtained feature map is used as the feature map of the residual error module to be output. In this embodiment, Relu activation is used for the activation of the second convolutional layer.
The attention module is used for acquiring a channel feature weight value from the received feature map by adopting a channel attention mechanism, then performing feature redistribution on the input feature map according to the acquired channel feature weight value, and sending the generated feature map to the capsule module. The channel attention mechanism can independently extract related information among all channels through the compression excitation module, gives different attention degrees to different channels in a weight mode, and can be embedded on the premise of not damaging an original neural network compared with other structure compression excitation modules. In the embodiment, the channel attention mechanism adopts Sigmoid and Relu activation.
And the capsule module is used for estimating the fault state corresponding to the input time-frequency image according to the received characteristic diagram.
As shown in fig. 2, the capsule module of the present invention comprises a main capsule layer, a digital capsule layer and an output layer, wherein:
the main capsule layer is used for extracting features of the feature map from the attention module and outputting the obtained feature map to the digital capsule layer.
The digital capsule layer comprises K +1 digital capsules, each digital capsule carries out feature extraction on the received feature map to obtain a one-dimensional vector, and the obtained K +1 one-dimensional vectors are output to the output layer.
The output layer is used for compressing the received K +1 one-dimensional vectors, namely solving the modular length of the corresponding digital capsule to obtain the probability value of the input time-frequency image belonging to K +1 states.
The main capsule layer and the digital capsule layer in the capsule module are key components in a capsule network, the capsule network is a novel neural network, and simply, the neurons in the neural network are replaced by capsules, namely the capsule network. The main capsule layer and the digital capsule layer are both composed of capsules, and the basic operation of each capsule comprises the following steps:
1) multiplying the input vector by a weight matrix, which encodes the spatial relationship of the low-level features and the high-level features;
2) weighting the input vector, using the weights to determine which higher level capsule the current capsule outputs to, is achieved by dynamic routing. The dynamic routing algorithm is the core of the transfer of the lower-level capsule to the higher-level capsule, the idea being to update the coupling coefficients by iterating continuously.
3) Summing after weighting;
4) nonlinear activation, using the square function: the vector is compressed to have a length between 0 and 1, and the direction remains unchanged.
For the principle of the capsule network and the working process of the Capsules, reference may be made to the document "Sabour S, frost N, Hinton G E.dynamic Routing Between Capsules [ J ].2017 ].
S103: training residual-capsule network:
and (4) training the residual error-capsule network constructed in the step (S102) by taking each time-frequency image obtained in the step (S101) as input and the corresponding fault state label as expected output to obtain the trained residual error-capsule network.
S104: fault diagnosis of the planetary gearbox:
when fault diagnosis needs to be carried out on the planetary gearbox, a time domain vibration signal with the time length of T is collected, time-frequency image conversion and scaling are carried out by adopting the same method in the step S101, and the obtained time-frequency image is input into the trained residual error-capsule network to obtain a fault diagnosis result.
Examples
In order to better illustrate the technical scheme and the technical effect of the invention, a specific example is adopted to analyze and illustrate the work flow and the technical effect of the invention. The planetary gearbox fault is a typical fault in a rotary machine, and the normal state and common 7 faults of the planetary gearbox are simulated by adopting a self-built fault simulation platform in the embodiment, wherein the normal state and common 7 faults comprise planetary wheel pitting, planetary wheel tooth breakage, planetary wheel tooth root crack, planetary wheel abrasion, gear ring tooth root breakage, gear ring crack and sun gear tooth root breakage. The acquired data are all one-dimensional acceleration vibration signals of the gearbox in the horizontal direction. Table 1 is a fault setting data table in the present embodiment.
TABLE 1
All fault signals of the self-built fault acquisition platform are acquired at 1260rpm, and the acquired time domain vibration signals are preprocessed by adopting z-score standardization. Fig. 6 is an exemplary diagram of the normalized time-domain vibration signal in the present embodiment. The time-domain vibration signals are converted into original time-frequency images, then the time-domain vibration signals are scaled into rectangles with the size of 100 x 100 in a unified mode through bilinear interpolation scaling, 600 pictures are generated in each category, and 4800 pictures are obtained in total.
A residual-capsule network is then constructed. Table 2 is a parameter configuration table of the residual error-capsule network in the present embodiment.
TABLE 2
In this embodiment, one batch layer is configured by default after each convolution layer.
The residual error-capsule network is built through an open source deep learning library Keras, the size of BatchSize is 32, the initial learning rate is 0.001, and the Adam algorithm is selected as the optimization algorithm.
In this embodiment, two fault diagnosis methods based on a Convolutional Neural Network (CNN) and a convolutional capsule network (Cap-Net) are used as comparison methods to compare with the diagnosis result of the present invention, and the selected evaluation indexes include accuracy, recall rate and F1 value. Table 3 is a table comparing the diagnosis results of the present invention and the comparative method in this example.
Model (model) | Rate of accuracy | Recall rate | F1 value |
CNN | 98.79% | 98.75% | 98.79% |
Cap-Net | 99.30% | 99.31% | 99.31% |
The |
100% | 100% | 100% |
TABLE 3
As shown in Table 3, each index of the invention is superior to the traditional convolutional neural network and the convolutional capsule network, and the residual error-capsule network provided by the invention can achieve 100% of accuracy, which is improved by 0.7% compared with the convolutional capsule network and 1.21% compared with the traditional convolutional neural network.
Further, in order to highlight the advantages of the invention in the aspect of noise immunity, a Gaussian noise mode is added to the original data set, and a trained model is used for diagnosing the noisy data. In the embodiment, 7 levels of noise data are set, and the signal-to-noise ratio ranges from-3 dB to 3 dB. Table 4 is a comparison table of the diagnostic results of the present invention and the comparison method for the noisy data of this example.
TABLE 4
As shown in Table 4, under the condition of noise addition, the residual error-capsule network provided by the invention has obvious advantages, when the signal-to-noise ratio is equal to 0dB, the noise power is equal to the signal power, the diagnosis accuracy rate of the invention can still reach more than 90%, and is far higher than that of a convolution capsule network and a common convolution neural network, and the invention is effective in improving the diagnosis performance of the planetary gearbox fault.
Although illustrative embodiments of the present invention have been described above to facilitate the understanding of the present invention by those skilled in the art, it should be understood that the present invention is not limited to the scope of the embodiments, and various changes may be made apparent to those skilled in the art as long as they are within the spirit and scope of the present invention as defined and defined by the appended claims, and all matters of the invention which utilize the inventive concepts are protected.
Claims (3)
1. A fault diagnosis method for a planetary gearbox based on a residual error-capsule network is characterized by comprising the following steps:
s1: setting K different fault states of the planetary gearbox according to needs, respectively acquiring a plurality of time domain vibration signals with preset time length T in a normal state and each fault state, then obtaining two-dimensional time frequency data for each time domain vibration signal through short-time Fourier transform to generate an original time frequency image, and then zooming the obtained original time frequency image to a preset size to obtain a time frequency image; setting the state label of each time-frequency image as a corresponding state code K, wherein K is 0 to represent a normal state, K is 1,2, …, and K represents a fault state serial number;
s2: constructing a residual error-capsule network, comprising a convolution module, a residual error module, an attention module and a capsule module, wherein:
the convolution module is used for performing convolution operation on the input time-frequency image, realizing feature extraction and data dimension reduction, and outputting the obtained feature image to the residual error module;
the residual error module is used for processing the received characteristic diagram and outputting the obtained characteristic diagram to the attention module;
the attention module is used for acquiring a channel characteristic weight value from the received characteristic diagram by adopting a channel attention mechanism, then performing characteristic redistribution on the input characteristic diagram according to the acquired channel characteristic weight value and sending the generated characteristic diagram to the capsule layer;
the capsule module is used for estimating a fault state corresponding to the input time-frequency image according to the received characteristic diagram; the capsule module includes main capsule layer, digital capsule layer and output layer, wherein:
the main capsule layer is used for extracting the characteristics of the characteristic diagram from the attention module and outputting the obtained characteristic diagram to the digital capsule layer;
the digital capsule layer comprises K +1 digital capsules, each digital capsule carries out feature extraction on the received feature map to obtain a one-dimensional vector, and the obtained K +1 one-dimensional vectors are output to the output layer;
the output layer is used for compressing the received K +1 one-dimensional vectors, namely solving the module length of the corresponding digital capsule to obtain the probability value of the input time-frequency image belonging to K +1 states;
s3: training the residual error-capsule network constructed in the step S2 by taking each time-frequency image obtained in the step S1 as input and the corresponding fault state label as expected output to obtain a trained residual error-capsule network;
s4: when fault diagnosis needs to be carried out on the planetary gearbox, a time domain vibration signal with the time length of T is collected, time-frequency image conversion and scaling are carried out by adopting the same method in the step S1, and the obtained time-frequency image is input into the trained residual error-capsule network to obtain a fault diagnosis result.
2. The planetary gearbox fault diagnosis method according to claim 1, wherein the convolution module in step S2 includes a first convolution layer, a pooling layer and a second convolution layer, wherein the first convolution layer convolves the input time-frequency image with a convolution kernel of size 7 × 7, the convolution step is 2, and the obtained feature map is output to the pooling layer; pooling the input feature maps by the pooling layer in a maximal pooling mode of 3 × 3, wherein the pooling step length is 2, and outputting the obtained feature maps to a second convolution layer; and the second convolution layer is convolved by adopting a convolution kernel with the size of 1 x 1, the convolution step is 1, and the obtained characteristic diagram is taken as the characteristic diagram of the convolution module to be output.
3. The planetary gearbox fault diagnosis method according to claim 1, wherein the residual error module in step S2 comprises a first convolutional layer, a block convolutional layer, a connection layer and a second convolutional layer, wherein:
the first convolution layer is used for convolving the received feature map by a convolution kernel with the size of 1 x 1, and outputting the obtained feature map to the blocking convolution layer and the second convolution layer respectively;
for block-wise convolutional layersCarrying out 4-path splitting on the received feature graph according to the channels, and respectively recording the obtained 4-path features as X1、X2、X3、X4Obtaining 4-path characteristic Y through block convolution layer processing1、Y2、Y3、Y4The specific method comprises the following steps:
characteristic Y1Is equal to X1;
Characteristic Y2From the feature X2Sequentially carrying out convolution processing on two convolution kernels with the sizes of 1 x 3 and 3 x 1 to obtain the convolution kernel;
characteristic Y3From the feature X3And feature Y2Adding the two convolution kernels with the sizes of 1 × 3 and 3 × 1 in sequence, and performing convolution processing to obtain the product;
characteristic Y4From the feature X4And feature Y3Adding the two convolution kernels with the sizes of 1 × 3 and 3 × 1 in sequence, and performing convolution processing to obtain the product;
will 4 way feature Y1、Y2、Y3、Y4Outputting to the connection layer;
the connection layer is used for splicing the received 4 paths of features to obtain a feature map and sending the feature map to the second convolution layer;
and the second convolution layer adds the two received feature maps and then carries out activation processing, and the obtained feature map is used as the feature map of the residual error module to be output.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210072530.1A CN114548153B (en) | 2022-01-21 | 2022-01-21 | Planetary gear box fault diagnosis method based on residual error-capsule network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210072530.1A CN114548153B (en) | 2022-01-21 | 2022-01-21 | Planetary gear box fault diagnosis method based on residual error-capsule network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114548153A true CN114548153A (en) | 2022-05-27 |
CN114548153B CN114548153B (en) | 2023-06-02 |
Family
ID=81671028
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210072530.1A Active CN114548153B (en) | 2022-01-21 | 2022-01-21 | Planetary gear box fault diagnosis method based on residual error-capsule network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114548153B (en) |
Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108710830A (en) * | 2018-04-20 | 2018-10-26 | 浙江工商大学 | A kind of intensive human body 3D posture estimation methods for connecting attention pyramid residual error network and equidistantly limiting of combination |
CN110059796A (en) * | 2018-01-19 | 2019-07-26 | 杭州海康威视数字技术股份有限公司 | The generation method and device of convolutional neural networks |
CN111179196A (en) * | 2019-12-28 | 2020-05-19 | 杭州电子科技大学 | Multi-resolution depth network image highlight removing method based on divide-and-conquer |
CN111310861A (en) * | 2020-03-27 | 2020-06-19 | 西安电子科技大学 | License plate recognition and positioning method based on deep neural network |
CN111652812A (en) * | 2020-04-30 | 2020-09-11 | 南京理工大学 | Image defogging and rain removing algorithm based on selective attention mechanism |
US20200342360A1 (en) * | 2018-06-08 | 2020-10-29 | Tencent Technology (Shenzhen) Company Limited | Image processing method and apparatus, and computer-readable medium, and electronic device |
CN112184577A (en) * | 2020-09-17 | 2021-01-05 | 西安理工大学 | Single image defogging method based on multi-scale self-attention generation countermeasure network |
CN112233026A (en) * | 2020-09-29 | 2021-01-15 | 南京理工大学 | SAR image denoising method based on multi-scale residual attention network |
CN112801270A (en) * | 2021-01-21 | 2021-05-14 | 中国人民解放军国防科技大学 | Automatic U-shaped network slot identification method integrating depth convolution and attention mechanism |
CN112884033A (en) * | 2021-02-06 | 2021-06-01 | 浙江净禾智慧科技有限公司 | Household garbage classification detection method based on convolutional neural network |
US20210248355A1 (en) * | 2019-04-02 | 2021-08-12 | Tencent Technology (Shenzhen) Company Limited | Face key point detection method and apparatus, storage medium, and electronic device |
CN113255882A (en) * | 2021-04-30 | 2021-08-13 | 南通大学 | Bearing fault diagnosis method based on improved convolution capsule network |
CN113344077A (en) * | 2021-06-08 | 2021-09-03 | 中国农业大学 | Anti-noise solanaceae disease identification method based on convolution capsule network structure |
CN113591638A (en) * | 2021-07-20 | 2021-11-02 | 天津理工大学 | Planetary gearbox fault diagnosis method based on convolution capsule network |
-
2022
- 2022-01-21 CN CN202210072530.1A patent/CN114548153B/en active Active
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110059796A (en) * | 2018-01-19 | 2019-07-26 | 杭州海康威视数字技术股份有限公司 | The generation method and device of convolutional neural networks |
CN108710830A (en) * | 2018-04-20 | 2018-10-26 | 浙江工商大学 | A kind of intensive human body 3D posture estimation methods for connecting attention pyramid residual error network and equidistantly limiting of combination |
US20200342360A1 (en) * | 2018-06-08 | 2020-10-29 | Tencent Technology (Shenzhen) Company Limited | Image processing method and apparatus, and computer-readable medium, and electronic device |
US20210248355A1 (en) * | 2019-04-02 | 2021-08-12 | Tencent Technology (Shenzhen) Company Limited | Face key point detection method and apparatus, storage medium, and electronic device |
CN111179196A (en) * | 2019-12-28 | 2020-05-19 | 杭州电子科技大学 | Multi-resolution depth network image highlight removing method based on divide-and-conquer |
CN111310861A (en) * | 2020-03-27 | 2020-06-19 | 西安电子科技大学 | License plate recognition and positioning method based on deep neural network |
CN111652812A (en) * | 2020-04-30 | 2020-09-11 | 南京理工大学 | Image defogging and rain removing algorithm based on selective attention mechanism |
CN112184577A (en) * | 2020-09-17 | 2021-01-05 | 西安理工大学 | Single image defogging method based on multi-scale self-attention generation countermeasure network |
CN112233026A (en) * | 2020-09-29 | 2021-01-15 | 南京理工大学 | SAR image denoising method based on multi-scale residual attention network |
CN112801270A (en) * | 2021-01-21 | 2021-05-14 | 中国人民解放军国防科技大学 | Automatic U-shaped network slot identification method integrating depth convolution and attention mechanism |
CN112884033A (en) * | 2021-02-06 | 2021-06-01 | 浙江净禾智慧科技有限公司 | Household garbage classification detection method based on convolutional neural network |
CN113255882A (en) * | 2021-04-30 | 2021-08-13 | 南通大学 | Bearing fault diagnosis method based on improved convolution capsule network |
CN113344077A (en) * | 2021-06-08 | 2021-09-03 | 中国农业大学 | Anti-noise solanaceae disease identification method based on convolution capsule network structure |
CN113591638A (en) * | 2021-07-20 | 2021-11-02 | 天津理工大学 | Planetary gearbox fault diagnosis method based on convolution capsule network |
Non-Patent Citations (3)
Title |
---|
HUAN LIU等: "Improved dual-scale residual network for image super-resolution", NEURAL NETWORKS * |
张慧美: "基于联合框架和分离框架的行人搜索算法研究", 中国优秀硕士学位论文全文数据库信息科技辑 * |
董建伟: "基于残差胶囊网络的滚动轴承故障诊断研究", 机电工程 * |
Also Published As
Publication number | Publication date |
---|---|
CN114548153B (en) | 2023-06-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111238814B (en) | Rolling bearing fault diagnosis method based on short-time Hilbert transform | |
CN111523509B (en) | Equipment fault diagnosis and health monitoring method integrating physical and depth expression characteristics | |
CN113567131B (en) | Bearing fault diagnosis method based on S transformation and miniature convolutional neural network model | |
CN116010900A (en) | Multi-scale feature fusion gearbox fault diagnosis method based on self-attention mechanism | |
CN114819315A (en) | Bearing degradation trend prediction method based on multi-parameter fusion health factor and time convolution neural network | |
CN114462446A (en) | Rolling bearing fault diagnosis method based on vibration signal and electronic equipment | |
CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
CN116451133A (en) | Heavy-duty robot speed reducer fault diagnosis method based on graph structure information | |
CN114048787B (en) | Method and system for intelligently diagnosing bearing fault in real time based on Attention CNN model | |
CN115204272A (en) | Industrial system fault diagnosis method and equipment based on multi-sampling rate data | |
Rahimilarki et al. | Time-series deep learning fault detection with the application of wind turbine benchmark | |
CN112634391B (en) | Gray image depth reconstruction and fault diagnosis system based on compressed sensing | |
CN111766513B (en) | Capsule network-based variable-working-condition multi-fault diagnosis method for three-phase induction motor | |
CN116821762A (en) | Mechanical fault diagnosis method based on multi-scale graph attention fusion network | |
CN111122160B (en) | Abnormal signal detection method based on stack noise reduction self-encoder | |
CN117171544A (en) | Motor vibration fault diagnosis method based on multichannel fusion convolutional neural network | |
CN114548153A (en) | Planetary gearbox fault diagnosis method based on residual error-capsule network | |
CN115165363A (en) | CNN-based light bearing fault diagnosis method and system | |
CN113505654A (en) | Bearing health state identification method based on multi-view attention network | |
CN113591625A (en) | Gearbox fault diagnosis method based on unbalanced sample importance weighting | |
CN113158769A (en) | CEEMDAN and FastICA-based electromechanical device bearing vibration signal denoising method | |
CN112161807A (en) | Fault diagnosis method and device for speed change gear box and storage medium | |
CN117992863B (en) | Rotary machine fault diagnosis method based on interpretable stationary wavelet packet convolution network | |
CN115481666B (en) | Gearbox small sample fault diagnosis method, system and equipment | |
CN116626491B (en) | Five-phase permanent magnet motor turn-to-turn short circuit fault diagnosis method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |