CN114509158B - Acoustic-vibration-fused blade crack fault detection method and application - Google Patents

Acoustic-vibration-fused blade crack fault detection method and application Download PDF

Info

Publication number
CN114509158B
CN114509158B CN202210001190.3A CN202210001190A CN114509158B CN 114509158 B CN114509158 B CN 114509158B CN 202210001190 A CN202210001190 A CN 202210001190A CN 114509158 B CN114509158 B CN 114509158B
Authority
CN
China
Prior art keywords
vibration
fusion
sample
samples
channels
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.)
Active
Application number
CN202210001190.3A
Other languages
Chinese (zh)
Other versions
CN114509158A (en
Inventor
宋狄
许飞云
胡建中
贾民平
黄鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Original Assignee
Southeast University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southeast University filed Critical Southeast University
Priority to CN202210001190.3A priority Critical patent/CN114509158B/en
Publication of CN114509158A publication Critical patent/CN114509158A/en
Application granted granted Critical
Publication of CN114509158B publication Critical patent/CN114509158B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/04—Analysing solids
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
    • G01N29/4481—Neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/02—Indexing codes associated with the analysed material
    • G01N2291/023—Solids
    • G01N2291/0234—Metals, e.g. steel
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E10/00—Energy generation through renewable energy sources
    • Y02E10/70—Wind energy
    • Y02E10/72—Wind turbines with rotation axis in wind direction

Abstract

The invention discloses an urgent application of a sound-vibration fusion blade crack fault detection method, which comprises the following steps: collecting two-channel acoustic signals and two-channel vibration signals, and dividing the signals into training and testing samples after standardized processing; respectively fusing two channels of vibration signal training and testing samples by using a data-level fusion method, and generating vibration data fusion training and testing samples; inputting the two-channel acoustic signals, the two-channel vibration signals and the vibration data fusion training sample into a one-dimensional convolutional neural network to obtain an initial detection result; and fusing the initial detection result by using a decision-level fusion method, and obtaining a final detection result to realize crack fault detection of the compressor blade. Compared with other existing crack detection technologies, the method is simple and easy to implement, and can be used for realizing crack fault detection of the compressor blade by fusing the sound vibration signals at a data level and a decision level.

Description

Acoustic-vibration-fused blade crack fault detection method and application
Technical Field
The invention relates to the technical field of analysis and fusion of acoustic vibration signals of rotary machinery, in particular to a blade crack fault detection method for acoustic vibration fusion.
Background
The compressor is widely applied to the fields of petrochemical industry, electric power and the like, and the blades are used as core components and are easy to generate crack faults under the actions of centrifugal force, friction and unstable airflow load, so that the normal operation of the whole compressor is influenced. Therefore, the method for detecting the crack faults of the blades in time has important significance for ensuring safe and stable operation of the compressor. In addition, in actual engineering, an acoustic signal, a vibration signal, or the like is often used for blade failure detection. However, a single acoustic signal or vibration signal is often doped with strong background noise, and fault analysis performed through a single signal can result in low detection accuracy and unreliable detection results. Therefore, how to effectively utilize and integrate the acoustic signals and the vibration signals and realize reliable and accurate detection of the crack faults of the compressor blades is worth of intensive research.
Traditional signal fusion includes 3 fusion modes: data level, feature level and decision level fusion. In general, data level fusion is applicable to signals of different positions of the same type, and is the lowest level fusion mode; the feature level and decision level fusion can fuse different types of signals, wherein the decision level fusion is the highest fusion mode and has the best fusion effect. A single fusion mode may not accurately detect blade cracks due to the presence of strong noise, and different similar fusions often have some conflict. Therefore, there is an urgent need for a method for detecting crack faults of a compressor blade, which can perform different degrees of sound vibration signal fusion according to different signal characteristics and realize reliability and accuracy.
Disclosure of Invention
Aiming at the defects, the invention provides a blade crack fault detection method by sound and vibration fusion, which overcomes the current situations that the existing single signal and single fusion mode are difficult to accurately reflect the crack fault characteristics of the compressor blade and cannot accurately detect the crack fault of the blade, and effectively realizes sound and vibration signal fusion and compressor blade crack fault detection.
In order to solve the technical problems, the invention adopts the following technical scheme:
a crack fault detection method for a sound-vibration integrated compressor blade comprises the following steps:
the blade crack fault detection method based on the sound and vibration fusion is characterized by comprising the following steps of:
step 1: collecting at least two channels of sound signals and at least two channels of vibration signals;
step 2: carrying out standardization processing on the collected at least two channels of sound signals and at least two channels of vibration signals, and dividing the standardized sound signals and vibration signals into training samples and test samples;
step 3: respectively fusing at least two channels of vibration signal training and testing samples by using a data-level fusion method, and generating vibration data fusion training and testing samples;
step 4: the training samples of the sound signals and the vibration signals processed in the step 2 and the vibration data generated in the step 3 are fused and input into a neural network to obtain an initial detection result;
step 5: and (3) fusing the initial detection result obtained in the step (4) by using a decision-level fusion method, and obtaining a final detection result to realize crack fault detection of the blade.
In the step 1, acoustic signals are acquired by adopting an acoustic pressure sensor, the acoustic pressure sensor is respectively located at a set distance from an air inlet and an air outlet, the set distance is 0.2-1.0 m, a set included angle is formed by inclining upwards, and the set included angle is 45 degrees.
In the step 2, the normalization process is a 0-1 normalization process, and the expression of the 0-1 normalization process is as follows:
Figure BDA0003454154710000031
wherein x is s (t) represents a normalized signal of 0-1, x (t) represents an original signal, and min (x) and max (x) represent a minimum value and a maximum value of the original signal, respectively.
The specific steps of the data level fusion method in the step 3 are as follows:
step 3.1, respectively calculating the Hoyer distance of the two-channel samples, wherein the Hoyer distance expression is as follows:
Figure BDA0003454154710000032
wherein H is x The Hoyer distance for sample x is represented, and t represents the length of sample x.
Step 3.2, respectively calculating the modified cosine similarity of the two-channel samples according to the historical samples and the Hoyer distance, wherein the modified cosine similarity expression is as follows:
Figure BDA0003454154710000033
Figure BDA0003454154710000034
wherein θ x Represents the cosine angle of sample x, x h Representing historical samples, G (x, x, H x )、G(x,x h ,H x ) And G (x) h ,x h ,H x ) Representing sample x, and history sample x between samples x h History sample x h Gaussian distance between them.
Step 3.3, fusing the two-channel samples according to the modified cosine similarity of the two-channel samples, wherein the fused expression is as follows:
Figure BDA0003454154710000041
Figure BDA0003454154710000042
wherein x is df Representing the fused samples, ω i Representing the weight of the i-th sample.
Preferably, in the step 4, the one-dimensional convolutional neural network includes 2 convolutional layers, 2 pooling layers, 1 full-connection layer and 1 Softmax layer, the convolutional layer filter size is 6*1, the step size of the convolutional layers is 1, the first layer of convolutional layer channel number is 4, the second layer of convolutional layer channel number is 8, the pooling layer filter size is 6*1, the step size of the pooling layers is 6, the first layer of pooling layer channel number is 4, the second layer of pooling layer channel number is 8, and the output number of Softmax layer is 3.
Preferably, the specific steps of the decision-level fusion method in step 5 are as follows:
step 5.1, respectively calculating the precision of different samples and the accuracy of different categories in the initial detection result, wherein the expressions of the precision and the accuracy are as follows:
Figure BDA0003454154710000043
Figure BDA0003454154710000044
wherein P is ij Representing the precision of the jth class in the ith sample, A i Representing the accuracy of the ith sample, TP ij 、TN ij 、FP ij And FN ij The numbers of the true positive sample, the true negative sample, the false positive sample and the false negative sample are respectively represented.
Step 5.2, calculating the credibility of the sample, wherein the expression of the credibility is as follows:
Figure BDA0003454154710000045
wherein, CP ij Indicating the trustworthiness of the jth class in the ith sample.
Step 5.3, calculating an initial decision stage fusion result, wherein the expression of the initial decision stage fusion is as follows:
Figure BDA0003454154710000051
wherein the FPT j And (3) representing the initial decision-stage fusion result of the j-th category, and m represents the number of initial detection results.
And 5.4, verifying an initial decision stage fusion result. When the probability of the initial decision level fusion result exceeds 60%, the effective result is considered, and the initial decision level fusion result is the final decision level fusion result; otherwise, removing the initial detection result with the lowest accuracy, and carrying out the steps 5.1-5.3 again, and verifying until the requirements are met.
The beneficial effects of the invention are as follows:
1) According to the method for detecting the crack fault of the compressor blade through the sound and vibration fusion, provided by the invention, the two-channel vibration signals are fused through the data level, the initial detection of the sound signals, the vibration signals and the vibration data level fusion signals is realized by combining the one-dimensional convolutional neural network, and finally, the decision level fusion is carried out on the initial detection result to realize the crack fault detection of the compressor blade. The defect that the single signal or the single fusion mode cannot accurately detect the crack fault of the blade is avoided, and the accuracy of detecting the crack fault of the blade of the compressor is improved by fusing the sound vibration signal at the data level and the decision level.
2) The method for fusing the data level provided by the invention can analyze the sparseness degree of the vibration signal and the similarity with the historical signal, further accurately reflect the crack fault characteristics of the blade and improve the reliability of the vibration signal.
3) The decision-level fusion method provided by the invention performs reliability distribution by calculating the precision and accuracy of the initial detection result, improves the reliability of sound vibration fusion on the basis of verifying the initial decision-level fusion result, and realizes high-accuracy detection of blade crack faults.
Drawings
FIG. 1 is a flow diagram of an embodiment of the present invention;
FIG. 2 is a flow chart of a data level fusion method in an embodiment of the invention;
FIG. 3 is a flow chart of a decision level fusion method in an embodiment of the invention.
Detailed Description
The following description is merely exemplary in nature and is not intended to limit the present disclosure, application, or uses. It should be understood that throughout the drawings, corresponding reference numerals indicate like or corresponding parts and features.
As shown in fig. 1, the method for detecting the crack fault of the sound and vibration fusion compressor blade comprises the following steps:
step 1: two sound pressure sensors are arranged at the air inlet and the air outlet of the compressor, two channels of sound signals are collected, two vibration sensors are arranged near the main shaft of the motor, and two channels of vibration signals are collected.
Step 2: and carrying out standardization processing on the four-way acoustic signal and the vibration signal, and dividing the standardized acoustic signal and the standardized vibration signal into a training sample and a test sample.
Step 3: and respectively fusing the two channels of vibration signal training and testing samples by using a data level fusion method, and generating vibration data fusion training and testing samples.
Step 4: and inputting the two-channel acoustic signals, the two-channel vibration signals and the vibration data fusion training sample into a one-dimensional convolutional neural network to obtain an initial detection result.
Step 5: and fusing the initial detection result by using a decision-level fusion method, and obtaining a final detection result to realize crack fault detection of the compressor blade.
In the above embodiment, two sound pressure sensors collect two-channel sound signals at the air inlet and the air outlet of the compressor, two vibration sensors collect two-channel vibration signals near the main shaft of the motor, the two vibration signals are divided into training and test samples after standardized processing, the two-channel vibration signals are fused by using a data level fusion method and are input into a one-dimensional convolutional neural network to obtain an initial detection result, and finally the initial detection result is fused by using a decision level fusion method to realize crack fault detection of the compressor blade.
It should be noted that, the sampling frequency of the acoustic signal and the vibration signal is 5120Hz, the sampling time is 1s, and the training sample and the test sample are according to 7:3, the training sample and the test sample may have certain difference due to the randomness of the division, but the subsequent processing is not affected.
In the step 1, the sound pressure sensor is respectively 0.5 meter away from the air inlet and the air outlet, and an included angle of 45 degrees is formed by inclining upwards.
The distance is referred to the center of the air inlet and the air outlet, and the 45 ° is referred to the horizontal direction.
In the step 2, the normalization process is a 0-1 normalization process, and the expression of the 0-1 normalization process is as follows:
Figure BDA0003454154710000071
wherein x is s (t) represents a normalized signal of 0-1, x (t) represents an original signal, and min (x) and max (x) represent a minimum value and a maximum value of the original signal, respectively.
In the step 3, the specific steps of the data-level fusion method are as follows:
step 3.1, respectively calculating the Hoyer distances of the two-channel samples, wherein the Hoyer distance expression is as follows:
Figure BDA0003454154710000081
wherein H is x The Hoyer distance for sample x is represented, and t represents the length of sample x.
Step 3.2, respectively calculating the modified cosine similarity of the two-channel samples according to the historical samples and the Hoyer distance, wherein the modified cosine similarity expression is as follows:
Figure BDA0003454154710000082
/>
Figure BDA0003454154710000083
wherein θ x Represents the cosine angle of sample x, x h Representing historical samples, G (x, x, H x )、G(x,x h ,H x ) And G (x) h ,x h ,H x ) Representing sample x, and history sample x between samples x h History sample x h Gaussian distance between them.
Step 3.3, fusing the two-channel samples according to the modified cosine similarity of the two-channel samples, wherein the fused expression is as follows:
Figure BDA0003454154710000084
Figure BDA0003454154710000085
wherein x is df Representing the fused samples, ω i Representing the weight of the i-th sample.
The history sample is the last sample of sample x, and the value of the history sample of the first sample is set to 0.
In the step 4, the one-dimensional convolutional neural network comprises 2 convolutional layers, 2 pooling layers, 1 full-connection layer and 1 Softmax layer, the size of the convolutional layer filter is 6*1, the step length of the convolutional layers is 1, the number of channels of the first layer of convolutional layers is 4, the number of channels of the second layer of convolutional layers is 8, the size of the pooling layer filter is 6*1, the step length of the pooling layers is 6, the number of channels of the first layer of pooling layers is 4, the number of channels of the second layer of pooling layers is 8, and the output number of Softmax layers is 3.
It should be noted that the one-dimensional convolutional neural network is built in the order of convolutional layer, pooling layer, full-connectivity and Softmax layer.
In step 5, the decision-level fusion method comprises the following specific steps:
step 5.1, respectively calculating the precision of different samples and the accuracy of different categories in the initial detection result, wherein the expressions of the precision and the accuracy are as follows:
Figure BDA0003454154710000091
Figure BDA0003454154710000092
wherein P is ij Representing the precision of the jth class in the ith sample, A i Representing the accuracy of the ith sample, TP ij 、TN ij 、FP ij And FN ij The numbers of the true positive sample, the true negative sample, the false positive sample and the false negative sample are respectively represented.
Step 5.2, calculating the sample credibility, wherein the expression of the credibility is as follows:
Figure BDA0003454154710000093
wherein, CP ij Indicating the trustworthiness of the jth class in the ith sample.
Step 5.3, calculating an initial decision stage fusion result, wherein the expression of the initial decision stage fusion is as follows:
Figure BDA0003454154710000094
wherein the FPT j And (3) representing the initial decision-stage fusion result of the j-th category, and m represents the number of initial detection results.
And 5.4, verifying an initial decision stage fusion result. When the probability of the initial decision-level fusion result exceeds 60%, the initial decision-level fusion result is considered as an effective result, and the initial decision-level fusion result is a final decision-level fusion result; otherwise, removing the initial detection result with the lowest accuracy, and carrying out the steps 5.1-5.3 again, and verifying until the requirements are met.
To further verify the effectiveness of the inventive method, the solution of the present application is further illustrated using a compressor blade crack test bench:
the compressor blade crack test bench had blades of 3 crack lengths total of 0, 10mm and 20mm and was operated at four speeds of 1200rpm, 1500rpm, 1800rpm and 2100 rpm. And acquiring two-channel acoustic signals and vibration signals, and dividing the signals into training samples and test samples after standardized processing. And the two-channel vibration samples are fused by adopting a data level fusion method, and the flow of the data level fusion method is shown in figure 2. And inputting the fusion sample of the acoustic signal, the vibration signal and the vibration data into a one-dimensional convolutional neural network for detection, and obtaining an initial detection result. And finally, fusing the initial detection result by using a decision-level fusion method, wherein the flow is shown in figure 3. And carrying out crack fault classification and detection according to the decision-stage fusion result, wherein the crack fault detection results of the compressor blade at 4 rotating speeds are shown in table 1. It has been found that the fault detection accuracy of a single acoustic or vibration signal sample is low due to noise interference. After data level fusion is carried out, the accuracy of crack fault detection of the compressor blade is greatly improved, and the accuracy is totally over 90 percent. Finally, after decision-level fusion, the accuracy of blade crack fault detection exceeds 98% at 4 rotational speeds, and 100% accuracy is achieved at 1500rpm, so that the accuracy and reliability of the method are verified.
Table 1 compressor blade crack failure detection results
Rotational speed 1200rpm 1500rpm 1800rpm 2100rpm
Acoustic signal sample 1 83.2％ 88.8％ 79.6％ 72.9％
Acoustic signal sample 2 92.3％ 76.3％ 84.6％ 77.4％
Vibration signal sample 1 95.4％ 87.0％ 84.3％ 85.7％
Vibration signal sample 2 95.2％ 88.4％ 85.1％ 87.1％
Data level fusion 97.1％ 93.2％ 90.2％ 95.6％
Decision level fusion 99.3％ 100％ 98.8％ 99.3％
To further illustrate the advantages of the method of the present invention in the detection of compressor blade cracks, a comparison was made with other fusion methods. The method comprises a feature level fusion method based on a one-dimensional convolutional neural network, a decision level fusion method based on voting, a fusion method based on a D-S evidence theory and a decision level fusion method based on the convolutional neural network. Based on the above experimental original sound vibration signals and initial results, four comparison methods were used to detect compressor blade cracks, respectively, the results are shown in table 2.
TABLE 2 comparison of different fusion methods
Figure BDA0003454154710000111
It can be found that the feature level fusion method based on the one-dimensional convolutional neural network is to fuse on the full-connected layer, and the average precision is 97.82%. However, the accuracy of this method decreases with increasing rotational speed, and is subject to strong noise interference at high rotational speeds, which is mainly because this method is too dependent on the characteristics of deep network extraction, and the accuracy of this method is not as good as the method of the present invention under all conditions.
For the voting-based decision-level fusion method, the accuracy at 1800rpm and 2100rpm was 94.92% and 95.74%, respectively. This is because the result is not analyzed for authenticity and the blindly obtained voting results are not accurate enough. In addition, if the voting results of the two tags are identical, randomness and confusion are easily generated.
For the fusion method based on the D-S evidence theory, the average precision is 97.76% as a classical decision-stage fusion method, but the method is lower than the method of the invention.
For the decision-level fusion method based on the convolutional neural network, the average precision is only 96.81%, which is the worst in all methods. By direct CNN fusion, acoustic overfitting is easily produced, resulting in poor accuracy.
For the method of the present invention, it has the highest accuracy under four working conditions and is superior to other comparative methods. The result shows that the method has good noise immunity and can accurately detect cracks.
Compared with the prior art, the method for detecting the crack fault of the compressor blade by adopting the sound and vibration fusion has the advantages that two channels of vibration signals are fused through a data level, the initial detection of the sound signals, the vibration signals and the vibration data level fusion signals is realized by combining a one-dimensional convolutional neural network, and finally, the decision level fusion is carried out on the initial detection result to realize the crack fault detection of the compressor blade. The defect that the single signal or single fusion mode cannot accurately detect the crack fault of the blade is avoided, and the accuracy of detecting the crack fault of the blade of the compressor is improved.
The embodiments described herein are only preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Equivalent changes and modifications of the invention are intended to be within the scope of the present invention.

Claims (7)

1. The blade crack fault detection method based on the sound and vibration fusion is characterized by comprising the following steps of:
step 1: collecting at least two channels of sound signals and at least two channels of vibration signals; the two channels of sound signals are signals at a set distance from the air inlet and the air outlet respectively;
step 2: carrying out standardization processing on the collected at least two channels of sound signals and at least two channels of vibration signals, and dividing the standardized sound signals and vibration signals into training samples and test samples;
step 3: respectively fusing at least two channels of vibration signal training and testing samples by using a data-level fusion method, and generating vibration data fusion training and testing samples;
step 4: the training samples of the sound signals and the vibration signals processed in the step 2 and the vibration data generated in the step 3 are fused and input into a neural network to obtain an initial detection result;
step 5: fusing the initial detection result obtained in the step 4 by using a decision-level fusion method, and obtaining a final detection result to realize crack fault detection of the blade;
in step 3, at least two channels of vibration signal training and testing samples are respectively fused by using a data level fusion method, including:
step 3.1, respectively calculating the Hoyer distance of the two-channel samples:
Figure FDA0004158601410000011
wherein H is x The Hoyer distance for sample x, t for sample x;
step 3.2, respectively calculating the modified cosine similarity of the two-channel samples according to the historical samples and the Hoyer distance:
Figure FDA0004158601410000021
wherein θ x Represents the cosine angle of sample x, x h Representing historical samples, G (x, x, H x ) Represents the Gaussian distance between samples x, G (x, x h ，H x ) Representing sample x and history sample x h Gaussian distance between each other, G (x h ，x h ，H x ) Representing historical sample x h A Gaussian distance therebetween;
sample x and history sample x h Gaussian distance G (x, x) h ，H x ) The method comprises the following steps:
Figure FDA0004158601410000022
step 3.3, fusing the two-channel samples according to the modified cosine similarity of the two-channel samples, wherein the fused expression is as follows:
Figure FDA0004158601410000023
wherein the method comprises the steps of，x df The sample after the fusion is represented as such,
Figure FDA0004158601410000024
representing sample x i Cosine angle of (2);
in step 5, the initial detection result obtained in step 4 is fused by using a decision-level fusion method, which comprises the following steps:
step 5.1, respectively calculating the precision of different samples and the accuracy of different categories in the initial detection result:
Figure FDA0004158601410000025
/>
Figure FDA0004158601410000026
wherein P is ij Representing the precision of the jth class in the ith sample, A i Representing the accuracy of the ith sample, TP ij 、TN ij 、FP ij And FN ij The numbers of the true positive samples, the true negative samples, the false positive samples and the false negative samples are respectively represented;
step 5.2, calculating the sample credibility:
Figure FDA0004158601410000031
wherein, CP ij Indicating the credibility of the jth category in the ith sample; pr (Pr) ij Representing the precision of the jth category in the ith sample in the initial detection result;
step 5.3, calculating an initial decision stage fusion result:
Figure FDA0004158601410000032
wherein the FPT j Representing the beginning of the jth categoryFusing results at an initial decision stage, wherein m represents the number of initial detection results;
step 5.4, verifying an initial decision stage fusion result; when the probability of the initial decision-stage fusion result exceeds a set value, the initial decision-stage fusion result is a final decision-stage fusion result; otherwise, removing the initial detection result with the lowest accuracy, and carrying out the steps 5.1-5.3 again, and verifying until the requirements are met.
2. The method for detecting crack failure of a blade by sound and vibration fusion according to claim 1, wherein the set distance between the two channels of sound signals and the air inlet and the air outlet is 0.2-1.0 m.
3. The method for detecting crack failure of a blade by sound and vibration fusion according to claim 1, wherein in the step 2, the normalization process is a 0-1 normalization process, and the expression of the 0-1 normalization process is as follows:
Figure FDA0004158601410000041
wherein x is s (t) represents a normalized signal of 0-1, x (t) represents an original signal, and min (x) and max (x) represent a minimum value and a maximum value of the original signal, respectively.
4. The method for detecting blade crack faults by sound and vibration fusion according to claim 1, wherein in step 5.4, the set value of the initial decision stage fusion result is verified to be 50% -70%.
5. The method for detecting blade crack failure by sound and vibration fusion according to claim 1, wherein in the step 4, the neural network is a one-dimensional convolutional neural network, and comprises 2 convolutional layers, 2 pooling layers, 1 full-connection layer and 1 Softmax layer.
6. The blade crack fault detection method for sound and vibration fusion according to claim 5, wherein the size of a convolution layer filter of a one-dimensional convolution neural network is 6*1, the step size of a convolution layer is 1, the number of channels of a first convolution layer is 4, the number of channels of a second convolution layer is 8, the size of a pooling layer filter is 6*1, the step size of a pooling layer is 6, the number of channels of the first pooling layer is 4, the number of channels of the second pooling layer is 8, and the output number of softmax layers is 3.
7. A blade crack fault detection method based on the sound-vibration fusion of any one of claims 1-6 for use in a compressor for crack fault identification of a compressor blade.
CN202210001190.3A 2022-01-04 2022-01-04 Acoustic-vibration-fused blade crack fault detection method and application Active CN114509158B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210001190.3A CN114509158B (en) 2022-01-04 2022-01-04 Acoustic-vibration-fused blade crack fault detection method and application

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210001190.3A CN114509158B (en) 2022-01-04 2022-01-04 Acoustic-vibration-fused blade crack fault detection method and application

Publications (2)

Publication Number Publication Date
CN114509158A CN114509158A (en) 2022-05-17
CN114509158B true CN114509158B (en) 2023-05-30

Family

ID=81549898

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210001190.3A Active CN114509158B (en) 2022-01-04 2022-01-04 Acoustic-vibration-fused blade crack fault detection method and application

Country Status (1)

Country Link
CN (1) CN114509158B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115166032B (en) * 2022-05-23 2024-04-19 东南大学 Device and method for detecting cracks of fan blades

Family Cites Families (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS55162698A (en) * 1979-06-06 1980-12-18 Matsushita Electric Ind Co Ltd Acoustic diaphragm plate
FR2815123B1 (en) * 2000-10-10 2003-02-07 Snecma Moteurs ACOUSTIC CONTROL OF MONOBLOCK BLADES
JP2004317408A (en) * 2003-04-18 2004-11-11 Calsonic Kansei Corp System and method for surveying vibration source of rotor
CN107016404A (en) * 2017-02-24 2017-08-04 沈阳工业大学 Wind power generating set failure prediction method based on D S evidence fusions
US10495693B2 (en) * 2017-06-01 2019-12-03 General Electric Company Wind turbine fault detection using acoustic, vibration, and electrical signals
US10916259B2 (en) * 2019-01-06 2021-02-09 3D Signals Ltd. Extracting overall equipment effectiveness by analysis of a vibro-acoustic signal
CN110131109A (en) * 2019-04-25 2019-08-16 浙江大学 A kind of pneumatic equipment bladess unbalance detection based on convolutional neural networks
CN112326210A (en) * 2019-07-17 2021-02-05 华北电力大学（保定） Large motor fault diagnosis method combining sound vibration signals with 1D-CNN
CN110376522B (en) * 2019-09-03 2021-07-16 宁夏西北骏马电机制造股份有限公司 Motor fault diagnosis method of data fusion deep learning network
CN111239249A (en) * 2020-03-18 2020-06-05 北京工业大学 Ventilator blade crack fault diagnosis method based on Hilbert-Huang transform
CN111612030A (en) * 2020-03-30 2020-09-01 华电电力科学研究院有限公司 Wind turbine generator blade surface fault identification and classification method based on deep learning
CN111678698B (en) * 2020-06-17 2022-03-04 沈阳建筑大学 Rolling bearing fault detection method based on sound and vibration signal fusion
CN112255538B (en) * 2020-10-14 2024-04-23 国网青海省电力公司海西供电公司 Multi-feature optimization fusion fault diagnosis method for high-voltage circuit breaker
CN112393800A (en) * 2020-10-28 2021-02-23 浙江工业大学 Monitoring device that sound vibration signal is united
CN112684012A (en) * 2020-12-02 2021-04-20 青岛科技大学 Equipment key force-bearing structural part fault diagnosis method based on multi-parameter information fusion
CN113033309B (en) * 2021-02-25 2023-12-19 北京化工大学 Fault diagnosis method based on signal downsampling and one-dimensional convolutional neural network
CN113255777B (en) * 2021-05-28 2023-01-31 郑州轻工业大学 Equipment fault early warning method and system based on multi-mode sensitive feature selection fusion
CN113469060A (en) * 2021-07-02 2021-10-01 浙大城市学院 Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method
CN113469281B (en) * 2021-07-22 2023-11-24 西北工业大学 Industrial gearbox multisource information fusion fault diagnosis method
CN113533962B (en) * 2021-07-29 2022-08-12 上海交通大学 Induction motor health diagnosis system based on decision fusion of multiple physical signal sensors

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
Damage detection in an operating Vestas V27 wind turbine blade by use of outlier analysis;Martin D. Ulriksen;2015 IEEE Workshop on Environmental, Energy, and Structural Monitoring Systems (EESMS) Proceedings;全文 *
Theoretical and experimental study of wind turbine drivetrain fault diagnosis by using torsional vibrations and modal estimation;Farid K. Moghadam;Journal of Sound and Vibration;全文 *
基于声振特征融合的机车走行部轴承状态评估方法研究;郭林;铁道技术监督;全文 *
翟栋 ; .基于多技术融合的风电大部件健康状态大数据分析.风能产业（2017年第1期 总第90期）.2017,全文. *
颉潭成 ; .基于信息融合的滚动轴承故障诊断.轴承.2015,全文. *

Also Published As

Publication number Publication date
CN114509158A (en) 2022-05-17

Similar Documents

Publication Publication Date Title
Pandya et al. Fault diagnosis of rolling element bearing with intrinsic mode function of acoustic emission data using APF-KNN
CN113375941B (en) Open-set fault diagnosis method for bearings of high-speed motor train unit
CN115688018B (en) Method for monitoring state and diagnosing faults of bearing under multiple working conditions
CN104897403A (en) Self-adaption fault diagnosis method based on permutation entropy (PE) and manifold-based dynamic time warping (MDTW)
CN115166032B (en) Device and method for detecting cracks of fan blades
CN114509158B (en) Acoustic-vibration-fused blade crack fault detection method and application
CN110259648B (en) Fan blade fault diagnosis method based on optimized K-means clustering
Mechefske et al. Fault detection and diagnosis in low speed rolling element bearings Part II: The use of nearest neighbour classification
Saimurugan et al. A comparative study of sound and vibration signals in detection of rotating machine faults using support vector machine and independent component analysis
Meng et al. Health indicator of bearing constructed by rms-CUMSUM and GRRMD-CUMSUM with multifeatures of envelope spectrum
Khan et al. System design for early fault diagnosis of machines using vibration features
Song et al. Data and decision level fusion-based crack detection for compressor blade using acoustic and vibration signal
CN109612726A (en) A kind of multiple superstage analysis method extracted for vibration signal characteristics
CN113314142B (en) Engine fault detection method, system, storage medium and equipment
CN111707472A (en) Rolling bearing fault diagnosis method based on FFT and full-connection layer-SVM
CN108760268B (en) Step fault diagnosis method for vertical mill operation data based on information entropy
CN110781612A (en) Fault diagnosis method and device for ball screw, computer device and storage medium
CN117076869B (en) Time-frequency domain fusion fault diagnosis method and system for rotary machine
CN110749443B (en) Rolling bearing fault diagnosis method and system based on high-order origin moment
Huo et al. Bearing fault diagnosis using multi-sensor fusion based on weighted DS evidence theory
CN116307938A (en) Health state assessment method for feeding system of machining center
CN114528868A (en) Crack fault detection method for compressor blade
CN109034028A (en) A kind of mechanical equipment fault feature extracting method based on symbolism and TF-IDF
Gelman et al. Novel decision-making technique for damage diagnosis
Zheng et al. Wavelet packet decomposition and neural network based fault diagnosis for elevator excessive vibration

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