CN110705456A - Micro motor abnormity detection method based on transfer learning - Google Patents
Micro motor abnormity detection method based on transfer learning Download PDFInfo
- Publication number
- CN110705456A CN110705456A CN201910932081.1A CN201910932081A CN110705456A CN 110705456 A CN110705456 A CN 110705456A CN 201910932081 A CN201910932081 A CN 201910932081A CN 110705456 A CN110705456 A CN 110705456A
- Authority
- CN
- China
- Prior art keywords
- motor
- time
- dimensional
- power spectrum
- transfer learning
- 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
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
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20048—Transform domain processing
- G06T2207/20056—Discrete and fast Fourier transform, [DFT, FFT]
Abstract
The invention discloses a micro motor abnormity detection method based on transfer learning, which comprises the following steps: performing feature extraction on a time domain vibration signal of the micro motor through short-time Fourier transform to obtain a two-dimensional time-frequency power spectrum matrix; performing characteristic analysis on the two-dimensional time-frequency power spectrum matrix to obtain a two-dimensional power spectrum of RGB three channels; and (3) carrying out anomaly detection on the motor by adopting a transfer learning method for improving a one-dimensional frequency domain convolution kernel network structure. Aiming at the condition that the number of motor samples is small, the network overfitting and the unreasonable parameter model are easily caused when the convolutional neural network is trained, the overfitting is effectively relieved through transfer learning, and meanwhile, the quality identification efficiency and the accuracy of the miniature direct current gear speed reduction motor are improved.
Description
Technical Field
The invention relates to the technical fields of Fourier transform, feature extraction, feature analysis, image processing, convolutional neural network, transfer learning and the like, in particular to a micro motor anomaly detection method based on transfer learning.
Background
The concept of ' fault diagnosis ' was formally proposed in the 70's of the 20 th century, the initial source was bionics, and when the system was at an abnormal level, the type, degree, position, etc. of the fault were determined by analyzing various state information generated during the operation, which is the core of the diagnosis technology. The running state of the motor can be obtained through vibration information, electrical information, thermodynamic information and auditory information collected by a state acquisition system, and due to the huge information amount, a method for efficiently detecting the abnormality of the motor is urgently needed.
(1) Fault diagnosis based on model analysis:
with the maturity of system identification and modeling and state space theory, a fault diagnosis method based on model analysis appears in the 80 s of the 20 th century, firstly, input and output of a motor are analyzed and modeled to obtain a mathematical model of the motor, ideal output is obtained under the condition of new input, whether the motor has faults or not and fault reasons are analyzed by comparing with actual output, but parameters of the motor can change in the actual operation process, and the nonlinear, multivariable and strong-coupling motor model is difficult to model.
(2) Fault diagnosis based on signal analysis:
in order to solve the problem that most complex motor systems cannot accurately obtain input and output models of the complex motor systems or model parameters are still difficult to estimate after the complex motor systems are simplified, a fault diagnosis method based on signal analysis is provided, the characteristic relation between motor faults and motor output signals is directly obtained through a corresponding signal processing technology, the input and output models of the motor do not need to be analyzed, and motor fault information, fault types, fault degrees and the like are directly obtained through mode recognition. At present, motor fault detection and diagnosis can be divided into the following four directions from the viewpoint of signal processing: <1> time domain signal analysis, such as determining a fault type by analyzing a root mean square, a skewness, a kurtosis, and a crest factor of the time domain signal; <2> frequency domain signal Analysis, which is to perform discrete Fourier transform on a time domain signal to obtain a spectrogram for Analysis, such as Motor Current Signature Analysis (MCSA); <3> frequency domain enhanced signal analysis, such as enhanced bispectrum technique using auxiliary frequencies; <4> time-frequency signal analysis, when the complete information can not be obtained by only observing the time domain or frequency domain waveform, the time domain and frequency domain signals are combined and analyzed together.
(3) Artificial intelligence fault diagnosis based on data:
information is extracted from a large amount of data, and an effective anomaly detection method including methods of an expert system, machine learning, an artificial neural network and the like is found by analyzing a large amount of measurement signals in a time domain or a frequency domain. The expert system is used for establishing an expert knowledge base for fault diagnosis by an expert and simulating expert reasoning; bayes classification, a support vector machine, a K-means clustering algorithm and the like are combined with a signal processing technology to form a hybrid system to diagnose motor faults, stator current spectrum analysis is adopted, and a fuzzy minimum maximum neural network and a classification and regression tree are combined to judge fault types of the induction motor. Frequency domain feature clustering information of motor vibration signals is mined by using an SOM method, and then a two-dimensional multi-classification SVM is used for realizing a fault type judgment error which is 1.48% extremely low. An end-to-end model of motor faults and diagnosis problems is realized by using a one-dimensional convolutional neural network, and the model replaces a feature extraction step and a feature classification model which use traditional machine learning and other methods in the FDD field. A self-encoding network is used for motor anomaly detection, and the network is trained by taking motor vibration frequency domain information as input. The deep learning shows the special capability in the aspect of feature extraction, can effectively and accurately represent the abnormal detection task of the complex motor, and carries out abnormal detection on the motor by combining short-time Fourier transform, a convolutional neural network and a transfer learning theory.
Disclosure of Invention
In view of the above technical problems, the present invention aims to provide a method for detecting an abnormality of a micro motor based on transfer learning, which effectively alleviates overfitting through transfer learning, and simultaneously improves the efficiency and accuracy of quality identification of the micro dc gear reduction motor.
The purpose of the invention is realized by at least one of the following technical schemes:
a micro motor abnormity detection method based on transfer learning comprises the following steps:
performing feature extraction on a time domain vibration signal of the micro motor through short-time Fourier transform to obtain a two-dimensional time-frequency power spectrum matrix;
performing characteristic analysis on the two-dimensional time-frequency power spectrum matrix to obtain a two-dimensional power spectrum of RGB three channels;
and (3) carrying out anomaly detection on the motor by adopting a transfer learning method for improving a one-dimensional frequency domain convolution kernel network structure.
Further, the step of characterizing the time-domain vibration signal of the micro-motor by short-time fourier transform specifically comprises: extracting and adopting short-time Fourier transform to analyze the local sine frequency and phase information of the time-varying signal, obtaining a five-second motor time-domain signal, and then carrying out windowing, framing and fast Fourier transform on the signal to obtain a two-dimensional time-frequency power spectrum matrix.
Further, when the characteristic analysis is performed on the two-dimensional time-frequency power spectrum matrix, in order to more intuitively understand the power spectrum matrix based on the short-time fourier transform, the matrix is converted into an RGB three-channel two-dimensional power spectrum according to the principle that the energy spectrum density value changes from low to high corresponding to the color from cold color to warm color.
Further, the training of the classification model of the transfer learning method specifically includes the steps of:
1) establishing a signal database of the miniature direct current gear reduction motor, namely collecting hundreds of motors with different types, processing to obtain five-second time-frequency data signals of the motors, and dividing a training set, a testing set and a verification set according to a certain proportion;
2) dividing the five-second time-frequency data signal into five parts of one-second signals, windowing, framing, performing fast Fourier transform and image standardization processing to obtain a two-dimensional power spectrum matrix;
3) converting the matrix into a two-dimensional power spectrogram of RGB three channels according to the principle that the energy spectrum density value changes from low to high corresponding to the color from cold color to warm color;
4) the method comprises the steps of pre-training AlexNet network convolution layer parameters on an ImageNet data set through transfer learning, adding an improved one-dimensional frequency domain convolution kernel, extracting the characteristics of an abnormal motor in the frequency domain direction, modifying the number of layers of SoftMax, initializing model hyper-parameters including learning rate, batch size, CNN kernel size and number, network frame position and number, and conducting L2 regularization on weight values;
5) inputting data of a training set into an initialized convolutional network deep model for training, and adjusting parameters such as weight of the convolutional network deep model through a BP algorithm when twenty periods or LOSS values of the network relatively change and are smaller than a threshold value;
6) inputting the data of the verification set into the trained model to verify the accuracy, if the accuracy is greater than the threshold value, keeping the model network parameters, otherwise, returning to the step 5) again, and adjusting the network framework and the hyper-parameters of the model.
Compared with the prior art, the invention at least comprises the following beneficial effects:
aiming at the problem that network overfitting and a parameter model are unreasonable when a convolutional neural network is trained easily under the condition of few motor samples, the invention integrates the theory of transfer learning into the convolutional neural network to effectively relieve overfitting, simultaneously improves the quality identification efficiency and accuracy of the miniature direct current gear speed reducing motor, and realizes the quality detection task of the miniature direct current gear speed reducing motor.
Drawings
Fig. 1 is a schematic flow chart of a method for detecting an abnormality of a micro motor according to an embodiment of the present invention.
Fig. 2 is an overall framework of the micro motor fault detection platform.
FIG. 3 is a color chromatogram.
Fig. 4(a) is a time domain diagram of a vibration signal of a normal motor.
Fig. 4(b) is a two-dimensional RGB power spectrum after the vibration signal time domain diagram processing of the normal motor.
Fig. 5(a) is a time domain diagram of the vibration signal of the abnormal motor.
Fig. 5(b) is a two-dimensional RGB power spectrum after the time domain diagram processing of the vibration signal of the abnormal motor.
Fig. 6 is an AlexNet network structure.
Fig. 7 is a framework of an anomaly detection classification flow based on transfer learning.
Detailed Description
The invention is further described below with reference to the figures and examples.
As shown in fig. 1, a method for detecting an abnormality of a micro motor based on transfer learning includes:
s1, extracting the characteristics of the time domain vibration signal of the micro motor through short-time Fourier transform to obtain a two-dimensional time-frequency power spectrum matrix;
s2, performing feature analysis on the two-dimensional time-frequency power spectrum matrix to obtain an RGB three-channel two-dimensional power spectrum;
and S3, carrying out anomaly detection on the motor by adopting a transfer learning method of an improved one-dimensional frequency domain convolution kernel network structure.
The whole set of micro motor fault detection platform for implementing the method consists of four parts, namely a direct current stabilized voltage power supply, a micro gear motor vibration signal measurement platform and a PC, and the schematic diagram is shown in FIG. 2.
The attached drawings 1 are sequentially as follows from left to right: (1) the direct current stabilized voltage supply is used for providing stable voltage for the tested micro gear speed reducing motor to enable the tested micro gear speed reducing motor to operate in a rated state; (2) the micro gear reduction motor vibration signal measuring platform is in rigid contact with a micro motor in normal operation through a vibration sensor to obtain an ideal vibration signal waveform for analysis; (3) the vibration signal conversion platform is used for processing discrete digital signals only by the PC, so that analog-to-digital conversion needs to be carried out on analog vibration signals subjected to signal amplification processing, and then the analog vibration signals are input into the PC for processing; (4) and the PC is used for building a motor fault detection system and training an abnormality detection classification model.
The step of carrying out the characteristics on the time domain vibration signal of the micro motor through short-time Fourier transform specifically comprises the following steps: extracting and adopting short-time Fourier transform to analyze the local sine frequency and phase information of the time-varying signal, obtaining a five-second motor time-domain signal, and then carrying out windowing, framing and fast Fourier transform on the signal to obtain a two-dimensional time-frequency power spectrum matrix.
When the characteristic analysis is carried out on the two-dimensional time-frequency power spectrum matrix, in order to more intuitively understand the power spectrum matrix based on the short-time Fourier transform, the matrix is converted into a RGB three-channel two-dimensional power spectrum according to the principle that the energy spectrum density value changes from low to high corresponding to the color from cold color to warm color.
The training of the classification model of the transfer learning method specifically comprises the following steps:
1) establishing a signal database of the miniature direct current gear reduction motor, namely collecting hundreds of motors with different types, processing to obtain five-second time-frequency data signals of the motors, and dividing a training set, a testing set and a verification set according to a certain proportion;
2) dividing the five-second time-frequency data signal into five parts of one-second signals, windowing, framing, performing fast Fourier transform and image standardization processing to obtain a two-dimensional power spectrum matrix;
3) converting the matrix into a two-dimensional power spectrogram of RGB three channels according to the principle that the energy spectrum density value changes from low to high corresponding to the color from cold color to warm color;
4) the method comprises the steps of pre-training AlexNet network convolution layer parameters on an ImageNet data set through transfer learning, adding an improved one-dimensional frequency domain convolution kernel, extracting the characteristics of an abnormal motor in the frequency domain direction, modifying the number of layers of SoftMax, initializing model hyper-parameters including learning rate, batch size, CNN kernel size and number, network frame position and number, and conducting L2 regularization on weight values;
5) inputting data of a training set into an initialized convolutional network deep model for training, and adjusting parameters such as weight of the convolutional network deep model through a BP algorithm when twenty periods or LOSS values of the network relatively change and are smaller than a threshold value;
6) inputting the data of the verification set into the trained model to verify the accuracy, if the accuracy is greater than the threshold value, keeping the model network parameters, otherwise, returning to the step 5) again, and adjusting the network framework and the hyper-parameters of the model.
Since the vibration signal of the micro gear reduction direct current motor belongs to a non-stationary time-varying signal, namely the frequency changes along with the change of time, in order to better describe the property, a Short Time Fourier Transform (STFT) is selected for analysis, and the STFT is a transform which is similar to the STFT and is used for analyzing the sinusoidal frequency and phase information of a local area of the time-varying signal. The working principle of short-time fourier transform is to divide a time domain signal into a plurality of sections, each section is approximated to a stable signal, further fourier transform can be performed respectively, time domain and frequency domain information is analyzed simultaneously, a window function is used for signal truncation, so to speak, STFT is a transform based on a window function, the shorter the length of the window function is, the more accurate the time domain information is, and the longer the length of the window function is, the more accurate the frequency domain information is, so that the angle of research needs to be measured according to actual research. Meanwhile, in order to improve the time domain characteristics to the maximum extent on the premise of ensuring the frequency domain information, the front and rear window functions may be partially overlapped, but the calculation amount is also increased.
Comprehensively considering sampling precision and model complexity, selecting a five-second time domain discrete vibration signal with the sampling frequency of 12Khz, dividing the five-second time domain discrete vibration signal into five signals of one second, and then adopting a discrete time short-time Fourier transform formula:
where x [ n ] is the discrete signal, γ [ n ] is the window function, and two window functions that are commonly used are the Hamming window and the rectangular window. The width of the main lobe of the Hamming window is 2 pi/N, the side lobe attenuation is large, the amplitude is small, and the frequency leakage is less. The rectangular window has the advantages that the width of a main lobe is small, the frequency resolution is high, the amplitude of a side lobe is large, the frequency leakage phenomenon exists, the selected window function is a Hamming window, the size is selected to be 256 sampling points, and the overlapping part is 128 sampling points.
The discrete signal is then subjected to a Discrete Fourier Transform (DFT), but the computational complexity and N of the DFT2Proportional, fast fourier transform with frequency decimation is used, since the sequence length N-256-28Dividing the sequence number k of the frequency domain X (k) according to the parity number to obtain:
wherein the content of the first and second substances,
and k is 1 when it is an even number and-1 when it is an odd number.
Let k be 2r, k be 2r +1, and r be 0,1, …, N/2-1, yielding:
thus, X (k) is divided into parity groups by the formula k, and the N-point DFT is divided into two N/2-point DFTs. In the same way, the parity grouping of N/2-point discrete Fourier transform can be carried out, and after 7 times of decomposition, the parity grouping is finally decomposed into 128 two-point discrete Fourier transforms, so that the calculation amount is greatly reduced compared with that of directly carrying out DFT algorithm.
Finally, a 92 × 129 two-dimensional time-frequency power spectrum matrix can be obtained by performing a short-time fourier transform-based feature extraction method on the vibration signal of the one-second miniature gear reduction motor, wherein 92 represents a time dimension, and 129 represents a frequency dimension.
In order to more intuitively understand the power spectrum matrix based on the short-time fourier transform, the matrix is converted into a two-dimensional power spectrum of RGB three channels according to the principle that the energy spectrum density value changes from low to high corresponding to the color from cold to warm as shown in fig. 3.
Fig. 4(a) and 4(b) are a time domain diagram of a vibration signal of a normal miniature gear reduction motor and an RGB three-channel two-dimensional power spectrum diagram after STFT processing. Fig. 5(a) and 5(b) are respectively a vibration signal time domain diagram of an abnormal micro gear reduction motor and an RGB three-channel two-dimensional power spectrum diagram after STFT processing.
It can be seen that only a few fixed fundamental frequency bands appear in the two-dimensional RGB power spectrogram of the normal motor, and the abnormal motor with large noise can have the phenomenon that the amplitude of the high frequency band is increased at a certain moment in the two-dimensional RGB power spectrogram, so that the two-dimensional RGB power spectrogram can be used as the input feature of the convolutional neural network to train a classification model, and an improved 7 x 1 one-dimensional frequency domain convolution kernel is adopted to extract the feature in the frequency domain direction.
The model-based transfer learning theory is applied to an abnormality detection classification model of the miniature gear reduction motor, and a transferred target network is a champion network AlexNet in ImageNet competition of 2012, as shown in FIG. 5.
Fig. 7 is a motor abnormality detection classification flow based on transfer learning:
1. frequency domain signal of vibration signal: windowing and framing the signals by using a window function, performing fast Fourier transform on each frame of signal to generate a five-second signal time-frequency diagram corresponding to three channels, and dividing the five-second signal time-frequency diagram into five parts and one second time-frequency diagrams;
2. normalization of the images: the method has the advantages that the centralization processing of each image is realized through the mean value, and according to the convex optimization theory and the data probability distribution related knowledge, the centralization of the data accords with the data distribution rule, so that the generalization effect after training is more easily obtained;
3. building an AlexNet network basic structure, modifying a SoftMax layer, adding an improved 7 multiplied by 1 one-dimensional frequency domain convolution kernel, loading a network parameter pre-trained by the convolution layer, fixing the network parameter, finely adjusting a neuron parameter of a full connection layer, setting a super parameter of the convolution neural network, including a learning rate, a batch size and a CNN kernel size and number, and training;
4. until the Loss value of the network is smaller than a certain threshold value or the training period reaches a certain number, saving network parameters, and inputting the data of the verification set into the trained model to verify the accuracy rate; if the occurrence accuracy rate is larger than the threshold value, the model network parameters are kept, otherwise, the training is returned again, and the network framework and the hyper-parameters of the model are adjusted.
The implementation of the invention aims at the problems of quality detection efficiency, precision and the like of the miniature direct current gear speed reducing motor, and incorporates the theory of transfer learning. The AlexNet network with the migration improved one-dimensional frequency domain convolution kernel is used for carrying out feature extraction and model learning on the vibration signal of the gear reduction motor, and the quality detection task of the miniature direct-current gear reduction motor is realized. The efficiency of quality identification of the miniature direct current gear speed reduction motor is improved while the accuracy rate is ensured.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (4)
1. A micro motor abnormity detection method based on transfer learning is characterized by comprising the following steps:
the time domain vibration signal of the micro motor is subjected to feature extraction through short-time Fourier transform,
obtaining a two-dimensional time-frequency power spectrum matrix;
performing characteristic analysis on the two-dimensional time-frequency power spectrum matrix to obtain a two-dimensional power spectrum of RGB three channels;
and (3) carrying out anomaly detection on the motor by adopting a transfer learning method for improving a one-dimensional frequency domain convolution kernel network structure.
2. The method for detecting the abnormality of the micro-motor based on the transfer learning according to claim 1, wherein the step of characterizing the time-domain vibration signal of the micro-motor by short-time fourier transform is specifically as follows: extracting and adopting short-time Fourier transform to analyze the local sine frequency and phase information of the time-varying signal, obtaining a five-second motor time-domain signal, and then carrying out windowing, framing and fast Fourier transform on the signal to obtain a two-dimensional time-frequency power spectrum matrix.
3. The method as claimed in claim 1, wherein when the two-dimensional time-frequency power spectrum matrix is subjected to feature analysis, in order to more intuitively understand the power spectrum matrix based on short-time fourier transform, the matrix is converted into a RGB three-channel two-dimensional power spectrum according to the principle that the energy spectrum density value changes from low to high corresponding to a color from cold to warm.
4. The micro-motor abnormality detection method based on the transfer learning of claim 1, wherein the training of the classification model of the transfer learning method specifically comprises the steps of:
1) establishing a signal database of the miniature direct current gear reduction motor, namely collecting hundreds of motors with different types, processing to obtain five-second time-frequency data signals of the motors, and dividing a training set, a testing set and a verification set according to a certain proportion;
2) dividing the five-second time-frequency data signal into five parts of one-second signals, windowing, framing, performing fast Fourier transform and image standardization processing to obtain a two-dimensional power spectrum matrix;
3) converting the matrix into a two-dimensional power spectrogram of RGB three channels according to the principle that the energy spectrum density value changes from low to high corresponding to the color from cold color to warm color;
4) the method comprises the steps of pre-training AlexNet network convolution layer parameters on an ImageNet data set through transfer learning, adding an improved one-dimensional frequency domain convolution kernel, extracting the characteristics of an abnormal motor in the frequency domain direction, modifying the number of layers of SoftMax, initializing model hyper-parameters including learning rate, batch size, CNN kernel size and number, network frame position and number, and conducting L2 regularization on weight values;
5) inputting data of a training set into an initialized convolutional network deep model for training, and adjusting parameters such as weight of the convolutional network deep model through a BP algorithm when twenty periods or LOSS values of the network relatively change and are smaller than a threshold value;
6) inputting the data of the verification set into the trained model to verify the accuracy, if the accuracy is greater than the threshold value, keeping the model network parameters, otherwise, returning to the step 5) again, and adjusting the network framework and the hyper-parameters of the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910932081.1A CN110705456A (en) | 2019-09-29 | 2019-09-29 | Micro motor abnormity detection method based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910932081.1A CN110705456A (en) | 2019-09-29 | 2019-09-29 | Micro motor abnormity detection method based on transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110705456A true CN110705456A (en) | 2020-01-17 |
Family
ID=69197310
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910932081.1A Pending CN110705456A (en) | 2019-09-29 | 2019-09-29 | Micro motor abnormity detection method based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110705456A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111666984A (en) * | 2020-05-20 | 2020-09-15 | 海南电网有限责任公司电力科学研究院 | Intelligent overvoltage identification method based on transfer learning |
| CN112016684A (en) * | 2020-08-06 | 2020-12-01 | 广西大学 | Electric power terminal fingerprint identification method of deep parallel flexible transmission network |
| CN112348052A (en) * | 2020-09-30 | 2021-02-09 | 国网山西省电力公司大同供电公司 | Power transmission and transformation equipment abnormal sound source positioning method based on improved EfficientNet |
| CN112464848A (en) * | 2020-12-07 | 2021-03-09 | 国网四川省电力公司电力科学研究院 | Information flow abnormal data monitoring method and device based on density space clustering |
| CN112697268A (en) * | 2020-12-04 | 2021-04-23 | 佳木斯电机股份有限公司 | Motor anomaly detection integrated algorithm based on T-SNE |
| CN112727704A (en) * | 2020-12-15 | 2021-04-30 | 北京天泽智云科技有限公司 | Method and system for monitoring corrosion of leading edge of blade |
| CN113761979A (en) * | 2020-06-04 | 2021-12-07 | 富士通株式会社 | Method, apparatus, and storage medium for optimizing a model |
| CN115017121A (en) * | 2022-08-05 | 2022-09-06 | 山东天意机械股份有限公司 | Concrete production equipment data storage system |
| CN115964670A (en) * | 2022-12-17 | 2023-04-14 | 中国人民解放军32802部队 | Frequency spectrum anomaly detection method |
| CN116702053A (en) * | 2023-08-09 | 2023-09-05 | 长春理工大学 | Agricultural machinery fault detection method based on multiple information and MCSA-SVM |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102005A (en) * | 2018-07-23 | 2018-12-28 | 杭州电子科技大学 | Small sample deep learning method based on shallow Model knowledge migration |
| CN109724984A (en) * | 2018-12-07 | 2019-05-07 | 上海交通大学 | A kind of defects detection identification device and method based on deep learning algorithm |
| CN109782167A (en) * | 2018-12-29 | 2019-05-21 | 华南理工大学 | Toothed gearing electric motor product examine device and method based on convolutional network depth model |
| CN109918752A (en) * | 2019-02-26 | 2019-06-21 | 华南理工大学 | Mechanical failure diagnostic method, equipment and medium based on migration convolutional neural networks |
| CN109975702A (en) * | 2019-03-22 | 2019-07-05 | 华南理工大学 | A kind of DC gear decelerating motor product examine method based on recirculating network disaggregated model |
| CN110175434A (en) * | 2019-06-19 | 2019-08-27 | 西南交通大学 | A kind of rail fastener system injury detection method based on convolutional neural networks |
-
2019
- 2019-09-29 CN CN201910932081.1A patent/CN110705456A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102005A (en) * | 2018-07-23 | 2018-12-28 | 杭州电子科技大学 | Small sample deep learning method based on shallow Model knowledge migration |
| CN109724984A (en) * | 2018-12-07 | 2019-05-07 | 上海交通大学 | A kind of defects detection identification device and method based on deep learning algorithm |
| CN109782167A (en) * | 2018-12-29 | 2019-05-21 | 华南理工大学 | Toothed gearing electric motor product examine device and method based on convolutional network depth model |
| CN109918752A (en) * | 2019-02-26 | 2019-06-21 | 华南理工大学 | Mechanical failure diagnostic method, equipment and medium based on migration convolutional neural networks |
| CN109975702A (en) * | 2019-03-22 | 2019-07-05 | 华南理工大学 | A kind of DC gear decelerating motor product examine method based on recirculating network disaggregated model |
| CN110175434A (en) * | 2019-06-19 | 2019-08-27 | 西南交通大学 | A kind of rail fastener system injury detection method based on convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| 于凤芹: "《数字信号处理简明教程》", 30 April 2017, 西安电子科技大学出版社 * |
| 吴定海等: "《柴油机振动信号分析与故障诊断研究》", 31 December 2012, 国防工业出版社 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111666984B (en) * | 2020-05-20 | 2023-08-25 | 海南电网有限责任公司电力科学研究院 | Overvoltage intelligent identification method based on transfer learning |
| CN111666984A (en) * | 2020-05-20 | 2020-09-15 | 海南电网有限责任公司电力科学研究院 | Intelligent overvoltage identification method based on transfer learning |
| CN113761979B (en) * | 2020-06-04 | 2023-11-14 | 富士通株式会社 | Method, apparatus and storage medium for optimizing model |
| CN113761979A (en) * | 2020-06-04 | 2021-12-07 | 富士通株式会社 | Method, apparatus, and storage medium for optimizing a model |
| CN112016684A (en) * | 2020-08-06 | 2020-12-01 | 广西大学 | Electric power terminal fingerprint identification method of deep parallel flexible transmission network |
| CN112016684B (en) * | 2020-08-06 | 2022-09-02 | 广西大学 | Electric power terminal fingerprint identification method of deep parallel flexible transmission network |
| CN112348052A (en) * | 2020-09-30 | 2021-02-09 | 国网山西省电力公司大同供电公司 | Power transmission and transformation equipment abnormal sound source positioning method based on improved EfficientNet |
| CN112697268A (en) * | 2020-12-04 | 2021-04-23 | 佳木斯电机股份有限公司 | Motor anomaly detection integrated algorithm based on T-SNE |
| CN112697268B (en) * | 2020-12-04 | 2022-06-24 | 佳木斯电机股份有限公司 | Motor anomaly detection integrated algorithm based on T-SNE |
| CN112464848A (en) * | 2020-12-07 | 2021-03-09 | 国网四川省电力公司电力科学研究院 | Information flow abnormal data monitoring method and device based on density space clustering |
| CN112727704A (en) * | 2020-12-15 | 2021-04-30 | 北京天泽智云科技有限公司 | Method and system for monitoring corrosion of leading edge of blade |
| CN115017121A (en) * | 2022-08-05 | 2022-09-06 | 山东天意机械股份有限公司 | Concrete production equipment data storage system |
| CN115964670A (en) * | 2022-12-17 | 2023-04-14 | 中国人民解放军32802部队 | Frequency spectrum anomaly detection method |
| CN116702053A (en) * | 2023-08-09 | 2023-09-05 | 长春理工大学 | Agricultural machinery fault detection method based on multiple information and MCSA-SVM |
| CN116702053B (en) * | 2023-08-09 | 2023-09-29 | 长春理工大学 | Agricultural machinery fault detection method based on multiple information and MCSA-SVM |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110705456A (en) | Micro motor abnormity detection method based on transfer learning | |
| Grezmak et al. | Interpretable convolutional neural network through layer-wise relevance propagation for machine fault diagnosis | |
| CN109977920B (en) | Water turbine set fault diagnosis method based on time-frequency spectrogram and convolutional neural network | |
| CN110792563B (en) | Wind turbine generator blade fault audio monitoring method based on convolution generation countermeasure network | |
| CN105275833A (en) | CEEMD (Complementary Empirical Mode Decomposition)-STFT (Short-Time Fourier Transform) time-frequency information entropy and multi-SVM (Support Vector Machine) based fault diagnosis method for centrifugal pump | |
| CN112435142B (en) | Power load identification method and load power utilization facility knowledge base construction method thereof | |
| CN112294341B (en) | Sleep electroencephalogram spindle wave identification method and system based on light convolutional neural network | |
| CN112766342A (en) | Abnormity detection method for electrical equipment | |
| CN108435819B (en) | Energy consumption abnormity detection method for aluminum profile extruder | |
| CN116226646B (en) | Method, system, equipment and medium for predicting health state and residual life of bearing | |
| Lv et al. | High-order synchroextracting transform for characterizing signals with strong AM-FM features and its application in mechanical fault diagnosis | |
| Liu et al. | Multi-feature fusion for fault diagnosis of rotating machinery based on convolutional neural network | |
| CN106502850B (en) | A kind of the multiple faults discrimination method and device of sparse tight frame dictionary learning model | |
| Yu et al. | A fault diagnosis approach for diesel engine valve train based on improved ITD and SDAG-RVM | |
| Huang et al. | Deep residual networks-based intelligent fault diagnosis method of planetary gearboxes in cloud environments | |
| CN111898644A (en) | Intelligent identification method for health state of aerospace liquid engine under fault-free sample | |
| CN113627375A (en) | Planetary gear fault diagnosis method and system, storage medium and computing device | |
| CN113158964A (en) | Sleep staging method based on residual learning and multi-granularity feature fusion | |
| CN115481657A (en) | Wind generating set communication slip ring fault diagnosis method based on electric signals | |
| Liu et al. | Composite multi-scale basic scale Entropy based on CEEMDAN and its application in hydraulic pump fault diagnosis | |
| Liu et al. | A novel dual-attention optimization model for points classification of power quality disturbances | |
| CN112798888B (en) | Non-invasive fault diagnosis method for vehicle-mounted electrical system of unmanned train | |
| Yang et al. | Fault Classification in Distribution Systems using Deep Learning with Data Preprocessing Methods based on Fast-DTW and Short-Time Fourier Transform | |
| Li et al. | Rotating machinery faults detection method based on deep echo state network | |
| Liu et al. | Bearing fault diagnosis based on multi-scale neural networks |
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: 20200117 |
|
| RJ01 | Rejection of invention patent application after publication |