CN114415018A - Self-learning grating interference spectrum analysis technology for motor fault early warning - Google Patents
Self-learning grating interference spectrum analysis technology for motor fault early warning Download PDFInfo
- Publication number
- CN114415018A CN114415018A CN202210059104.4A CN202210059104A CN114415018A CN 114415018 A CN114415018 A CN 114415018A CN 202210059104 A CN202210059104 A CN 202210059104A CN 114415018 A CN114415018 A CN 114415018A
- Authority
- CN
- China
- Prior art keywords
- motor
- wavelet
- frequency
- vibration
- spectrum
- 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
- 238000005516 engineering process Methods 0.000 title claims abstract description 19
- 238000010183 spectrum analysis Methods 0.000 title claims abstract description 14
- 239000013307 optical fiber Substances 0.000 claims abstract description 24
- 238000000354 decomposition reaction Methods 0.000 claims description 36
- 238000001228 spectrum Methods 0.000 claims description 30
- 238000000034 method Methods 0.000 claims description 29
- 238000001845 vibrational spectrum Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 10
- 230000009466 transformation Effects 0.000 claims description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 230000001902 propagating effect Effects 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000003745 diagnosis Methods 0.000 claims description 3
- 230000009191 jumping Effects 0.000 claims description 3
- 230000000737 periodic effect Effects 0.000 claims description 3
- 238000013139 quantization Methods 0.000 claims description 3
- 230000035939 shock Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 2
- 238000013507 mapping Methods 0.000 claims 3
- 230000006870 function Effects 0.000 abstract description 28
- 238000001514 detection method Methods 0.000 abstract description 8
- 230000002159 abnormal effect Effects 0.000 abstract description 3
- 230000007613 environmental effect Effects 0.000 abstract description 3
- 238000010801 machine learning Methods 0.000 abstract description 3
- 239000000306 component Substances 0.000 description 5
- 230000001133 acceleration Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000004020 conductor Substances 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 230000001627 detrimental effect Effects 0.000 description 1
- 230000005674 electromagnetic induction Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000192 social effect Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/34—Testing dynamo-electric machines
Abstract
The invention discloses a self-learning grating interference spectrum analysis technology for motor fault early warning in the technical field of motor fault detection, wherein a wavelet basis function is determined according to a motor and environmental parameters, a judgment model is trained firstly by utilizing data of the motor in normal operation, and fault characteristics are pushed based on the model, so that a signal frequency band with abnormal energy can be more accurately determined as a frequency band where the fault characteristics are located, the frequency of the fault characteristics is further identified by utilizing wavelet packet node reconstruction, and the fault type is judged by utilizing machine learning; only two signals of the vibration sensor and the temperature sensor are fused, and the two signals are both optical fiber sensors, so that the influence of electromagnetic interference can be completely avoided.
Description
Technical Field
The invention relates to the technical field of motor fault detection, in particular to a self-learning grating interference spectrum analysis technology for motor fault early warning.
Background
Electric motors are in widespread use in production, life and scientific research. In the modern industrial process, the motor is used as an important driving device and has the characteristics of severe working environment, more component parts and the like. When the motor runs under complicated and severe conditions, the performance of the motor is reduced, various faults of the motor occur, and core components of the motor are damaged, so that the normal operation of the whole production, life and other systems is influenced, and serious catastrophic accidents, adverse social effects and huge economic losses are caused. The method and the device can detect the parameters representing the fault state of the motor in time, improve the fault detection technology of the motor, and have great significance for avoiding the generation of great economic loss and catastrophic accidents caused by the fault of the motor.
The vibration monitoring method is an important method for detecting motor faults, and many faults of the motor can be reflected in the vibration process, but the existing monitoring has two main problems:
(1) in a complex electromagnetic environment, the sensor can generate an interference signal due to electromagnetic induction to influence judgment: the multi-sensor technology can effectively extract motor fault characteristic signals through signal fusion, and the problem is that all sensors contained in the multi-sensor have links which are easily subjected to electromagnetic interference, such as power supply and conductive circuits. In the detection process of the vibration acceleration sensor, the frequency spectrum leakage problem exists, the fault characteristic signal is relatively small, and the characteristic frequency spectrum is easily submerged by a fundamental wave component and is difficult to identify.
(2) The vibration characteristic of the motor is influenced by various factors, is not only related to the process of the motor, but also has a larger relation with the running environment, when a vibration signal is detected, the part of the content has great interference on judging the motor fault, the running state of the motor is judged by using a fixed model, and the motor vibration detection method has larger limitation and is not strong in applicability.
Aiming at the related technologies, the invention provides a self-learning grating interference spectrum analysis technology for motor fault early warning.
Disclosure of Invention
The invention provides a self-learning grating interference spectrum analysis technology for motor fault early warning, which detects motor vibration by using the grating interference technology, realizes no power supply access from a sensor to a demodulation device, and only uses optical fibers to transmit signals, thereby avoiding electromagnetic interference. By utilizing wavelet packet analysis and the self-learning function of artificial intelligence, the frequency spectrum characteristic value during fault is deduced according to basic vibration and a vibration signal in normal operation, the fault type is judged, and the influence of vibration redundant signals and interference signals is reduced. To solve the above-mentioned problems.
The invention provides a self-learning grating interference spectrum analysis technology for motor fault early warning, which adopts the following technical scheme: the method comprises the following steps:
the method comprises the following steps that optical fiber signals sent by a laser source through a coupler reach a grating arranged on a motor shell and an optical fiber temperature sensor arranged in a motor, the optical fiber signals received by the coupler are connected with an upper computer through a grating demodulator, and the upper computer analyzes optical fiber spectrum signals converted by the grating demodulator to determine motor faults;
the grating demodulator comprises a vibration demodulator and a temperature demodulator, a small wave base is determined by an optical fiber spectrum signal converted by the vibration demodulator according to motor parameters and working environment parameters, the analysis and decomposition of the spectrum signal are carried out on optical fiber interference by utilizing small wave packet transformation, meanwhile, a temperature compensation signal converted by the temperature demodulator is input, the spectrum signal of the normal operation of the motor is learned according to a convolutional neural network model, then training is carried out, an initial vibration spectrum model is modified, and vibration spectrum characteristic values of the motor under different fault conditions are deduced according to the modified vibration spectrum model;
when the motor normally runs, the vibration frequency spectrum model is used for diagnosing the characteristic value in the running process, whether the motor fails or not and the fault type of the motor are determined, and when the vibration signal of the motor in normal running changes due to changes of various reasons, the previous vibration frequency spectrum model can be dynamically corrected through model training.
Optionally, the wavelet basis is a Complex Morlet wavelet, and the wavelet function is described as:
wherein f isbIs a band width parameter, fcThe center frequency of the wave.
Optionally, the analyzing and decomposing of the spectral signal for the optical fiber interference by using wavelet packet transform includes:
the spectrum information is decomposed into two parts of low-frequency information a1 and high-frequency information d1, in the decomposition, the information lost in the low-frequency a1 is captured by the high-frequency d1, in the decomposition of the next layer, the a1 is decomposed into two parts of low-frequency a2 and high-frequency d2, the information lost in the low-frequency a2 is captured by the high-frequency d2, and the like, the deeper decomposition is carried out,
the signal is represented as:
wherein g is the coefficient of the low-pass filter, l is the number of different frequency bands, u is the wavelet packet family about the coefficient of the high-pass filter, and t represents the independent variable time;
the decomposition algorithm for the wavelet can be given by:
wherein, ak-2lAnd bk-2lAll represent the coefficient of the wavelet decomposition conjugate filter, k is the decomposition layer number, and j and n are wavelet packet nodes;
the wavelet packet reconstruction specifically solves the formula as follows:
wherein h isk-2lAnd gk-2lAll represent the coefficient of the wavelet reconstruction conjugate filter, k is the decomposition layer number, and j and n are wavelet packet nodes.
Optionally, the method further includes:
the reconstruction factor of the motor vibration wavelet base is selected as follows:
wherein s (t) is periodic impact signal component, n (t) is background Gaussian noise, the initial impact amplitude A0 is set to 0.5, and the frequency conversion f is set tor50Hz, a signal attenuation index C of 1000, and a system natural frequency fnIs 4000Hz, tauiThe random small fluctuation of the impact period T of the ith impact conforms to normal distribution, the mean value is 0, and the standard deviation is fr0.5% of; the SNR is-11 dB, and the adopted frequency is 16000 Hz;
the similarity function of the wavelet basis function to the shock signal is:
wherein s isiIs the area enclosed by each peak of the function after the absolute value of the basis function is taken, hiIs the maximum of each peak of the basis function after taking the absolute value, aiIs the weighting coefficient of each peak of the basis function after the absolute value is taken, and n is the number of peaks included after the absolute value of the basis function is taken;
selecting a wavelet basis function according to the reconstruction factor and the similarity function;
selecting a soft threshold value by a threshold value function, wherein the expression is as follows:
in the formula, wi,jIs the decomposition coefficient before the de-noising,
the decomposition coefficient after the threshold is acted, λ is a selected threshold, μ + β is 1, and m-n is 1, the original wavelet decomposition coefficient and the wavelet coefficient after the threshold quantization processing are controlled by adjusting the values of λ and β, the initial value μ is 0.3, and m is 5;
the threshold rule takes heursure, and 4 layers are selected as the decomposition layer number.
Optionally, the feature spectrum signal after wavelet processing is convolved with three trainable filters and an applicable bias, three feature maps are generated at a C1 level after convolution, then four pixels of each group in the feature maps are summed, weighted, biased, and a Sigmoid function is used to obtain three feature maps at S2 levels; the maps are further filtered to obtain a C3 layer, the hierarchy is then generated as in S2 to generate S4, and finally, the pixel values are rasterized and connected into a vector input to the convolutional neural network model to obtain an output.
Optionally, the diagnosing the characteristic value in the operation process by using the vibration spectrum model specifically includes the following steps:
1) collecting a motor vibration signal;
2) converting the vibration signal into a spectrogram through wavelet packet transformation;
3) processing the graph, namely deleting the non-characteristic part of the graph, and then compressing the graph into a square with a proper size;
4) establishing a network and initializing network parameters, constructing a network model with proper depth according to samples and requirements, and determining the network parameters;
5) network training and forward propagation, wherein a sample is input into a network, and the error between the network output and an expected target is obtained through the forward propagation;
6) judging whether the network is converged, if the network is converged, executing the step 8), otherwise executing the step 7);
7) reversely propagating and modifying the weight, reversely propagating the error obtained in the step 5) to each node layer by utilizing a BP (back propagation) algorithm, updating the weight, and repeatedly executing the steps 5) to 7) until the network is converged;
8) judging whether the network meets the actual requirements according to the accuracy of the test sample, if so, executing the step 9), otherwise, jumping to the step 4), and modifying the network parameters;
9) the output network is used for motor fault diagnosis.
In summary, the invention includes at least one of the following advantages:
(1) the wavelet basis function is determined according to the motor and environmental parameters, the judgment model is trained by using data of the motor in normal operation, the fault characteristics are obtained on the basis of the model, the signal frequency band with abnormal energy can be more accurately determined as the frequency band of the fault characteristics, the frequency of the fault characteristics is further identified by using wavelet packet node reconstruction, and the fault type is judged by using machine learning.
(2) Only two signals of the vibration sensor and the temperature sensor are fused, and the two signals are both optical fiber sensors, so that the influence of electromagnetic interference can be completely avoided.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic diagram of a structure for detecting motor vibration by using a grating interference technique according to the present invention;
FIG. 2 is a flow chart of the present invention utilizing wavelet packet analysis and artificial intelligence for self-learning;
FIG. 3 is a diagram of a spectrum signal converted by a vibration demodulator for an optical fiber signal according to the present invention;
FIG. 4 is a flow chart of wavelet packet denoising in accordance with the present invention;
FIG. 5 is a flow chart of a wavelet processed signature spectrum convolved with three trainable filters and possibly biased according to the present invention.
Detailed Description
The present invention is described in further detail below with reference to figures 1-5.
Referring to fig. 1-2, the invention discloses a self-learning grating interference spectrum analysis technology for motor fault early warning, which comprises the following steps:
the connection mode can ensure that a detection system has no power supply, no conductor and no influence of on-site electromagnetic interference in a complicated electromagnetic environment field. The optical fiber signal received by the coupler is connected with an upper computer through a grating demodulator, and the upper computer analyzes the optical fiber spectrum signal converted by the grating demodulator to determine the motor fault;
the grating demodulator comprises a vibration demodulator and a temperature demodulator, a wavelet base is determined according to motor parameters (including but not limited to motor type, rated capacity, rated rotating speed and the like) and working environment parameters (including but not limited to temperature, humidity, altitude, air pressure and the like) through the optical fiber spectrum signals converted by the vibration demodulator, the optical fiber interference is analyzed and decomposed by utilizing wavelet packet transformation, meanwhile, temperature compensation signals converted by the temperature demodulator are input, the spectrum signals of the motor in normal operation are learned according to a convolutional neural network model, then training is carried out, an initial vibration spectrum model is modified, and vibration spectrum characteristic values under different fault conditions of the motor are deduced according to the modified vibration spectrum model;
when the motor normally runs, the vibration spectrum model is used for diagnosing the characteristic value in the running process, determining whether the motor has faults and the fault type of the motor, and when the vibration signal changes in the normal running process due to the change of various reasons (including but not limited to working age, working environment and working load), the previous vibration spectrum model can be dynamically corrected through model training.
Example (b):
the motor parameters are as follows:
the working environment is as follows:
room temperature 20 ℃, altitude 153m and humidity 63%.
The fiber signal passes through a vibration demodulator, and the spectrum signal diagram is shown in fig. 3.
Depending on the motor and the working environment, a support length of 6 is chosen because too long a support length creates boundary problems, and too short a support length gives too low a vanishing moment, which is detrimental to the concentration of signal energy. In this embodiment, a Complex Morlet wavelet is selected, and the wavelet function is described as:
wherein f isbIs a band width parameter, fcThe center frequency of the wave.
The spectrum information is decomposed into two parts of low-frequency information a1 and high-frequency information d1, in the decomposition, the information lost in the low-frequency a1 is captured by the high-frequency d1, in the decomposition of the next layer, the a1 is decomposed into two parts of low-frequency a2 and high-frequency d2, the information lost in the low-frequency a2 is captured by the high-frequency d2, and the like, the deeper decomposition is carried out,
the signal is represented as:
wherein g is the coefficient of the low-pass filter, l is the number of different frequency bands, u is the wavelet packet family about the coefficient of the high-pass filter, and t represents the independent variable time;
the decomposition algorithm for the wavelet can be given by:
wherein, ak-2lAnd bk-2lAll represent the coefficient of the wavelet decomposition conjugate filter, k is the decomposition layer number, and j and n are wavelet packet nodes;
the wavelet packet reconstruction specifically solves the formula as follows:
wherein h isk-2lAnd gk-2lAll represent the coefficient of the wavelet reconstruction conjugate filter, k is the decomposition layer number, and j and n are wavelet packet nodes.
Referring to fig. 4, a wavelet packet noise reduction method is also included:
the reconstruction factor of the motor vibration wavelet base is selected as follows:
wherein s (t) is periodic impact signal component, n (t) is background Gaussian noise, the initial impact amplitude A0 is set to 0.5, and the frequency conversion f is set tor50Hz, a signal attenuation index C of 1000, and a system natural frequency fnIs 4000Hz, tauiThe random small fluctuation of the impact period T of the ith impact conforms to normal distribution, the mean value is 0, and the standard deviation is fr0.5% of; the SNR is-11 dB, and the adopted frequency is 16000 Hz;
the similarity function of the wavelet basis function to the shock signal is:
wherein s isiIs the area enclosed by each peak of the function after the absolute value of the basis function is taken, hiIs the maximum of each peak of the basis function after taking the absolute value, aiIs the weighting coefficient of each peak of the basis function after the absolute value is taken, and n is the number of peaks included after the absolute value of the basis function is taken;
selecting a Db10 wavelet as a wavelet basis function according to the reconstruction factor and the similarity function;
selecting a soft threshold value by a threshold value function, wherein the expression is as follows:
in the formula, wi,jIs the decomposition coefficient before the de-noising,
the decomposition coefficient after the threshold is acted, λ is a selected threshold, μ + β is 1, and m-n is 1, the original wavelet decomposition coefficient and the wavelet coefficient after the threshold quantization processing are controlled by adjusting the values of λ and β, the initial value μ is 0.3, and m is 5;
the threshold rule takes heursure, and 4 layers are selected as the decomposition layer number.
Referring to fig. 5, after wavelet processing, the feature spectrum signal is convolved with three trainable filters and an applicable bias, three feature maps are generated at a C1 level after convolution, then four pixels of each group in the feature maps are summed, weighted, biased, and a Sigmoid function is used to obtain three feature maps at S2 levels; the maps are further filtered to obtain a C3 layer, the hierarchy is then generated as in S2 to generate S4, and finally, the pixel values are rasterized and connected into a vector input to the convolutional neural network model to obtain an output.
The convolutional network performs training with a pilot, so its sample set is formed by: vector pairs of (input vector, ideal output vector). All the vector pairs are derived from the operation vibration results of the motor at different working temperatures, and part of data is collected from an actual operation system. Before training begins, all weights are initialized with the data of the fixed model.
After training is finished, data of the optical fiber after wavelet packet processing can be directly input into a network, and complexity of data reconstruction in the feature extraction and classification processes is avoided.
The method for diagnosing the characteristic value in the operation process by using the vibration spectrum model specifically comprises the following steps:
1) collecting a motor vibration signal;
2) converting the vibration signal into a spectrogram through wavelet packet transformation;
3) processing the graph, namely deleting the non-characteristic part of the graph, and then compressing the graph into a square with a proper size;
4) establishing a network and initializing network parameters, constructing a network model with proper depth according to samples and requirements, and determining the network parameters;
5) network training and forward propagation, wherein a sample is input into a network, and the error between the network output and an expected target is obtained through the forward propagation;
6) judging whether the network is converged, if the network is converged, executing the step 8), otherwise executing the step 7);
7) reversely propagating and modifying the weight, reversely propagating the error obtained in the step 5) to each node layer by utilizing a BP (back propagation) algorithm, updating the weight, and repeatedly executing the steps 5) to 7) until the network is converged;
8) judging whether the network meets the actual requirements according to the accuracy of the test sample, if so, executing the step 9), otherwise, jumping to the step 4), and modifying the network parameters;
9) the output network is used for motor fault diagnosis.
1. Compared with a fault detection method which only depends on a vibration acceleration sensor, the method has the advantages that (1) the vibration acceleration sensor has the problem of frequency spectrum leakage in the detection process, the characteristic signal of the fault is relatively small, and the characteristic frequency spectrum is easily submerged by fundamental wave components and is difficult to identify. (2) The vibration characteristic of the motor is influenced by various factors, is not only related to the process of the motor, but also has a larger relation with the running environment, and the running state of the motor is judged by using a fixed model, so that the motor has larger limitation and is not strong in applicability. (3) According to the method, the wavelet basis function is determined according to the motor and environmental parameters, the judgment model is trained by using data of the motor in normal operation, the fault characteristics are obtained based on the model, the signal frequency band with abnormal energy can be more accurately determined as the frequency band of the fault characteristics, the fault characteristic frequency is further identified by using wavelet packet node reconstruction, and the fault type is judged by using machine learning.
2. The multi-sensor technology can effectively extract motor fault characteristic signals through signal fusion, and the problem is that all sensors contained in the multi-sensor have links which are easily subjected to electromagnetic interference, such as power supply and conductive circuits. The invention has the advantages that only two signals of the vibration sensor and the temperature sensor are fused, and the two signals are both optical fiber sensors, so that the influence of electromagnetic interference can be completely avoided.
The above are all preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby, so: all equivalent changes made according to the structure, shape and principle of the invention are covered by the protection scope of the invention.
Claims (6)
1. A self-learning grating interference spectrum analysis technology for motor fault early warning is characterized in that: the method comprises the following steps:
the method comprises the following steps that optical fiber signals sent by a laser source through a coupler reach a grating arranged on a motor shell and an optical fiber temperature sensor arranged in a motor, the optical fiber signals received by the coupler are connected with an upper computer through a grating demodulator, and the upper computer analyzes optical fiber spectrum signals converted by the grating demodulator to determine motor faults;
the grating demodulator comprises a vibration demodulator and a temperature demodulator, a small wave base is determined by an optical fiber spectrum signal converted by the vibration demodulator according to motor parameters and working environment parameters, the analysis and decomposition of the spectrum signal are carried out on optical fiber interference by utilizing small wave packet transformation, meanwhile, a temperature compensation signal converted by the temperature demodulator is input, the spectrum signal of the normal operation of the motor is learned according to a convolutional neural network model, then training is carried out, an initial vibration spectrum model is modified, and vibration spectrum characteristic values of the motor under different fault conditions are deduced according to the modified vibration spectrum model;
when the motor normally runs, the vibration frequency spectrum model is used for diagnosing the characteristic value in the running process, whether the motor fails or not and the fault type of the motor are determined, and when the vibration signal of the motor in normal running changes due to changes of various reasons, the previous vibration frequency spectrum model can be dynamically corrected through model training.
2. The self-learning grating interference spectrum analysis technology for motor fault early warning as claimed in claim 1, wherein: the wavelet base is a Complex Morlet wavelet, and the wavelet function is described as follows:
wherein f isbIs a band width parameter, fcThe center frequency of the wave.
3. The self-learning grating interference spectrum analysis technology for motor fault early warning as claimed in claim 1, wherein: the analysis and decomposition of the spectrum signal of the optical fiber interference by utilizing the wavelet packet transformation comprises the following steps:
the spectrum information is decomposed into two parts of low-frequency information a1 and high-frequency information d1, in the decomposition, the information lost in the low-frequency a1 is captured by the high-frequency d1, in the decomposition of the next layer, the a1 is decomposed into two parts of low-frequency a2 and high-frequency d2, the information lost in the low-frequency a2 is captured by the high-frequency d2, and the like, the deeper decomposition is carried out,
the signal is represented as:
wherein g is the coefficient of the low-pass filter, l is the number of different frequency bands, u is the wavelet packet family about the coefficient of the high-pass filter, and t represents the independent variable time;
the decomposition algorithm for the wavelet can be given by:
wherein, ak-2lAnd bk-2lAll represent the coefficient of the wavelet decomposition conjugate filter, k is the decomposition layer number, and j and n are wavelet packet nodes;
the wavelet packet reconstruction specifically solves the formula as follows:
wherein h isk-2lAnd gk-2lAll represent the coefficient of the wavelet reconstruction conjugate filter, k is the decomposition layer number, and j and n are wavelet packet nodes.
4. The self-learning grating interference spectrum analysis technology for motor fault early warning as claimed in claim 1, wherein: the method also comprises a wavelet packet noise reduction method:
the reconstruction factor of the motor vibration wavelet base is selected as follows:
wherein s (t) is periodic impact signal component, n (t) is background Gaussian noise, the initial impact amplitude A0 is set to 0.5, and the frequency conversion f is set tor50Hz, a signal attenuation index C of 1000, and a system natural frequency fnIs 4000Hz, tauiThe random small fluctuation of the impact period T of the ith impact conforms to normal distribution, the mean value is 0, and the standard deviation is fr0.5% of; the SNR is-11 dB, and the adopted frequency is 16000 Hz;
the similarity function of the wavelet basis function to the shock signal is:
wherein s isiIs the area enclosed by each peak of the function after the absolute value of the basis function is taken, hiIs a basis function after taking the absolute valueNumber of maxima of each peak, aiIs the weighting coefficient of each peak of the basis function after the absolute value is taken, and n is the number of peaks included after the absolute value of the basis function is taken;
selecting a wavelet basis function according to the reconstruction factor and the similarity function;
selecting a soft threshold value by a threshold value function, wherein the expression is as follows:
in the formula, wi,jIs the decomposition coefficient before the de-noising,
the decomposition coefficient after the threshold is acted, λ is a selected threshold, μ + β is 1, and m-n is 1, the original wavelet decomposition coefficient and the wavelet coefficient after the threshold quantization processing are controlled by adjusting the values of λ and β, the initial value μ is 0.3, and m is 5;
the threshold rule takes heursure, and 4 layers are selected as the decomposition layer number.
5. The self-learning grating interference spectrum analysis technology for motor fault early warning as claimed in claim 1, wherein: convolving the characteristic spectrum signals subjected to wavelet processing with three trainable filters and an applicable bias, generating three characteristic mapping maps at a C1 layer after convolution, then summing four pixels in each group in the characteristic mapping maps, weighting values, adding the bias, and obtaining three characteristic mapping maps at S2 layers through a Sigmoid function; the maps are further filtered to obtain a C3 layer, the hierarchy is then generated as in S2 to generate S4, and finally, the pixel values are rasterized and connected into a vector input to the convolutional neural network model to obtain an output.
6. The self-learning grating interference spectrum analysis technology for motor fault early warning as claimed in claim 1, wherein: the method for diagnosing the characteristic value in the operation process by using the vibration spectrum model specifically comprises the following steps:
1) collecting a motor vibration signal;
2) converting the vibration signal into a spectrogram through wavelet packet transformation;
3) processing the graph, namely deleting the non-characteristic part of the graph, and then compressing the graph into a square with a proper size;
4) establishing a network and initializing network parameters, constructing a network model with proper depth according to samples and requirements, and determining the network parameters;
5) network training and forward propagation, wherein a sample is input into a network, and the error between the network output and an expected target is obtained through the forward propagation;
6) judging whether the network is converged, if the network is converged, executing the step 8), otherwise executing the step 7);
7) reversely propagating and modifying the weight, reversely propagating the error obtained in the step 5) to each node layer by utilizing a BP (back propagation) algorithm, updating the weight, and repeatedly executing the steps 5) to 7) until the network is converged;
8) judging whether the network meets the actual requirements according to the accuracy of the test sample, if so, executing the step 9), otherwise, jumping to the step 4), and modifying the network parameters;
9) the output network is used for motor fault diagnosis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210059104.4A CN114415018A (en) | 2022-01-19 | 2022-01-19 | Self-learning grating interference spectrum analysis technology for motor fault early warning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210059104.4A CN114415018A (en) | 2022-01-19 | 2022-01-19 | Self-learning grating interference spectrum analysis technology for motor fault early warning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114415018A true CN114415018A (en) | 2022-04-29 |
Family
ID=81273190
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210059104.4A Pending CN114415018A (en) | 2022-01-19 | 2022-01-19 | Self-learning grating interference spectrum analysis technology for motor fault early warning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114415018A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114646907A (en) * | 2021-12-24 | 2022-06-21 | 中铁二院工程集团有限责任公司 | Rail transit low-frequency magnetic field measuring method based on machine learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN203250006U (en) * | 2013-05-24 | 2013-10-23 | 温州大学 | A large-power motor fault comprehensive monitoring system based on fiber grating sensing technology |
| CN106503642A (en) * | 2016-10-18 | 2017-03-15 | 长园长通新材料股份有限公司 | A kind of model of vibration method for building up for being applied to optical fiber sensing system |
| CN108426713A (en) * | 2018-02-26 | 2018-08-21 | 成都昊铭科技有限公司 | Rolling bearing Weak fault diagnostic method based on wavelet transformation and deep learning |
-
2022
- 2022-01-19 CN CN202210059104.4A patent/CN114415018A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN203250006U (en) * | 2013-05-24 | 2013-10-23 | 温州大学 | A large-power motor fault comprehensive monitoring system based on fiber grating sensing technology |
| CN106503642A (en) * | 2016-10-18 | 2017-03-15 | 长园长通新材料股份有限公司 | A kind of model of vibration method for building up for being applied to optical fiber sensing system |
| CN108426713A (en) * | 2018-02-26 | 2018-08-21 | 成都昊铭科技有限公司 | Rolling bearing Weak fault diagnostic method based on wavelet transformation and deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 王晓龙: "基于振动信号处理的滚动轴承故障诊断方法研究", 中国博士学位论文全文数据库 工程科技Ⅱ辑, no. 12, pages 31 * |
| 陈长征;孙长城;费朝阳;周勃;: "基于应力波的低速滚动轴承故障诊断新方法研究", 机械强度, no. 06, pages 4 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114646907A (en) * | 2021-12-24 | 2022-06-21 | 中铁二院工程集团有限责任公司 | Rail transit low-frequency magnetic field measuring method based on machine learning |
| CN114646907B (en) * | 2021-12-24 | 2023-10-20 | 中铁二院工程集团有限责任公司 | Rail transit low-frequency magnetic field on-site measurement method based on machine learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112906644B (en) | Mechanical fault intelligent diagnosis method based on deep migration learning | |
| CN107657250B (en) | Bearing fault detection and positioning method and detection and positioning model implementation system and method | |
| CN110376522B (en) | Motor fault diagnosis method of data fusion deep learning network | |
| CN110849626A (en) | Self-adaptive sparse compression self-coding rolling bearing fault diagnosis system | |
| CN110866448A (en) | Flutter signal analysis method based on convolutional neural network and short-time Fourier transform | |
| CN113052000B (en) | Feature diagnosis method for early weak fault signals of ship mechanical equipment | |
| Chen et al. | Fault feature extraction and diagnosis of gearbox based on EEMD and deep briefs network | |
| Li et al. | Gear pitting fault diagnosis using raw acoustic emission signal based on deep learning | |
| CN116010900A (en) | Multi-scale feature fusion gearbox fault diagnosis method based on self-attention mechanism | |
| CN111783531A (en) | Water turbine set fault diagnosis method based on SDAE-IELM | |
| CN112287780A (en) | Spectral kurtosis algorithm-based mechanical equipment fault diagnosis method and system and readable storage medium | |
| CN114415018A (en) | Self-learning grating interference spectrum analysis technology for motor fault early warning | |
| CN112945546A (en) | Accurate diagnosis method for complex fault of gear box | |
| Zhang et al. | Rolling bearing fault diagnosis using improved deep residual shrinkage networks | |
| Ding et al. | Gear fault diagnosis based on VMD sample entropy and discrete hopfield neural network | |
| CN115165274A (en) | Self-adaptive intelligent monitoring device and method for vibration state of engineering mechanical equipment | |
| CN114510960A (en) | Method for recognizing distributed optical fiber sensor system mode | |
| CN114169368A (en) | Signal noise reduction method based on signal noise reduction self-encoder SDE | |
| CN111766513B (en) | Capsule network-based variable-working-condition multi-fault diagnosis method for three-phase induction motor | |
| Ghosh | Comparative DNN Model Analysis for Detection of Various types of Optical Noise | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium | |
| CN112802011A (en) | Fan blade defect detection method based on VGG-BLS | |
| CN112801033A (en) | AlexNet network-based construction disturbance and leakage identification method along long oil and gas pipeline | |
| CN115356599B (en) | Multi-mode urban power grid fault diagnosis method and system | |
| CN117030263A (en) | Bearing fault diagnosis method based on improved residual error network under multi-sensor signal fusion |
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 |