CN117992777A - Motor electrical fault diagnosis method based on image fusion - Google Patents
Motor electrical fault diagnosis method based on image fusion Download PDFInfo
- Publication number
- CN117992777A CN117992777A CN202410167172.1A CN202410167172A CN117992777A CN 117992777 A CN117992777 A CN 117992777A CN 202410167172 A CN202410167172 A CN 202410167172A CN 117992777 A CN117992777 A CN 117992777A
- Authority
- CN
- China
- Prior art keywords
- time
- frequency
- image
- images
- fusion
- 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
- 230000004927 fusion Effects 0.000 title claims abstract description 89
- 238000003745 diagnosis Methods 0.000 title claims abstract description 58
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 14
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 18
- 238000011176 pooling Methods 0.000 claims description 10
- 230000009466 transformation Effects 0.000 claims description 8
- 238000012935 Averaging Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 30
- 230000016507 interphase Effects 0.000 description 6
- 230000008901 benefit Effects 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000013480 data collection Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 241001391944 Commicarpus scandens Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000005272 metallurgy Methods 0.000 description 1
- -1 metallurgy Substances 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Landscapes
- Tests Of Circuit Breakers, Generators, And Electric Motors (AREA)
Abstract
The invention belongs to the technical field of industrial fault diagnosis, and in particular relates to an electric fault diagnosis method of a motor based on image fusion, which comprises the steps of respectively acquiring three-phase current time sequence data and three-phase voltage time sequence data during normal operation and fault operation of the motor; converting the three-phase current time sequence data and the three-phase voltage time sequence data into a time-frequency image by using short-time Fourier transform; and then carrying out three-phase current time-frequency image fusion and three-phase voltage time-frequency image fusion, and fusing the fused current time-frequency image and the fused voltage time-frequency image to form a time-frequency fusion characteristic image: integrating all time-frequency fusion characteristic images as a characteristic set; constructing a fault diagnosis network model of the time-frequency fusion characteristic image; and performing motor electrical fault diagnosis after training a fault diagnosis network model of the time-frequency fusion characteristic image. The method solves the problems of the prior art that the fault diagnosis model has higher calculation complexity and lower fault prediction precision, and the motor fault feature extraction is difficult.
Description
Technical Field
The invention relates to the technical field of industry, in particular to a motor electrical fault diagnosis method based on image fusion.
Background
With the development of economy and industrialization, the motor is widely applied in the industrial fields of coal, metallurgy, petroleum, chemical industry and the like, and becomes an indispensable power device. The load of the motor is often changed in the running process, and the motor continuously works in a complex working environment for a long time, so that the motor is easy to break down. Once a certain part of the motor fails, the motor can be stopped in the middle and the production process is stopped, so that the production efficiency is influenced, and meanwhile, the economic benefit and the personal safety are greatly influenced. The traditional manual fault diagnosis technology has limitations, and can not accurately discover faults in time.
Chinese patent No. CN114553639a discloses a Morse signal detection and recognition method, which includes performing short-time fourier transform on an audio time domain signal to obtain a first standard time-frequency image; performing energy sorting and feature extraction on the first standard time-frequency image, and determining the frequency point of the Morse signal; digitally filtering the audio time domain signal based on the frequency point of the Morse signal to obtain an interference-free signal, performing short-time Fourier transform to obtain a second standard time-frequency image, and obtaining a second effective time-frequency image; and carrying out image enhancement processing on the second effective time-frequency image, extracting a characteristic sequence, obtaining the characteristic sequence arranged according to the time domain, and then carrying out message prediction.
Conventional motor fault diagnosis methods are generally based on expert experience and require manual analysis of vibration data and current data of the motor to discover potential faults. However, the method has the problems of strong subjectivity, large workload, low diagnosis accuracy and the like, and limits the efficiency and accuracy of fault diagnosis. The deep learning technology is adopted, so that the diagnosis accuracy can be improved, the labor cost can be reduced, the diagnosis process is simplified, and the economic loss is avoided. However, the motor fault signal often includes a plurality of signal dimensions such as three-phase current, three-phase voltage, stator current, rotor current, motor rotation speed, etc., which may cause difficulty in extracting motor fault characteristics, and the efficiency and accuracy of fault diagnosis are low.
Disclosure of Invention
In order to solve the technical problems, the invention adopts the following technical scheme: the motor electrical fault diagnosis method based on image fusion specifically comprises the following steps:
S1, acquiring three-phase current time sequence data and three-phase voltage time sequence data of a motor in normal operation and fault operation;
s2, multi-sensor time sequence data conversion: respectively converting three-phase current time sequence data Ia, ib and Ic and three-phase voltage time sequence data Ua, ub and Uc into time-frequency images by using short-time Fourier transformation;
s3, respectively carrying out three-phase current time-frequency image fusion and three-phase voltage time-frequency image fusion, and then fusing the fused current time-frequency image and the fused voltage time-frequency image to form a time-frequency fusion characteristic image:
S4, integrating all the time-frequency fusion characteristic images to be used as a characteristic set;
The method for randomly dividing the data set into a training set and a testing set is that 80% of samples in the data set are randomly selected as the training set of the algorithm, and 20% of samples are selected as the testing set of the algorithm;
S5, constructing a fault diagnosis network model of the time-frequency fusion characteristic image;
s6, performing motor electrical fault diagnosis after training a fault diagnosis network model of the time-frequency fusion characteristic image;
The invention relates to a fault diagnosis network model training method, which adopts SGD (Stochast IC GRAD IENT DESCENT) optimization algorithm to update model parameters, the SGD algorithm updates the parameters of the fault diagnosis model by calculating the gradient of each training sample, and the parameter updating formula is as follows:
Where, θ a represents the model parameters of the a-th iteration, η is the learning rate, Is the gradient of the objective function f with respect to the parameter θ a;
the SGD optimization algorithm has the advantage that only one sample gradient is needed to be calculated in each iteration, so that the calculated amount of the fault diagnosis model can be further reduced.
Specifically, the different fault types of the motor electrical faults simulate different types of motor faults by designing a motor fault circuit, the collection of motor fault time sequence data is completed by using a data collection module, and the fault types collected by experiments comprise: the motor phase failure fault, interphase short circuit fault, phase-to-ground fault, rotor exciting voltage fault and rotor exciting current fault.
Further, converting the three-phase current time series data and the three-phase voltage time series data into time-frequency images by using short-time fourier transform specifically includes:
For the three-phase current time sequence data I a,Ib,Ic and the three-phase voltage time sequence data U a,Ub,Uc of the motor, a time domain-frequency domain conversion method is respectively used, wherein Fourier transformation is the most common method for time domain-frequency domain conversion, but the Fourier transformation is limited to frequency spectrum analysis of signals and has no resolution capability on the time domain.
The invention uses Short time fourier transform (STFT, short-Time Fourier Transform) to time-frequency convert the motor signal. The short-time fourier transform is a fourier-based transform that can spread a one-dimensional time series to a two-dimensional time-frequency plane.
The three-phase current time sequence data and the three-phase voltage time sequence data are subjected to windowing processing by using short-time Fourier transformation to reflect the time-frequency domain characteristics of the three-phase current time sequence data and the three-phase voltage time sequence data:
Where G (t, ω) is the complex value of frequency ω at time t, t is time, ω is frequency, f (u) is the original time domain signal, i.e., three-phase current time series data, three-phase voltage time series data, e -iωt is the complex exponent, describing the frequency of the signal, G (u-t) is the window function, multiple window function selections are used in calculating STFT, and Hamming windows are used in the present invention, the formula is as follows:
In equation (2), g (t) is the window function value at time t;
the STFT takes a window function g (t), expands the original time domain signals, namely three-phase current time sequence data and three-phase voltage time sequence data f (U), from a time domain to a time frequency domain, respectively converts the three-phase current time sequence data I a,Ib,Ic and the three-phase voltage time sequence data U a,Ub,Uc into a three-phase current time frequency image and a three-phase voltage time frequency image by using short-time Fourier transformation.
Further, the fusion of S3 is specifically:
S31, forming an initial picture set PI by using the time-frequency images, wherein the initial picture set PI comprises N groups of current time-frequency image sets, and each group of current time-frequency image sets comprises 3 current time-frequency images, and P iN＝{PiaN,PibN,PicN; n groups of voltage time-frequency image sets, wherein each group of voltage time-frequency image set comprises 3 voltage time-frequency images, P uN＝{PuaN,PubN,PucN;
Wherein, P iN represents a total of N groups of current time-frequency image sets, P iaN represents a total of N groups of a phase current time-frequency images, P ibN represents a total of N groups of b phase current time-frequency images, P icN represents a total of N groups of c phase current time-frequency images, P uN represents a total of N groups of voltage time-frequency image sets, P uaN represents a total of N groups of a phase voltage time-frequency images, P ubN represents a total of N groups of b phase voltage time-frequency images, and P ucN represents a total of N groups of c phase voltage time-frequency images;
Each image P iaN,PibN,PicN,PuaN,PubN,PucN is represented in RGB space by a three-dimensional vector.
S32, acquiring a fusion of an R value of P ian in an nth group of images in a current time-frequency image set P iN, a G value of P ibn in the nth group of images and a B value of P icn in the nth group of images to form a picture, wherein the new picture imgI can be expressed as follows: imgI = (P ianr1,Pibng2,Picnb3), wherein 1 Σn is equal to or less than N;
wherein, (P ianr1,Pibng2,Picnb3) represents the R value of the nth group of current images, the G value of the nth group of current images and the B value of the nth group of current images respectively;
After primary fusion, the incompleteness of expressing fault characteristics of a single current sensor can be avoided, and after the data of a plurality of similar current sensors are fused, the fault data can contain more diversified fault characteristics, so that the capability of expressing faults is improved.
S33, R values of P uan in an nth group of images in the voltage time-frequency image set P uN are acquired, G values of P ubn in the nth group of images are acquired, B values of P ucn in the nth group of images are fused into one image, and then a new image imgU can be expressed as follows: imgU = (P uanr1,Pubng2,Pucnb3);
Wherein, (P uanr1,Pubng2,Pucnb3) represents the R value of the nth group of voltage images, the G value of the nth group of voltage images and the B value of the nth group of voltage images respectively;
after primary fusion, the incompleteness of expressing fault characteristics of a single voltage sensor can be avoided, and after the data of a plurality of similar voltage sensors are fused, the fault data can contain more diversified fault characteristics, so that the capability of expressing faults is improved.
And S34, respectively taking one of the two groups of images imgI and imgU, averaging the two images, and adding and fusing the two images to form a group of time-frequency fusion characteristic images, wherein the formula is as follows:
imgn＝(imgIn+imgUn)/2 (3)；
Wherein imgI n is the nth current time-frequency image fused by S32, imgU n is the nth voltage time-frequency image fused by S33, img n is the nth time-frequency fusion characteristic image;
after further fusion, the incompleteness of the similar voltage or current sensors expressing fault characteristics can be avoided, and the fault data can be further diversified after the data of a plurality of different types of sensors are fused.
The step S5 specifically comprises the following steps:
S51, the fault diagnosis network in the invention firstly uses a global average pooling method to integrate three-phase current and three-phase voltage characteristics contained in each channel in the input time-frequency fusion characteristic image, and the global average pooling expression is as follows
Where P is the output after global averaging pooling, 1/(W H) is a normalization factor, where W and H are the width and height of the property map, respectively, and U m,j is an element of the property map, where m and j represent the index.
S52, carrying out convolution calculation on the input time-frequency fusion characteristic image by using one-dimensional convolution with the convolution kernel size of K, extracting local characteristics of the time-frequency fusion characteristic image, wherein the size of the convolution kernel K can be defined according to differences of different time-frequency fusion characteristic images, and the calculation complexity can be reduced while the performance is not reduced by using one-dimensional convolution.
The activation value of the one-dimensional convolution output is calculated using S igmo id functions, S igmo id formula is as follows:
Where x is the value of the input argument, and the value range of Sigmoid (x) is (0, 1).
S53, finally, in order to weight the three-phase current and three-phase voltage characteristics in each channel of the time-frequency fusion image, multiplying the weight parameters and the characteristics of each channel of the time-frequency fusion image one by one to obtain weighted characteristicsThe important features are given a large weight, and the ineffective features are given a small weight so as to embody the importance of different features.
Advantageous effects
Compared with the prior art, the invention has the following advantages:
(1) The invention provides an image fusion-based motor electrical fault diagnosis method, which provides a new solution for motor fault diagnosis with multi-signal dimension as fault characteristics by using an image fusion technology, when fault diagnosis is carried out by utilizing multi-dimensional data, various detection signals are required to be captured by adopting a plurality of sensors, different information is accumulated, each sensor makes unique contribution to a fault diagnosis result, and the multi-signal dimension is subjected to dimension reduction processing by using the image fusion technology, so that a fault diagnosis model can utilize less data and lower calculation complexity, realize faster fault diagnosis and obtain higher fault prediction precision.
(2) The invention provides a motor electrical fault diagnosis method based on image fusion, which constructs a fault diagnosis model aiming at a time-frequency fusion characteristic image by using a method of global average pooling, self-defining convolution kernel size and weighting characteristic parameters, improves the extraction efficiency of motor fault characteristics, effectively digs effective characteristics of a time-frequency diagram, improves the precision of a fault diagnosis process and reduces the calculation complexity of the fault diagnosis process.
Drawings
Fig. 1 is a flowchart of a fault diagnosis method according to the present invention.
Fig. 2 is a waveform diagram of fault three-phase current time series data according to the present embodiment.
Fig. 3 is a waveform diagram of the fault three-phase voltage timing data according to the present embodiment.
Fig. 4 is a time-frequency diagram of the current of the fault time series data after short-time fourier transform according to the embodiment.
Fig. 5 is a voltage time-frequency diagram of the fault time sequence data after short-time fourier transform according to the embodiment.
Fig. 6 is a time-frequency fusion feature diagram after the fault feature fusion in this embodiment.
Fig. 7 is a structural diagram of the failure diagnosis network according to the present embodiment.
Fig. 8 is a graph of the failure diagnosis network loss according to the present embodiment.
Fig. 9 is a failure diagnosis network learning graph according to the present embodiment.
Detailed Description
Example 1
The embodiment provides a motor electrical fault diagnosis method based on image fusion, as shown in fig. 1, comprising the following specific steps:
S1, acquiring three-phase current time sequence data and three-phase voltage time sequence data of a motor in normal operation and fault operation;
Further, the data in S1 are collected through a motor fault experimental platform, and specific motor parameters are shown in the following table:
| Parameters of the motor | Value of | Unit (B) |
| Rated output power of motor | 300 | W |
| Rated rotation speed of motor | 1345 | rpm |
| Rated frequency of motor | 50 | Hz |
| Rated voltage of motor | 230(△)；400(Y) | V |
| Rated current of motor | 0.75 | A |
Specifically, the different fault types of the motor electrical faults simulate different types of motor faults by designing a motor fault circuit, the collection of motor fault time sequence data is completed by using a data collection module, and the fault types collected by experiments comprise: the motor phase failure fault, interphase short circuit fault, phase-to-ground fault, rotor exciting voltage fault and rotor exciting current fault.
The collected fault data are motor three-phase current time sequence data and motor three-phase voltage time sequence data, each group of time sequence data comprises 400000 continuous sampling points, wherein 80 motor faults are randomly simulated, and the sampling data when the motor faults are needed to be screened out by a manual screening mode.
Further, the three-phase current time series data collected in S1 are shown in fig. 2:
Fig. 2 (a) is a motor open-phase fault three-phase current diagram, fig. 2 (b) is a relative ground fault three-phase current diagram, fig. 2 (c) is an interphase short-circuit fault three-phase current diagram, fig. 2 (d) is a rotor exciting voltage fault three-phase current diagram, and fig. 2 (e) is a rotor exciting current fault three-phase current diagram.
Further, the three-phase voltage time sequence data collected in S1 is shown in fig. 3:
Fig. 3 (a) is a motor open-phase fault three-phase voltage diagram, fig. 3 (b) is a relative ground fault three-phase voltage diagram, fig. 3 (c) is an interphase short-circuit fault three-phase voltage diagram, fig. 3 (d) is a rotor exciting voltage fault three-phase voltage diagram, and fig. 3 (e) is a rotor exciting current fault three-phase voltage diagram.
S2, multi-sensor time sequence data conversion: three-phase current time sequence data I a,Ib,Ic and three-phase voltage time sequence data U a,Ub,Uc are converted into time-frequency images by short-time Fourier transformation;
further, converting the three-phase current time series data and the three-phase voltage time series data into time-frequency images by using short-time fourier transform specifically includes:
The three-phase current time sequence data and the three-phase voltage time sequence data are subjected to windowing processing by using short-time Fourier transformation to reflect the time-frequency domain characteristics of the three-phase current time sequence data and the three-phase voltage time sequence data:
Where G (t, ω) is the complex value of frequency ω at time t, t is time, ω is frequency, f (u) is the original time domain signal, i.e., three-phase current time series data, three-phase voltage time series data, e -iωt is the complex exponent, describing the frequency of the signal, G (u-t) is the window function, multiple window function selections are used in calculating STFT, and Hamming windows are used in the present invention, the formula is as follows:
where g (t) is the window function value at time t;
STFT takes a window function g (t), expands the original time domain signals, namely three-phase current time sequence data and three-phase voltage time sequence data f (U), from time domain to time domain, respectively converts the three-phase current time sequence data I a、Ib、Ic and the three-phase voltage time sequence data U a、Ub、Uc into three-phase current time frequency images and three-phase voltage time frequency images by using short-time Fourier transform, as shown in RGB color diagrams of fig. 4 and 5;
Fig. 4 (a) is a motor open-phase fault current time-frequency diagram, fig. 4 (b) is a relative ground fault current time-frequency diagram, fig. 4 (c) is an interphase short-circuit fault current time-frequency diagram, fig. 4 (d) is a rotor exciting voltage fault current time-frequency diagram, and fig. 4 (e) is a rotor exciting current fault current time-frequency diagram;
Fig. 5 (a) is a motor open-phase fault voltage time-frequency diagram, fig. 5 (b) is a relative ground fault voltage time-frequency diagram, fig. 5 (c) is an inter-phase short-circuit fault voltage time-frequency diagram, fig. 5 (d) is a rotor exciting voltage fault voltage time-frequency diagram, and fig. 5 (e) is a rotor exciting current fault voltage time-frequency diagram.
S3, performing image fusion to form a time-frequency fusion characteristic image:
respectively carrying out three-phase current time-frequency image fusion and three-phase voltage time-frequency image fusion, and then fusing the fused current time-frequency image and the fused voltage time-frequency image to form a time-frequency fusion characteristic image;
Further, the fusion of S3 is specifically:
S31, forming an initial picture set PI by using the time-frequency images, wherein the initial picture set PI comprises N groups of current time-frequency image sets, and each group of current time-frequency image sets comprises 3 current time-frequency images, and P iN＝{PiaN,PibN,PicN; n groups of voltage time-frequency image sets, wherein each group of voltage time-frequency image set comprises 3 voltage time-frequency images, P uN＝{PuaN,PubN,PucN;
Wherein, P iN represents a total N groups of current time-frequency image sets, P iaN represents a total N groups of a-phase current time-frequency images, P ibN represents a total N groups of b-phase current time-frequency images, and P icN represents a total N groups of c-phase current time-frequency images;
P uN represents a total of N sets of voltage time-frequency images, P uaN represents a total of N sets of a-phase voltage time-frequency images, P ubN represents a total of N sets of b-phase voltage time-frequency images, and P ucN represents a total of N sets of c-phase voltage time-frequency images;
Each image P iaN,PibN,PicN,PuaN,PubN,PucN is represented in RGB space by a three-dimensional vector.
S32, acquiring a fusion of an R value of P ian in an nth group of images in a current time-frequency image set P iN, a G value of P ibn in the nth group of images and a B value of P icn in the nth group of images to form a picture, and then expressing a new picture, namely a current time-frequency fusion picture imgI, as follows: imgI = (P ianr1,Pibng2,Picnb3), wherein 1 Σ N is equal to or less than N, as shown in fig. 6 (a) a current time-frequency fusion diagram of five fault types;
wherein, (P ianr1,Pibng2,Picnb3) represents the R value of the nth group of current images, the G value of the nth group of current images and the B value of the nth group of current images respectively;
After primary fusion, the incompleteness of expressing fault characteristics of a single current sensor can be avoided, and after the data of a plurality of similar current sensors are fused, the fault data can contain more diversified fault characteristics, so that the capability of expressing faults is improved.
S33, R values of P uan in an nth group of images in the voltage time-frequency image set P uN are acquired, G values of P ubn in the nth group of images are acquired, B values of P ucn in the nth group of images are fused into one image, and then a new image imgU can be expressed as follows: imgU = (P uanr1,Pubng2,Pucnb3), a voltage time-frequency fusion diagram of five fault types as shown in fig. 6 (b);
Wherein, (P uanr1,Pubng2,Pucnb3) represents the R value of the nth group of voltage images, the G value of the nth group of voltage images and the B value of the nth group of voltage images respectively;
after primary fusion, the incompleteness of expressing fault characteristics of a single voltage sensor can be avoided, and after the data of a plurality of similar voltage sensors are fused, the fault data can contain more diversified fault characteristics, so that the capability of expressing faults is improved.
And S34, respectively taking one of the two groups of images imgI and imgU, averaging the two images, and adding and fusing the two images to form a group of time-frequency fusion characteristic images, wherein the formula is as follows:
imgn＝(imgIn+imgUn)/2 (3)；
Wherein imgI n is the nth current time-frequency image fused by S32, imgU n is the nth voltage time-frequency image fused by S33, img n is the nth time-frequency fusion feature image, as shown in fig. 6 (c), of five fault types;
after further fusion, the incompleteness of the similar voltage or current sensors expressing fault characteristics can be avoided, and the fault data can be further diversified after the data of a plurality of different types of sensors are fused.
S4, integrating all the time-frequency fusion characteristic images to be used as a characteristic set;
The data set is randomly divided into a training set and a testing set in such a way that 80% of samples in the data set are randomly selected as the training set of the algorithm and 20% of samples are selected as the testing set of the algorithm.
S5, constructing a fault diagnosis network model of the time-frequency fusion characteristic image, as shown in FIG. 7;
The step S5 specifically comprises the following steps:
S51, the fault diagnosis network in the invention firstly uses a global average pooling method to integrate three-phase current and three-phase voltage characteristics contained in each channel in the input time-frequency fusion characteristic image, and the global average pooling expression is as follows
Where P is the output after global averaging pooling, 1/(W H) is a normalization factor, where W and H are the width and height of the property map, respectively, and U m,j is an element of the property map, where m and j represent the index.
S52, carrying out convolution calculation on the input time-frequency fusion characteristic image by using one-dimensional convolution with the convolution kernel size of K, extracting local characteristics of the time-frequency fusion characteristic image, wherein the size of the convolution kernel K can be defined according to differences of different time-frequency fusion characteristic images, and the calculation complexity can be reduced while the performance is not reduced by using one-dimensional convolution.
The activation value of the one-dimensional convolution output is calculated using S igmo id functions, S igmo id formula is as follows:
S53, finally, in order to weight the three-phase current and three-phase voltage characteristics in each channel of the time-frequency fusion image, multiplying the weight parameters and the characteristics of each channel of the time-frequency fusion image one by one to obtain weighted characteristics The important features are given a large weight, and the ineffective features are given a small weight so as to embody the importance of different features.
S6, performing motor electrical fault diagnosis after training the fault diagnosis network model of the time-frequency fusion characteristic image.
Further, S6 is specifically:
The invention relates to a fault diagnosis network model training method, which adopts an SGD (Stochast IC GRAD IENT DESCENT) optimization algorithm to update model parameters, wherein the SGD algorithm updates the parameters of a fault diagnosis model by calculating the gradient of each training sample. The parameter update formula is as follows:
Where, θ a represents the model parameters of the a-th iteration, η is the learning rate, Is the gradient of the objective function f with respect to the parameter θ a.
The SGD optimization algorithm has the advantage that only one sample gradient is needed to be calculated in each iteration, so that the calculated amount of the fault diagnosis model can be further reduced.
Further, the values of the parameter selection convolution kernels k of the training network model are 3, 5 and 7 in sequence, the batch size is 16, the learning rate is 0.1, and the training number is 100.
The loss function curve processed by the fault diagnosis model is shown in fig. 8, the electric motor fault diagnosis network learning curve is shown in fig. 9, the fault diagnosis accuracy of fault data processed by the fault diagnosis model is 98.87%, and the electric motor fault diagnosis method based on graph fusion has good diagnosis effect.
Claims (6)
1. The motor electrical fault diagnosis method based on image fusion is characterized by comprising the following steps of;
S1, acquiring three-phase current time sequence data and three-phase voltage time sequence data of a motor in normal operation and fault operation;
s2, converting the three-phase current time sequence data and the three-phase voltage time sequence data into time-frequency images;
s3, respectively carrying out three-phase current time-frequency image fusion and three-phase voltage time-frequency image fusion, and then fusing the fused current time-frequency image and the fused voltage time-frequency image to form a time-frequency fusion characteristic image;
S4, integrating all the time-frequency fusion characteristic images to be used as a characteristic set;
s5, constructing a fault diagnosis network model aiming at the time-frequency fusion characteristic image;
S6, performing motor electrical fault diagnosis after training the fault diagnosis network model of the time-frequency fusion characteristic image.
2. The method for diagnosing an electrical fault of a motor based on image fusion according to claim 1, wherein the converting three-phase current time series data and three-phase voltage time series data into time-frequency images is specifically as follows:
the time-frequency domain characteristics of the three-phase current time sequence data and the three-phase voltage time sequence data are reflected by windowing the three-phase current time sequence data and the three-phase voltage time sequence data by using short-time Fourier transformation;
in formula (1), G (t, ω) is a complex value of frequency ω at time t, f (u) is three-phase current time series data, three-phase voltage time series data which are original time-domain signals, e -iωt is a complex exponent, i is an imaginary unit, G (u-t) is a window function, and Hamming window is used in calculating STFT:
In equation (2), g (t) is the window function value at time t;
STFT takes a window function g (t), and expands the original time domain signals, namely three-phase current time sequence data and three-phase voltage time sequence data f (u), from a time domain to a time domain;
Three-phase current time series data I a,Ib,Ic and three-phase voltage time series data U a,Ub,Uc are converted into three-phase current time-frequency images and three-phase voltage time-frequency images respectively by using short-time fourier transform.
3. The method for diagnosing an electrical fault of a motor based on image fusion according to claim 2, wherein the steps of performing three-phase current time-frequency image fusion, three-phase voltage time-frequency image fusion, fusion of the fused current time-frequency image and the fused voltage time-frequency image, respectively, and forming a time-frequency fusion feature image include:
S31, forming an initial picture set PI by using the time-frequency images, wherein the initial picture set PI comprises N groups of current time-frequency image sets, and each group of current time-frequency image sets comprises 3 current time-frequency images, and P iN＝{PiaN,PibN,PicN; n groups of voltage time-frequency image sets, wherein each group of voltage time-frequency image set comprises 3 voltage time-frequency images, P uN＝{PuaN,PubN,PucN;
Wherein, P iN represents a total of N groups of current time-frequency image sets, P iaN represents a total of N groups of a phase current time-frequency images, P ibN represents a total of N groups of b phase current time-frequency images, P icN represents a total of N groups of c phase current time-frequency images, P uN represents a total of N groups of voltage time-frequency image sets, P uaN represents a total of N groups of a phase voltage time-frequency images, P ubN represents a total of N groups of b phase voltage time-frequency images, and P ucN represents a total of N groups of c phase voltage time-frequency images;
each image P iaN,PibN,PicN,PuaN,PubN,PucN is represented in RGB space by a three-dimensional vector;
S32, acquiring a fusion of an R value of P ian in an nth group of images in a current time-frequency image set P iN, a G value of P ibn in the nth group of images and a B value of P icn in the nth group of images to form a picture, wherein the new picture imgI can be expressed as follows: imgI = (P ianr1,Pibng2,Picnb3), wherein 1 Σn is equal to or less than N;
wherein, (P ianr1,Pibng2,Picnb3) represents the R value of the nth group of current images, the G value of the nth group of current images and the B value of the nth group of current images respectively;
S33, R values of P uan in an nth group of images in the voltage time-frequency image set P uN are acquired, G values of P ubn in the nth group of images are acquired, B values of P ucn in the nth group of images are fused into one image, and then a new image imgU can be expressed as follows: imgU = (P uanr1,Pubng2,Pucnb3);
Wherein, (P uanr1,Pubng2,Pucnb3) represents the R value of the nth group of voltage images, the G value of the nth group of voltage images and the B value of the nth group of voltage images respectively;
And S34, respectively taking one of the two groups of images imgI and imgU, averaging the two images, and adding and fusing the two images to form a group of time-frequency fusion characteristic images, wherein the formula is as follows:
imgn＝(imgIn+imgUn)/2 (3)；
In the formula (3), imgI n is an nth current time-frequency image fused by S32, imgU n is an nth voltage time-frequency image fused by S33, and img n is an nth time-frequency fusion characteristic image.
4. A method for diagnosing an electrical fault of a motor based on image fusion as recited in claim 3, wherein said constructing a fault diagnosis network model of a time-frequency fusion feature image comprises:
s51, firstly integrating three-phase currents and three-phase voltage characteristics contained in each channel in the input time-frequency fusion characteristic image by using a global average pooling method, wherein the global average pooling expression is as follows:
In equation (4), P is the output result after global averaging pooling, 1/(w×h) is a normalization factor, where W and H are the width and height of the feature map, respectively, and U m,j is an element of the feature map, where m and j represent indexes;
S52, carrying out convolution calculation on the input time-frequency fusion characteristic image by using one-dimensional convolution with the convolution kernel size of K, and extracting local characteristics of the time-frequency fusion image;
calculating an activation value of the one-dimensional convolution output by using a Sigmoid function, wherein the Sigmoid formula is as follows:
S53, finally, multiplying the weight parameter calculated according to the local feature of the time-frequency fusion image and the feature of each channel of the time-frequency fusion image one by one to obtain a weighted feature Large weights are given to important features and small weights are given to invalid features.
5. A method for diagnosing electric faults of a motor based on image fusion as claimed in claim 3, wherein the fault diagnosis network model of the time-frequency fusion characteristic image is updated by adopting SGD optimization algorithm:
In the formula (6), θ a represents the model parameters of the a-th iteration, η is the learning rate, Is the gradient of the objective function f with respect to the parameter θ a.
6. A method of diagnosing an electrical fault in a motor based on image fusion as recited in claim 3, wherein the fault types include: motor open-phase faults, phase-to-phase short-circuit faults, phase-to-ground faults, rotor field voltage faults, and rotor field current faults.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410167172.1A CN117992777A (en) | 2024-02-06 | 2024-02-06 | Motor electrical fault diagnosis method based on image fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410167172.1A CN117992777A (en) | 2024-02-06 | 2024-02-06 | Motor electrical fault diagnosis method based on image fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117992777A true CN117992777A (en) | 2024-05-07 |
Family
ID=90897017
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410167172.1A Pending CN117992777A (en) | 2024-02-06 | 2024-02-06 | Motor electrical fault diagnosis method based on image fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117992777A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118194241A (en) * | 2024-05-20 | 2024-06-14 | 广汽埃安新能源汽车股份有限公司 | Electric automobile motor fault diagnosis method and device |
-
2024
- 2024-02-06 CN CN202410167172.1A patent/CN117992777A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118194241A (en) * | 2024-05-20 | 2024-06-14 | 广汽埃安新能源汽车股份有限公司 | Electric automobile motor fault diagnosis method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Han et al. | Intelligent fault diagnosis method for rotating machinery via dictionary learning and sparse representation-based classification | |
| CN117992777A (en) | Motor electrical fault diagnosis method based on image fusion | |
| CN106556781A (en) | Shelf depreciation defect image diagnostic method and system based on deep learning | |
| CN110147760B (en) | Novel efficient electric energy quality disturbance image feature extraction and identification method | |
| CN102279358A (en) | MCSKPCA based neural network fault diagnosis method for analog circuits | |
| CN109655266B (en) | Wind turbine generator bearing fault diagnosis method based on AVMD and spectrum correlation analysis | |
| Jiang et al. | A fault diagnostic method for induction motors based on feature incremental broad learning and singular value decomposition | |
| CN114216682B (en) | Service life prediction method and device of rolling bearing based on TCN and BLS | |
| CN112881942A (en) | Abnormal current diagnosis method and system based on wavelet decomposition and empirical mode decomposition | |
| CN112926728B (en) | Small sample turn-to-turn short circuit fault diagnosis method for permanent magnet synchronous motor | |
| CN113255458A (en) | Bearing fault diagnosis method based on multi-view associated feature learning | |
| CN115481657A (en) | Wind generating set communication slip ring fault diagnosis method based on electric signals | |
| CN113884844A (en) | Transformer partial discharge type identification method and system | |
| CN115130514A (en) | Construction method and system for health index of engineering equipment | |
| CN112160877A (en) | Fan bearing fault diagnosis method based on SFA and CNN | |
| CN113610119B (en) | Method for identifying power transmission line development faults based on convolutional neural network | |
| CN111428772A (en) | Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting | |
| CN111025100A (en) | Transformer ultrahigh frequency partial discharge signal mode identification method and device | |
| CN117171544B (en) | Motor vibration fault diagnosis method based on multichannel fusion convolutional neural network | |
| Yang et al. | Fault Classification in Distribution Systems Using Deep Learning With Data Preprocessing Methods Based on Fast Dynamic Time Warping and Short-Time Fourier Transform | |
| CN114118149A (en) | Induction motor fault diagnosis system based on finite element simulation and symmetric feature migration | |
| CN112162196A (en) | Motor fault diagnosis method based on graph attention network | |
| CN116773952A (en) | Transformer voiceprint signal fault diagnosis method and system | |
| CN116610990A (en) | Method and device for identifying hidden danger of breaker based on characteristic space differentiation | |
| CN116226704A (en) | Ship variable rotation speed bearing fault diagnosis method based on multi-feature fusion and improved SheffeNetV 2 |
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 |