CN117607683A - Intelligent abnormality detection method for motor - Google Patents
Intelligent abnormality detection method for motor Download PDFInfo
- Publication number
- CN117607683A CN117607683A CN202311578889.7A CN202311578889A CN117607683A CN 117607683 A CN117607683 A CN 117607683A CN 202311578889 A CN202311578889 A CN 202311578889A CN 117607683 A CN117607683 A CN 117607683A
- Authority
- CN
- China
- Prior art keywords
- motor
- frequency
- delta
- detection
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 150
- 230000005856 abnormality Effects 0.000 title claims abstract description 53
- 230000005236 sound signal Effects 0.000 claims abstract description 49
- 230000002159 abnormal effect Effects 0.000 claims abstract description 47
- 238000000034 method Methods 0.000 claims abstract description 12
- 238000013527 convolutional neural network Methods 0.000 claims description 31
- 238000012549 training Methods 0.000 claims description 29
- 238000010586 diagram Methods 0.000 claims description 28
- 238000001228 spectrum Methods 0.000 claims description 28
- 230000006870 function Effects 0.000 claims description 20
- 230000004044 response Effects 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 8
- 230000005284 excitation Effects 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 3
- 230000003595 spectral effect Effects 0.000 claims description 3
- 230000006872 improvement Effects 0.000 description 15
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000012423 maintenance Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000010248 power generation Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000004075 alteration Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 239000000446 fuel Substances 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000001931 thermography Methods 0.000 description 1
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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention provides an intelligent abnormality detection method of a motor, which is characterized by comprising the following steps of: s1: acquiring a sound signal of a motor to be detected during operation through an audio collector array; s2: inputting the sound signal into an FP detection model to obtain a corresponding motor abnormality detection result; if the detection result is normal, turning to step S3; if the detection result is abnormal, the motor is abnormal; s3: inputting the sound signal into an AI detection model to obtain a corresponding motor abnormality detection result, wherein if the detection result is normal, the motor is not abnormal, otherwise, the motor is abnormal. Firstly, detecting by using an FP detection model, and when abnormality is detected, indicating that the motor is abnormal and ending detection; if the detection is normal, the AI model is further used for detection, if the detection results of the two models are normal, the motor is indicated to have no abnormality, and if the detection results of the two models are normal, the motor is indicated to have abnormality. The method can improve the detection accuracy and the detection efficiency of the abnormal motor and reduce the detection cost.
Description
Technical Field
The invention relates to the field of motor abnormality detection, in particular to an intelligent abnormality detection method for a motor.
Background
In recent years, new energy policies including india, norway, france and uk governments have been continuously pushed out all over the world, and the sales of fuel vehicles have been gradually prohibited until 2040 years, and the use of clean energy by terminal energy has become a new trend, and the use of electricity has also been gradually improved. The international energy agency estimates that the renewable energy power generation amount in 2020-2030 years gradually exceeds that of coal, and then wind power generation gradually becomes a main power generation source. The wind power generation mainly uses wind energy to drive fan blades to rotate by a wind power generator to convert the wind energy into electric energy, but because the wind power generator is leaked in an outdoor environment all the year round, the influence of severe weather is added, and the influence on the fan blades of the wind power generator is larger; however, from the structural and production point of view, the sector damage is difficult to eliminate only from the construction and production point of view. Therefore, it is extremely important to detect abnormality of the wind turbine.
In the prior art, the motor abnormality detection method includes Bragg grating (fiber Bragg grating, FBG), vibration analysis, infrared thermal imaging, optical image, X-ray, ultrasonic and the like. For example, the deformation of the fan blade is reversely pushed by the change of the refractive index in the Bragg light grating, but the installation in the mode is complex; while the vibration analysis method is effective and has a large measurement range, the sensor is required to be installed on the motor, and for the motor with complex and various parts, the interaction influence among different parts can cause that the measured vibration signal is not easy to judge; in addition, the method can be realized by adopting an acoustic emission technology (Acoustic emission, AE), but is limited by the characteristic of quick attenuation of acoustic emission signals, the possible occurrence position must be predicted before the sensor is installed, and the sensor is installed nearby, so that the overall detection efficiency is low, and the detection cost is high.
The motor abnormality detection has a great influence on the normal operation of the motor, so that the abnormal motor needs to be efficiently detected in industrial production practice, so that maintenance personnel can intervene in maintenance work earlier, and meanwhile, the normal motor needs to avoid missed detection as much as possible, and therefore, a more accurate detection result is needed. Therefore, how to improve the detection efficiency of abnormal detection of the motor, reduce the detection cost and simultaneously consider the accuracy of the detection result is a technical problem to be solved in the field.
Disclosure of Invention
The invention aims to provide a combined detection method based on an artificial intelligence technology and a sound spectrum characteristic, which can solve the technical problems mentioned in the background technology.
The invention provides an intelligent abnormality detection method for a motor, which is characterized by comprising the following steps:
s1: acquiring a sound signal of a motor to be detected during operation through an audio collector array;
s2: inputting the sound signal into an FP detection model to obtain a corresponding motor abnormality detection result; if the detection result is normal, turning to step S3; if the detection result is abnormal, the motor is abnormal;
s3: inputting the sound signal into an AI detection model to obtain a corresponding motor abnormality detection result, wherein if the detection result is normal, the motor is not abnormal, otherwise, the motor is abnormal;
the detection method of the FP detection model comprises the following steps:
s21: acquiring a signal to be detected, and calculating time-frequency response and power spectrum of the signal;
s22: calculating the frequency characteristic and the power characteristic of the signal to be detected according to the time-frequency response and the power spectrum;
s23: calculating multichannel characteristics corresponding to the frequency characteristics and the power characteristics according to the frequency characteristics and the power characteristics;
s24: detecting motor abnormality based on the multi-channel characteristics, and obtaining a detection result;
the detection method of the AI detection model comprises the following steps:
s31: acquiring a sound signal of historical operation of a motor, and calculating an instantaneous amplitude time-frequency diagram of the sound signal, wherein the sound signal also comprises a corresponding normal or abnormal mark;
s32: preprocessing and normalizing the instantaneous amplitude time-frequency diagram;
s33: taking the output data of the step S32 as training data;
s34: inputting the training data into a convolutional neural network for training, and continuously updating training parameters of the convolutional neural network;
s35: and (3) acquiring a sound signal of the motor to be detected, repeating the steps S31-S32 to obtain a time-frequency diagram corresponding to the signal to be detected, inputting the trained convolutional neural network, and outputting a classification result to obtain a detection result of motor abnormality.
As a further improvement of the present invention, the step S21 specifically includes: the audio collector array forms a multichannel collector and calculates the time-frequency response of the signal x (n), wherein the time-frequency response is calculated in the following manner:where l is the time index, k is the frequency index, w (N) is the Hamming window function, its length is equal to the length of the DFT transform, N FT Equal; the normalized power spectrum of the signal is:wherein when k=0 or N FT At/2, a=1, the remaining cases a=2.
As a further improvement of the present invention, the step S22 specifically includes:
acquiring the soundFrequency characteristic and power characteristic set fp= [ f ] of signal 1 ,f 2 …f 6 ]Comprising:
f 1 which is indicative of the power characteristics of the device,wherein k is s1 At 480Hz, k e Is 36KHz, l s1 0,l of a shape of 0,l e1 0.03s;
f 2 which is indicative of the characteristics of the high frequency,wherein k is s1 At 8KHz, k e Is 36KHz, l s1 0,l of a shape of 0,l e1 0.03s;
f 3 which is indicative of a power increase characteristic,wherein k is s At 8KHz, k e Is 36KHz, l s1 0,l of a shape of 0,l e1 0.03s, l s2 Is-0.1 s, l e2 Is-0.03 s;
f 4 which is indicative of the flatness of the frequency spectrum,wherein k is s At 8KHz, k e 36KHz;
f 5 the characteristic of the spectral shift is represented,wherein->s (l) represents the centroid of the power spectrum, P dB (k, l) denotes the logarithm of the power spectrum, F is the center frequency of all bands, k s At 8KHz, k e Is 36KHz, l s3 0,l of a shape of 0,l e3 0.01s, l s2 Is-0.1 s, l e2 Is-0.03 s;
f 6 which is indicative of a power spectrum drop characteristic, wherein k is s At 8KHz, k e Is 36KHz, l s4 0,l of a shape of 0,l e4 0.1s.
As a further improvement of the present invention, the step S23 specifically includes: the multi-channel features include a single-channel feature d ij And joint channel characteristics d ij
Where i is the index of the frequency signature and j is the index of the audio collector.
As a further improvement of the present invention, the step S24 specifically includes:
inputting the multi-channel characteristics into a multi-channel classifier, and obtaining a classification result, wherein the classification result is whether abnormality exists, and the classification rule of the classifier is as follows:
when all the following judgment rules are true, the classification result is abnormal, otherwise, the classification result is normal; the decision rule includes:
a if it is determined that 1j >δ 1 And then judge d 1 Whether or not it is greater than delta 1m Wherein delta 1 And delta 1m Is a corresponding threshold value;
b if it is determined that 2j >δ 2 And then judge d 2 Whether or not it is greater than delta 2m Wherein delta 2 And delta 2m Is a corresponding threshold value;
c if it is determined that 3j >δ 3 And then judge d 3 Whether or not it is greater than delta 3m Wherein delta 3 And delta 3m Is a corresponding threshold value;
d if it is determined that 4j <δ 4 And then judge d 4 Whether or not less than delta 4m Wherein delta 4 And delta 4m For the corresponding thresholdA value;
e if it is determined that d 5j <δ 5 And then judge d 5 Whether or not less than delta 5m Wherein delta 5 And delta 5m Is a corresponding threshold value;
f if it is determined that 6j <δ 6 And then judge d 6 Whether or not less than delta 6m Wherein delta 6 And delta 6m Is the corresponding threshold value.
As a further improvement of the present invention, the step S31 specifically includes: acquiring a historical sound signal p (n) when the motor operates, and calculating the time-frequency response of the sound signal p (n):where T is time, f is frequency index, τ is normalized time index, β is normalized frequency index, Λ is time window length, p (n) duration is T, w is window function, and the function value is [ T, t+L ]]The outer is 0, Δt is the sampling interval of the signal, Δf is the frequency interval, t=τΔt, f=βΔf, l=ΛΔt, p d ＝p((τ+d)Δt)，w d ＝w(-dΔt)，/>Points transformed for DFT; the amplitude of the sound signal is +.>The time-frequency diagram of the amplitude is: />
As a further improvement of the present invention, the step S32 includes: sampling rate 1/Δt=25600 Hz, the number of points of dft is 512, the frequency interval is 50Hz, and the window function length is 512; converting the amplitude to normalized powerWherein P is ref The sound pressure is used as a time-frequency diagram; removing the part lower than 4000Hz in the time-frequency diagram, and recording as S Γ*177 。
As a further improvement of the invention, the convolutional neural network comprises a data input layer, a convolutional calculation layer, a pooling layer, a full connection layer and an output layer; the data input layer is used for preprocessing original data, the convolution calculation layer is used for carrying out spatial domain filtering on the preprocessed data, the excitation layer is used for carrying out nonlinear mapping on the result of the convolution calculation layer, and the pooling layer is used for compressing data on the output result of the excitation layer; the full connection layer is used for mapping the final output to a linearly separable space; the output layer is used for outputting the classification result.
As a further improvement of the present invention, the convolutional neural network continuously updates training parameters through an error function loss, and specifically includes the following steps:
the error function loss is: wherein x (i) is an input of the convolutional neural network and y (i) is an output of the convolutional neural network;
the update formula of the training parameters is as follows:
wherein w is t Is the training parameter at the moment t, mu i (i=1, 2) is an exponential decay factor, l r Is the rate of learning to be performed,m t is a first order motion vector, V t Is a second order motion vector, m t ＝g t +μ 1 m t-1 -μ 1 g t ，
The invention provides an intelligent abnormality detection method of a motor, which is characterized by comprising the following steps of: s1: acquiring a sound signal of a motor to be detected during operation through an audio collector array; s2: inputting the sound signal into an FP detection model to obtain a corresponding motor abnormality detection result; if the detection result is normal, turning to step S3; if the detection result is abnormal, the motor is abnormal; s3: inputting the sound signal into an AI detection model to obtain a corresponding motor abnormality detection result, wherein if the detection result is normal, the motor is not abnormal, otherwise, the motor is abnormal; the detection method of the FP detection model comprises the following steps: s21: acquiring a signal to be detected, and calculating time-frequency response and power spectrum of the signal; s22: calculating the frequency characteristic and the power characteristic of the signal to be detected according to the time-frequency response and the power spectrum; s23: calculating multichannel characteristics corresponding to the frequency characteristics and the power characteristics according to the frequency characteristics and the power characteristics; s24: detecting motor abnormality based on the multi-channel characteristics, and obtaining a detection result; the detection method of the AI detection model comprises the following steps: s31: acquiring a sound signal of historical operation of a motor, and calculating an instantaneous amplitude time-frequency diagram of the sound signal, wherein the sound signal also comprises a corresponding normal or abnormal mark; s32: preprocessing and normalizing the instantaneous amplitude time-frequency diagram; s33: taking the output data of the step S32 as training data; s34: inputting the training data into a convolutional neural network for training, and continuously updating training parameters of the convolutional neural network;
s35: and (3) acquiring a sound signal of the motor to be detected, repeating the steps S31-S32 to obtain a time-frequency diagram corresponding to the signal to be detected, inputting the trained convolutional neural network, and outputting a classification result to obtain a detection result of motor abnormality. Firstly, detecting by using an FP detection model, and when abnormality is detected, indicating that the motor is abnormal and ending detection; if the detection is normal, the AI model is further used for detection, if the detection results of the two models are normal, the motor is indicated to have no abnormality, and if the detection results of the two models are normal, the motor is indicated to have abnormality.
Compared with the prior art, the invention has the main beneficial effects that:
1. the FP detection model utilizes a plurality of power features and frequency features of sound spectrum to detect the motor abnormality, and compared with the traditional method, the FP detection model is more efficient, has low algorithm complexity and strong adaptability, can effectively reduce the detection cost, can achieve more than 90% of detection accuracy, and meanwhile, the AI model of the invention utilizes a convolutional neural network model to detect, has high detection accuracy which can achieve more than 95%, and does not need to modify the motor structure;
2. aiming at the requirement of high detection efficiency of an abnormal motor, the design of the invention firstly utilizes the FP detection model to detect, compared with the AI model, the algorithm complexity is low, and after the FP detection model outputs an abnormal result, the FP detection model does not need to utilize the AI model to carry out secondary detection, thus the detection efficiency is high, the detection cost is low, and the maintenance personnel can conveniently intervene as soon as possible; for a normal motor, two detection models are needed to be used successively to ensure the detection accuracy, and meanwhile, the motor is determined to be abnormal when the detection is normal, so that the detection accuracy can be further improved, and the abnormal motor is avoided.
Drawings
Fig. 1 is a flowchart of steps of a method for detecting intelligent abnormality of a motor according to embodiment 1 of the present invention.
Fig. 2 is a block diagram of a convolutional neural network according to embodiment 1 of the present invention.
Fig. 3 is a block diagram of an intelligent abnormality detection apparatus for a motor according to embodiment 2 of the present invention.
Fig. 4 is an intelligent abnormality detection apparatus for a motor according to another embodiment 3 of the present invention.
Detailed Description
The invention is further illustrated below with reference to examples.
Example 1:
the first aspect of the present invention provides an intelligent anomaly detection method for a motor, as shown in fig. 1, which is characterized by comprising the following steps:
s1: acquiring a sound signal of a motor to be detected during operation through an audio collector array;
s2: inputting the sound signal into an FP detection model to obtain a corresponding motor abnormality detection result; if the detection result is normal, turning to step S3; if the detection result is abnormal, the motor is abnormal;
s3: inputting the sound signal into an AI detection model to obtain a corresponding motor abnormality detection result, wherein if the detection result is normal, the motor is not abnormal, otherwise, the motor is abnormal;
the detection method of the FP detection model comprises the following steps:
s21: acquiring a signal to be detected, and calculating time-frequency response and power spectrum of the signal;
s22: calculating the frequency characteristic and the power characteristic of the signal to be detected according to the time-frequency response and the power spectrum;
s23: calculating multichannel characteristics corresponding to the frequency characteristics and the power characteristics according to the frequency characteristics and the power characteristics;
s24: detecting motor abnormality based on the multi-channel characteristics, and obtaining a detection result;
the detection method of the AI detection model comprises the following steps:
s31: acquiring a sound signal of historical operation of a motor, and calculating an instantaneous amplitude time-frequency diagram of the sound signal, wherein the sound signal also comprises a corresponding normal or abnormal mark;
s32: preprocessing and normalizing the instantaneous amplitude time-frequency diagram;
s33: taking the output data of the step S32 as training data;
s34: inputting the training data into a convolutional neural network for training, and continuously updating training parameters of the convolutional neural network;
s35: and (3) acquiring a sound signal of the motor to be detected, repeating the steps S31-S32 to obtain a time-frequency diagram corresponding to the signal to be detected, inputting the trained convolutional neural network, and outputting a classification result to obtain a detection result of motor abnormality.
Compared with the prior art, the invention has the main beneficial effects that:
1. the FP detection model utilizes a plurality of power features and frequency features of sound spectrum to detect the motor abnormality, and compared with the traditional method, the FP detection model is more efficient, has low algorithm complexity and strong adaptability, can effectively reduce the detection cost, can achieve more than 90% of detection accuracy, and meanwhile, the AI model of the invention utilizes a convolutional neural network model to detect, has high detection accuracy which can achieve more than 95%, and does not need to modify the motor structure;
2. aiming at the requirement of high detection efficiency of an abnormal motor, the design of the invention firstly utilizes the FP detection model to detect, compared with the AI model, the algorithm complexity is low, and after the FP detection model outputs an abnormal result, the FP detection model does not need to utilize the AI model to carry out secondary detection, thus the detection efficiency is high, the detection cost is low, and the maintenance personnel can conveniently intervene as soon as possible; for a normal motor, two detection models are needed to be used successively to ensure the detection accuracy, and meanwhile, the motor is determined to be abnormal when the detection is normal, so that the detection accuracy can be further improved, and the abnormal motor is avoided.
As a further improvement of the present invention, the step S21 specifically includes: the audio collector array forms a multichannel collector and calculates the time-frequency response of the signal x (n), wherein the time-frequency response is calculated in the following manner:where l is the time index, k is the frequency index, w (N) is the Hamming window function, its length is equal to the length of the DFT transform, N FT Equal; the normalized power spectrum of the signal is:wherein when k=0 or N FT At/2, a=1, the remaining cases a=2.
As a further improvement of the present invention, the step S22 specifically includes:
acquiring a frequency characteristic and a power characteristic set FP= [ f ] of the sound signal 1 ,f 2 …f 6 ]Comprising:
f 1 which is indicative of the power characteristics of the device,wherein k is s1 At 480Hz, k e Is 36KHz, l s1 0,l of a shape of 0,l e1 0.03s;
f 2 which is indicative of the characteristics of the high frequency,wherein k is s1 At 8KHz, k e Is 36KHz, l s1 0,l of a shape of 0,l e1 0.03s;
f 3 which is indicative of a power increase characteristic,wherein k is s At 8KHz, k e Is 36KHz, l s1 0,l of a shape of 0,l e1 0.03s, l s2 Is-0.1 s, l e2 Is-0.03 s;
f 4 which is indicative of the flatness of the frequency spectrum,wherein k is s At 8KHz, k e 36KHz;
f 5 the characteristic of the spectral shift is represented,wherein->s (l) represents the centroid of the power spectrum, P dB (k, l) denotes the logarithm of the power spectrum, F is the center frequency of all bands, k s At 8KHz, k e Is 36KHz, l s3 0,l of a shape of 0,l e3 0.01s, l s2 Is-0.1 s, l e2 Is-0.03 s;
f 6 which is indicative of a power spectrum drop characteristic, wherein the method comprises the steps ofk s At 8KHz, k e Is 36KHz, l s4 0,l of a shape of 0,l e4 0.1s.
The sound of the abnormal motor has the following characteristics: 1. the abnormal sound belongs to burst pulse and is characterized in that the energy is suddenly increased and the energy is different. 2. This increase occurs over a wide frequency range, typically exceeding the hearing threshold for high frequencies in humans. 3. The frequency at which maximum power occurs may vary. 4. Starting from the frequency with the greatest power, its power decreases exponentially from the high frequency. 5. The decay of power over time is relatively slow and depends on the location of the lesions and sensors and the geometry of the sectors. The audio characteristics are calculated by using two frequency ranges. A range from about 480Hz to 36kHz is used because the ambient noise level in frequencies below about 480Hz is too high and 36kHz is the high frequency limit of the microphone. The second frequency range is from about 8 to 36kHz. In this reduced frequency range, the power of the ambient noise signal is significantly lower. The disadvantage is that the power decreases more with distance, especially when sound has to propagate at corners. Nevertheless, the reduced amount of ambient noise in this frequency range helps to provide more stable results for detection and classification purposes, so it is used for five of the six features. The time ranges for calculating different parameters are slightly different, and the total time ranges are between-0.1 s and 0.1s.
As a further improvement of the present invention, the step S23 specifically includes: the multi-channel features include a single-channel feature d ij And joint channel characteristics d ij
Where i is the index of the frequency signature and j is the index of the audio collector.
As a further improvement of the present invention, the step S24 specifically includes:
inputting the multi-channel characteristics into a multi-channel classifier, and obtaining a classification result, wherein the classification result is whether abnormality exists, and the classification rule of the classifier is as follows:
when all the following judgment rules are true, the classification result is abnormal, otherwise, the classification result is normal; the decision rule includes:
a if it is determined that 1j >δ 1 And then judge d 1 Whether or not it is greater than delta 1m Wherein delta 1 And delta 1m Is a corresponding threshold value;
b if it is determined that 2j >δ 2 And then judge d 2 Whether or not it is greater than delta 2m Wherein delta 2 And delta 2m Is a corresponding threshold value;
c if it is determined that 3j >δ 3 And then judge d 3 Whether or not it is greater than delta 3m Wherein delta 3 And delta 3m Is a corresponding threshold value;
d if it is determined that 4j <δ 4 And then judge d 4 Whether or not less than delta 4m Wherein delta 4 And delta 4m Is a corresponding threshold value;
e if it is determined that d 5j <δ 5 And then judge d 5 Whether or not less than delta 5m Wherein delta 5 And delta 5m Is a corresponding threshold value;
f if it is determined that 6j <δ 6 And then judge d 6 Whether or not less than delta 6m Wherein delta 6 And delta 6m Is the corresponding threshold value.
As a further improvement of the present invention, the step S31 specifically includes: acquiring a historical sound signal p (n) when the motor operates, and calculating the time-frequency response of the sound signal p (n):where T is time, f is frequency index, τ is normalized time index, β is normalized frequency index, Λ is time window length, p (n) duration is T, w is window function, and the function value is [ T, t+L ]]The outer is 0, Δt is the sampling interval of the signal, Δf is the frequency interval, t=τΔt, f=βΔf, l=ΛΔt, p d ＝p((τ+d)Δt)，w d ＝w(-dΔt)，/>Points transformed for DFT; the amplitude of the sound signal is +.>The time-frequency diagram of the amplitude is: />
As a further improvement of the present invention, the step S32 includes: sampling rate 1/Δt=25600 Hz, the number of points of dft is 512, the frequency interval is 50Hz, and the window function length is 512; converting the amplitude to normalized powerWherein P is ref The sound pressure is used as a time-frequency diagram; removing the part lower than 4000Hz in the time-frequency diagram, and recording as S Γ*177 。
FIG. 2 is a block diagram of a convolutional neural network provided by the present invention, the convolutional neural network including a data input layer, a convolutional calculation layer, a pooling layer, a full connection layer, and an output layer; the data input layer is used for preprocessing original data, the convolution calculation layer is used for carrying out spatial domain filtering on the preprocessed data, the excitation layer is used for carrying out nonlinear mapping on the result of the convolution calculation layer, and the pooling layer is used for compressing data on the output result of the excitation layer; the full connection layer is used for mapping the final output to a linearly separable space; the output layer is used for outputting the classification result.
As a further improvement of the present invention, the convolutional neural network continuously updates training parameters through an error function loss, and specifically includes the following steps:
the error function loss is: wherein x (i) isAn input of the convolutional neural network, y (i) being an output of the convolutional neural network;
the update formula of the training parameters is as follows:
wherein w is t Is the training parameter at the moment t, mu i (i=1, 2) is an exponential decay factor, l r Is the rate of learning to be performed,m t is a first order motion vector, V t Is a second order motion vector, m t ＝g t +μ 1 m t-1 -μ 1 g t ，
Example 2:
as shown in fig. 3, a second aspect of the present invention provides an intelligent anomaly detection device for a motor, including an acquisition module, an FP detection module, and an AI detection module, where:
the acquisition module is used for acquiring sound signals of the motor to be detected during operation through the audio collector array;
the FP detection module is used for inputting the sound signal into an FP detection model to obtain a corresponding motor abnormality detection result; if the detection result is normal, turning to step S3; if the detection result is abnormal, the motor is abnormal; the FP detection model is also used to implement the following functions: acquiring a signal to be detected, and calculating time-frequency response and power spectrum of the signal; calculating the frequency characteristic and the power characteristic of the signal to be detected according to the time-frequency response and the power spectrum; calculating multichannel characteristics corresponding to the frequency characteristics and the power characteristics according to the frequency characteristics and the power characteristics; and detecting motor abnormality based on the multichannel characteristics, and obtaining a detection result.
The AI detection module is used for inputting the sound signal into an AI detection model to obtain a corresponding motor abnormality detection result, if the detection result is normal, the motor is not abnormal, otherwise, the motor is abnormal; the AI detection module is also used for realizing the following functions: acquiring a sound signal of historical operation of a motor, and calculating an instantaneous amplitude time-frequency diagram of the sound signal, wherein the sound signal also comprises a corresponding normal or abnormal mark; preprocessing and normalizing the instantaneous amplitude time-frequency diagram; taking the output data of the last step as training data; inputting the training data into a convolutional neural network for training, and continuously updating training parameters of the convolutional neural network; and acquiring a sound signal of the motor to be detected, obtaining a time-frequency diagram corresponding to the signal to be detected, inputting the trained convolutional neural network, and outputting a classification result to obtain a detection result of motor abnormality.
Example 3
As shown in fig. 4, a third aspect of the present invention provides an intelligent abnormality detection apparatus for a motor, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program and implements the detection method as described in embodiment 1.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
Claims (9)
1. The intelligent abnormality detection method for the motor is characterized by comprising the following steps of:
s1: acquiring a sound signal of a motor to be detected during operation through an audio collector array;
s2: inputting the sound signal into an FP detection model to obtain a corresponding motor abnormality detection result; if the detection result is normal, turning to step S3; if the detection result is abnormal, the motor is abnormal;
s3: inputting the sound signal into an AI detection model to obtain a corresponding motor abnormality detection result, wherein if the detection result is normal, the motor is not abnormal, otherwise, the motor is abnormal;
the detection method of the FP detection model comprises the following steps:
s21: acquiring a signal to be detected, and calculating time-frequency response and power spectrum of the signal;
s22: calculating the frequency characteristic and the power characteristic of the signal to be detected according to the time-frequency response and the power spectrum;
s23: calculating multichannel characteristics corresponding to the frequency characteristics and the power characteristics according to the frequency characteristics and the power characteristics;
s24: detecting motor abnormality based on the multi-channel characteristics, and obtaining a detection result;
the detection method of the AI detection model comprises the following steps:
s31: acquiring a sound signal of historical operation of a motor, and calculating an instantaneous amplitude time-frequency diagram of the sound signal, wherein the sound signal also comprises a corresponding normal or abnormal mark;
s32: preprocessing and normalizing the instantaneous amplitude time-frequency diagram;
s33: taking the output data of the step S32 as training data;
s34: inputting the training data into a convolutional neural network for training, and continuously updating training parameters of the convolutional neural network;
s35: and (3) acquiring a sound signal of the motor to be detected, repeating the steps S31-S32 to obtain a time-frequency diagram corresponding to the signal to be detected, inputting the trained convolutional neural network, and outputting a classification result to obtain a detection result of motor abnormality.
2. The method for detecting intelligent abnormality of motor according to claim 1, wherein said step S21 specifically includes: by a means ofThe audio collector array forms a multichannel collector and calculates the time-frequency response of the signal x (n), wherein the time-frequency response is calculated in the following manner: where l is the time index, k is the frequency index, w (N) is the Hamming window function, its length is equal to the length of the DFT transform, N FT Equal; the normalized power spectrum of the signal is: wherein when k=0 or N FT At/2, a=1, the remaining cases a=2.
3. The method for detecting intelligent abnormality of motor according to claim 2, wherein said step S22 is specifically:
acquiring a frequency characteristic and a power characteristic set FP= [ f ] of the sound signal 1 ，f 2 …f 6 ]Comprising:
f 1 which is indicative of the power characteristics of the device,wherein k is s1 At 480Hz, k e Is 36KHz, l s1 0,l of a shape of 0,l e1 0.03s;
f 2 which is indicative of the characteristics of the high frequency,wherein k is s1 At 8KHz, k e Is 36KHz, l s1 0,l of a shape of 0,l e1 0.03s;
f 3 which is indicative of a power increase characteristic,wherein k is s At 8KHz, k e Is 36KHz, l s1 0,l of a shape of 0,l e1 0.03s, l s2 Is-0.1 s, l e2 Is-0.03 s;
f 4 which is indicative of the flatness of the frequency spectrum,wherein k is s At 8KHz, k e 36KHz;
f 5 the characteristic of the spectral shift is represented,wherein->s (l) represents the centroid of the power spectrum, P dB (k, l) denotes the logarithm of the power spectrum, F is the center frequency of all bands, k s At 8KHz, k e Is 36KHz, l s3 0,l of a shape of 0,l e3 0.01s, l s2 Is-0.1 s, l e2 Is-0.03 s;
f 6 which is indicative of a power spectrum drop characteristic, wherein k is s At 8KHz, k e Is 36KHz, l s4 0,l of a shape of 0,l e4 0.1s.
4. The method for detecting intelligent abnormality of motor according to claim 3, wherein said step S23 is specifically: the multi-channel features include a single-channel feature d ij And joint channel characteristics d ij
Where i is the index of the frequency signature and j is the index of the audio collector.
5. The method for detecting intelligent abnormality of motor according to claim 4, wherein said step S24 is specifically:
inputting the multi-channel characteristics into a multi-channel classifier, and obtaining a classification result, wherein the classification result is whether abnormality exists, and the classification rule of the classifier is as follows:
when all the following judgment rules are true, the classification result is abnormal, otherwise, the classification result is normal; the decision rule includes:
a if it is determined that 1j ＞δ 1 And then judge d 1 Whether or not it is greater than delta 1m Wherein delta 1 And delta 1m Is a corresponding threshold value;
b if it is determined that 2j ＞δ 2 And then judge d 2 Whether or not it is greater than delta 2m Wherein delta 2 And delta 2m Is a corresponding threshold value;
c if it is determined that 3j ＞δ 3 And then judge d 3 Whether or not it is greater than delta 3m Wherein delta 3 And delta 3m Is a corresponding threshold value;
d if it is determined that 4j ＜δ 4 And then judge d 4 Whether or not less than delta 4m Wherein delta 4 And delta 4m Is a corresponding threshold value;
e if it is determined that d 5j ＜δ 5 And then judge d 5 Whether or not less than delta 5m Wherein delta 5 And delta 5m Is a corresponding threshold value;
f if it is determined that 6j ＜δ 6 And then judge d 6 Whether or not less than delta 6m Wherein delta 6 And delta 6m Is the corresponding threshold value.
6. The intelligent abnormality detection method of a motor according to claim 1The method is characterized in that the step S31 specifically comprises the following steps: acquiring a historical sound signal p (n) when the motor operates, and calculating the time-frequency response of the sound signal p (n):where T is time, f is frequency index, τ is normalized time index, β is normalized frequency index, Λ is time window length, p (n) duration is T, w is window function, and the function value is [ T, t+L ]]The outer is 0, Δt is the sampling interval of the signal, Δf is the frequency interval, t=τΔt, f=βΔf, l=ΛΔt, p d ＝p((τ+d)Δt)，w d ＝w(-dΔt)，/>Points transformed for DFT; the amplitude of the sound signal isThe amplitude time-frequency diagram is as follows: />
7. The intelligent anomaly detection method for a motor according to claim 6, wherein the step S32 includes: sampling rate 1/Δt=25600 Hz, the number of points of dft is 512, the frequency interval is 50Hz, and the window function length is 512; converting the amplitude to normalized powerWherein P is ref As a time-frequency diagram, Γ is the number of time points of the time-frequency diagram; removing the part lower than 4000Hz in the time-frequency diagram, and recording as S Γ*177 。
8. The intelligent anomaly detection method of the motor according to claim 7, wherein the convolutional neural network comprises a data input layer, a convolutional calculation layer, a pooling layer, a full connection layer and an output layer; the data input layer is used for preprocessing original data, the convolution calculation layer is used for carrying out spatial domain filtering on the preprocessed data, the excitation layer is used for carrying out nonlinear mapping on the result of the convolution calculation layer, and the pooling layer is used for compressing data on the output result of the excitation layer; the full connection layer is used for mapping the final output to a linearly separable space; the output layer is used for outputting the classification result.
9. The method for detecting intelligent anomalies of a motor according to claim 8, characterized in that said convolutional neural network continuously updates training parameters through an error function loss, comprising the steps of:
the error function loss is: wherein x (i) is an input of the convolutional neural network and y (i) is an output of the convolutional neural network;
the update formula of the training parameters is as follows:
wherein w is t Is the training parameter at the moment t, mu i (i=1, 2) is an exponential decay factor, l r Is the rate of learning to be performed,m t is a first order motion vector, V t Is a second order motion vector, m t ＝g t +μ 1 m t-1 -μ 1 g t ，
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311578889.7A CN117607683B (en) | 2023-11-24 | 2023-11-24 | Intelligent abnormality detection method for motor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311578889.7A CN117607683B (en) | 2023-11-24 | 2023-11-24 | Intelligent abnormality detection method for motor |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117607683A true CN117607683A (en) | 2024-02-27 |
| CN117607683B CN117607683B (en) | 2024-05-10 |
Family
ID=89945720
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311578889.7A Active CN117607683B (en) | 2023-11-24 | 2023-11-24 | Intelligent abnormality detection method for motor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117607683B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109416849A (en) * | 2016-06-16 | 2019-03-01 | 卡塔尔大学 | By convolutional neural networks come the method and apparatus of actuating motor fault detection |
| CN112924749A (en) * | 2021-02-04 | 2021-06-08 | 西安电子科技大学 | Unsupervised counterstudy electromagnetic spectrum abnormal signal detection method |
| CN116168720A (en) * | 2023-02-28 | 2023-05-26 | 盐城工学院 | Motor sound abnormality fault diagnosis method, system and storable medium |
-
2023
- 2023-11-24 CN CN202311578889.7A patent/CN117607683B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109416849A (en) * | 2016-06-16 | 2019-03-01 | 卡塔尔大学 | By convolutional neural networks come the method and apparatus of actuating motor fault detection |
| CN112924749A (en) * | 2021-02-04 | 2021-06-08 | 西安电子科技大学 | Unsupervised counterstudy electromagnetic spectrum abnormal signal detection method |
| CN116168720A (en) * | 2023-02-28 | 2023-05-26 | 盐城工学院 | Motor sound abnormality fault diagnosis method, system and storable medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117607683B (en) | 2024-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2898931C (en) | System and method for enhanced operation of wind parks | |
| JP7199608B2 (en) | Methods and apparatus for inspecting wind turbine blades, and equipment and storage media therefor | |
| CN109763944B (en) | Non-contact monitoring system and monitoring method for blade faults of offshore wind turbine | |
| AU2011253741A1 (en) | Wind turbine generator system vibration detecting device and method | |
| Chen et al. | Acoustical damage detection of wind turbine blade using the improved incremental support vector data description | |
| CN113298134B (en) | System and method for remotely and non-contact health monitoring of fan blade based on BPNN | |
| KR102235808B1 (en) | Operation prediction system, method and computer program of industrial plant facility using machine learning model based on signal group | |
| CN111415070A (en) | Wind turbine generator gearbox oil temperature over-temperature fault early warning method based on SCADA data | |
| CN112067701B (en) | Fan blade remote auscultation method based on acoustic diagnosis | |
| CN109061302A (en) | A kind of wind power generator incorporated in power network group harmonic measure system converted based on EEMD and Hilbert | |
| CN104101652A (en) | Audio signal based wind power blade damage monitoring method and system | |
| CN115249972B (en) | Performance evaluation system and method for wind generating set of wind power plant under big data | |
| Lin et al. | Condition monitoring and fault diagnosis of roller element bearing | |
| CN110554090A (en) | Wind turbine generator and crack monitoring system and method of variable-pitch bearing of wind turbine generator | |
| CN112052712B (en) | Power equipment state monitoring and fault identification method and system | |
| CN117607683B (en) | Intelligent abnormality detection method for motor | |
| Sánchez et al. | Wind Turbines Acoustic Inspections performed with UAV and sound frequency domain analysis | |
| Zhao et al. | Damage detection of wind turbine blade based on wavelet analysis | |
| CN115859148A (en) | Fan blade vibration alarm method and device | |
| CN113919525A (en) | Power station fan state early warning method, system and application thereof | |
| CN116861219B (en) | Wind turbine generator pitch-variable fault diagnosis method | |
| CN111398810A (en) | Slip ring electric spark detection and diagnosis system and detection and diagnosis method | |
| CN114785414B (en) | Identification method and identification system for external acoustic interference of optical fiber composite submarine cable | |
| CN116092518A (en) | Wind driven generator blade state identification method, device and storage medium | |
| CN112780503B (en) | Fan blade protective paint damage monitoring method and system based on audio signals |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |