CN116168720A - Motor sound abnormality fault diagnosis method, system and storable medium - Google Patents
Motor sound abnormality fault diagnosis method, system and storable medium Download PDFInfo
- Publication number
- CN116168720A CN116168720A CN202310187876.0A CN202310187876A CN116168720A CN 116168720 A CN116168720 A CN 116168720A CN 202310187876 A CN202310187876 A CN 202310187876A CN 116168720 A CN116168720 A CN 116168720A
- Authority
- CN
- China
- Prior art keywords
- random forest
- motor
- matrix
- training
- signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000003745 diagnosis Methods 0.000 title claims abstract description 28
- 230000005856 abnormality Effects 0.000 title claims description 21
- 238000007637 random forest analysis Methods 0.000 claims abstract description 73
- 239000011159 matrix material Substances 0.000 claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 63
- 230000002159 abnormal effect Effects 0.000 claims abstract description 41
- 230000009467 reduction Effects 0.000 claims abstract description 34
- 238000001514 detection method Methods 0.000 claims abstract description 24
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 238000012847 principal component analysis method Methods 0.000 claims abstract description 14
- 238000000926 separation method Methods 0.000 claims abstract description 13
- 238000000605 extraction Methods 0.000 claims abstract description 12
- 239000013598 vector Substances 0.000 claims description 17
- 230000005236 sound signal Effects 0.000 claims description 14
- 230000001186 cumulative effect Effects 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 13
- 238000005070 sampling Methods 0.000 claims description 8
- 230000035945 sensitivity Effects 0.000 claims description 8
- 238000003066 decision tree Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 4
- 238000013100 final test Methods 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 description 7
- 230000009466 transformation Effects 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 238000010183 spectrum analysis Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/45—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of analysis window
-
- 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
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L17/00—Speaker identification or verification techniques
- G10L17/04—Training, enrolment or model building
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention belongs to the technical field of noise detection, and discloses a motor sound abnormal fault diagnosis method, which comprises the following steps: window-separated time domain feature extraction is carried out on the acquired motor noise signals; performing dimension reduction on the extracted features by using a principal component analysis method; generating a random forest neural network training data set according to the feature matrix after dimension reduction; training the random forest network model through a training set, and evaluating the classification performance of the classifier by using the Kappa coefficient; and detecting motor noise by using the trained random forest network model. The time domain feature extraction of window separation is carried out on the motor noise signals, so that the motor noise can be rapidly and effectively identified, and the method is suitable for industrial application.
Description
Technical Field
The invention belongs to the technical field of noise detection, and relates to a motor sound abnormal fault diagnosis method, a motor sound abnormal fault diagnosis system and a storable medium, which are particularly suitable for an intelligent seat back motor of an automobile.
Background
The motor is an important electromechanical device in modern production and life, and the working state of the motor directly influences the stability of the whole system. If the motor fails during operation, very serious consequences such as production stalls, machine damage and significant economic losses can result.
As shown in fig. 1, the existing method for diagnosing abnormal sound faults of the intelligent seat back motor of the automobile is mainly developed from three aspects of frequency spectrum analysis, time-frequency analysis and machine learning modeling of signals. In the aspect of frequency spectrum analysis of motor abnormal sound, a frequency spectrum diagram of a motor audio signal is obtained through Fourier transformation, and then the abnormal sound signal can be distinguished according to corresponding fault characteristic frequency information under various faults, but because motor vibration signals belong to non-stationary signals, the effect of using a common frequency spectrum analysis method such as Fourier transformation needs to be improved, and the Fourier transformation can only carry out frequency spectrum analysis of the whole section of signals and cannot locate the occurrence time of the abnormal sound. Although the time-frequency analysis method can diagnose abnormal sound faults of the motor better than the frequency spectrum analysis method, the time-frequency analysis method has various problems, such as: the short-time Fourier transform has a certain time resolution, but once the window is selected, the window cannot be changed any more, and the window does not have an automatic adjusting function;
the wavelet transformation is used as a new time-frequency analysis method, inherits and develops the concept of short-time Fourier transformation localization, overcomes the defects that the window size does not change along with frequency and the like, and is very suitable for being applied to non-stationary signals, but the wavelet analysis only carries out fine division in a low frequency band and lacks high-frequency signal analysis and the like.
The machine learning and neural network analysis technology is a mainstream motor abnormal sound fault diagnosis method at present, a wavelet transformation is adopted to construct a data set, after the data set is divided, a convolutional neural network and a migration learning method are utilized to diagnose and analyze motor bearing faults, and finally the obtained accuracy is very high, but the machine learning and neural network abnormal sound fault diagnosis method has the defects that a large sample is required to train, the training cost is very high, and the time is long.
Disclosure of Invention
The invention aims to provide a motor sound abnormal fault diagnosis method and system, which are used for extracting time domain features of window separation of motor noise signals, can be used for rapidly and effectively identifying motor noise, and are suitable for industrial application.
The technical solution for realizing the purpose of the invention is as follows:
a motor sound abnormality fault diagnosis method includes the following steps:
s01: window-separated time domain feature extraction is carried out on the acquired motor noise signals;
s02: performing dimension reduction on the extracted features by using a principal component analysis method;
s03: generating a random forest neural network training data set according to the feature matrix after dimension reduction;
s04: training the random forest network model through a training set, and evaluating the classification performance of the classifier by using the Kappa coefficient;
s05: and detecting motor noise by using the trained random forest network model.
In a preferred embodiment, the time domain feature includes: peak-to-peak, average, mean square, standard deviation, effective, peak factor, pulse factor, waveform factor, margin factor, skewness factor, and kurtosis factor.
In a preferred technical solution, the method for dimension reduction of the extracted features in step S02 by using a principal component analysis method includes:
s21: calculating the characteristic equation det (lambda i E- Σx) =0, resulting in a eigenvalue λ i Wherein E represents an identity matrix, det is a square matrix function for determining a determinant of a square matrix, c x Is the characteristic matrix of the original data, sigma x Is covariance matrix;
s22: solving (lambda) i E-∑x)ω i =0, resulting in a corresponding eigenvalue λ i Feature vector omega of (2) i The feature vector omega i Constructed as a W matrix;
s23: by calculating c y ＝c x W T To obtain a data representation of the data in the new feature space;
s24: the characteristic values are arranged from big to small to obtain first m principal components as y 1 ,y 2 ,Λ,y m Then the cumulative variance contribution is found to be:
where k=1, 2,..n, n is the number of feature values before dimension reduction, k=1, 2,..m;
s25: after the cumulative variance contribution rate is calculated, when phi (m) is larger than a set value, the first m principal components are obtained to represent information contained in the original signal.
In a preferred technical solution, the training method of the random forest network model in step S04 includes:
s41: setting N samples in an original training set, and randomly sampling the N samples by using a bootstrap method to form the training set;
s42: setting D characteristics of a sample, randomly extracting D characteristics at each node of each tree, wherein D is less than D, selecting a variable with the most classification capability from the D characteristics, and determining a variable classification threshold by checking each classification point;
s43: constructing a random forest network model by using the extracted features;
s44: and inputting data, judging and classifying the new data set by the random forest network model, wherein the classification result is determined according to the number of votes of the tree classifier.
In a preferred embodiment, the method for evaluating the classification performance of the classifier in step S04 by using the kappa coefficient includes:
s041: marking the rest samples divided into test sets as I and inputting the I into a trained random forest model;
s042: each decision tree in the random forest judges and votes each sample of the test set, and finally, the result with more votes is output as the final test result of the random forest through the principle of 'minority compliance with majority';
s043: for an input sample I and an output result O obtained by testing, evaluating the classification performance of the classifier by using specificity, sensitivity, accuracy and kappa coefficient;
the specificity Sp represents the proportion of the original signal that belongs to the abnormal signal and is correctly recognized:
sensitivity Se represents the proportion of the original signal that belongs to the normal signal and is correctly identified:
accuracy Ac refers to the correct proportion of classifier predictions:
wherein TP, TN, FP, FN is a parameter of the confusion matrix, which represents true positive, true negative, false positive, false negative, respectively, namely: TP is the number of normal signals classified as the recognition result in the case that the real signal is the normal signal; TN is the number of abnormal sound signals identified in the case where the real signal is an abnormal sound signal; FP is the number of normal signals classified as the recognition result in the case where the true signal is the abnormal signal; FN is the number of signals identified as normal in the case where the true signal is an abnormal signal.
In a preferred technical solution, the method for detecting motor noise in step S05 by using a trained random forest network model includes:
s51: obtaining a feature matrix of the motor noise signal after dimension reduction, setting the number of frames of the matrix after window separation as T, inputting the matrix into a trained random forest neural network model, obtaining output values of T random forest neural network models, and marking the output values as r;
s52: when r is greater than or equal to a set threshold, the noise detection mark of the motor to be detected is 1, namely noise exists and the motor detection result is unqualified; and when r is smaller than a set threshold value, the motor noise detection mark to be detected is set to be 0, namely no noise exists, and the motor detection result is qualified.
The invention also discloses a motor sound abnormal fault diagnosis system, which comprises:
the feature extraction module is used for extracting time domain features of window separation of the acquired motor noise signals;
the dimension reduction module is used for reducing dimension of the extracted features by using a principal component analysis method;
the training data set generation module is used for generating a random forest neural network training data set according to the feature matrix after dimension reduction;
the training evaluation module is used for training the random forest network model through the training set and evaluating the classification performance of the classifier by using the kappa coefficient;
and the detection module is used for detecting motor noise by using the trained random forest network model.
In a preferred technical solution, the method for dimension reduction of the extracted features by using the principal component analysis method in the dimension reduction module includes:
s21: calculating the characteristic equation det (lambda i E- Σx) =0, resulting in a eigenvalue λ i Wherein E represents an identity matrix, det is a square matrix function for determining a determinant of a square matrix, c x Is the characteristic matrix of the original data, sigma x Is covariance matrix;
s22: solving (lambda) i E-∑x)ω i =0, resulting in a corresponding eigenvalue λ i Feature vector omega of (2) i The feature vector omega i Constructed as a W matrix;
s23: by calculating c y ＝c x W T To obtain data in newIs represented by data in a feature space of (a);
s24: the characteristic values are arranged from big to small to obtain first m principal components as y 1 ,y 2 ,Λ,y m Then the cumulative variance contribution is found to be:
where i=1, 2,..n, n is the number of feature values before dimension reduction, k=1, 2,..m;
s25: after the cumulative variance contribution rate is calculated, when phi (m) is larger than a set value, the first m principal components are obtained to represent information contained in the original signal.
In a preferred technical scheme, the training method of the random forest network model in the training evaluation module comprises the following steps:
s41: setting N samples in an original training set, and randomly sampling the N samples by using a bootstrap method to form the training set;
s42: setting D characteristics of a sample, randomly extracting D characteristics at each node of each tree, wherein D is less than D, selecting a variable with the most classification capability from the D characteristics, and determining a variable classification threshold by checking each classification point;
s43: constructing a random forest network model by using the extracted features;
s44: and inputting data, judging and classifying the new data set by the random forest network model, wherein the classification result is determined according to the number of votes of the tree classifier.
The invention also discloses a computer storage medium, on which a computer program is stored, which when executed implements the motor sound abnormality fault diagnosis method described above.
Compared with the prior art, the invention has the remarkable advantages that:
according to the invention, window separation time domain feature extraction is carried out on the motor noise signals, so that the motor noise can be rapidly and effectively identified. The training cost is not very high, a large number of samples are not needed for training, and the method is suitable for industrial application.
Drawings
FIG. 1 is a diagram of a prior art method for diagnosing abnormal sound faults of a motor of an intelligent seat back of a vehicle;
FIG. 2 is a flow chart of a motor sound abnormality fault diagnosis method of the preferred embodiment;
FIG. 3 is a schematic block diagram of a motor sound abnormality fault diagnosis system of the preferred embodiment;
fig. 4 is a flowchart showing the operation of the motor sound abnormality fault diagnosis system of the preferred embodiment.
Detailed Description
The principle of the invention is as follows: the time domain feature extraction of window separation is carried out on the motor noise signals, so that the motor noise can be rapidly and effectively identified. The training cost is not very high, a large number of samples are not needed for training, and the method is suitable for industrial application.
Example 1:
as shown in fig. 2, a motor sound abnormality fault diagnosis method includes the steps of:
s01: window-separated time domain feature extraction is carried out on the acquired motor noise signals;
s02: performing dimension reduction on the extracted features by using a principal component analysis method;
s03: generating a random forest neural network training data set according to the feature matrix after dimension reduction;
s04: training the random forest network model through a training set, and evaluating the classification performance of the classifier by using the Kappa coefficient;
s05: and detecting motor noise by using the trained random forest network model.
In one embodiment, the time domain features include: peak-to-peak, average, mean square, standard deviation, effective, peak factor, pulse factor, waveform factor, margin factor, skewness factor, and kurtosis factor.
In one embodiment, the method for dimension reduction of the extracted features in step S02 using the principal component analysis method includes:
s21: calculating the characteristic equation det (lambda i E-∑x)＝0，Obtaining a characteristic value lambda i Wherein E represents an identity matrix, det is a square matrix function for determining a determinant of a square matrix, c x Is the characteristic matrix of the original data, sigma x Is covariance matrix;
s22: solving (lambda) i E-∑x)ω i =0, resulting in a corresponding eigenvalue λ i Feature vector omega of (2) i The feature vector omega i Constructed as a W matrix;
s23: by calculating c y ＝c x W T To obtain a data representation of the data in the new feature space;
s24: the characteristic values are arranged from big to small to obtain first m principal components as y 1 ,y 2 ,Λ,y m Then the cumulative variance contribution is found to be:
where i=1, 2,..n, n is the number of feature values before dimension reduction, k=1, 2,..m;
s25: after the cumulative variance contribution rate is calculated, when phi (m) is larger than a set value, the first m principal components are obtained to represent information contained in the original signal.
In one embodiment, the training method of the random forest network model in step S04 includes:
s41: setting N samples in an original training set, and randomly sampling the N samples by using a bootstrap method to form the training set;
s42: setting D characteristics of a sample, randomly extracting D characteristics at each node of each tree, wherein D is less than D, selecting a variable with the most classification capability from the D characteristics, and determining a variable classification threshold by checking each classification point;
s43: constructing a random forest network model by using the extracted features;
s44: and inputting data, judging and classifying the new data set by the random forest network model, wherein the classification result is determined according to the number of votes of the tree classifier.
In one embodiment, the method for evaluating the classification performance of the classifier by using the kappa coefficient in step S04 includes:
s041: marking the rest samples divided into test sets as I and inputting the I into a trained random forest model;
s042: each decision tree in the random forest judges and votes each sample of the test set, and finally, the result with more votes is output as the final test result of the random forest through the principle of 'minority compliance with majority';
s043: for an input sample I and an output result O obtained by testing, evaluating the classification performance of the classifier by using specificity, sensitivity, accuracy and kappa coefficient;
the specificity Sp represents the proportion of the original signal that belongs to the abnormal signal and is correctly recognized:
sensitivity Se represents the proportion of the original signal that belongs to the normal signal and is correctly identified:
accuracy Ac refers to the correct proportion of classifier predictions:
wherein TP, TN, FP, FN is a parameter of the confusion matrix, which represents true positive, true negative, false positive, false negative, respectively, namely: TP is the number of normal signals classified as the recognition result in the case that the real signal is the normal signal; TN is the number of abnormal sound signals identified in the case where the real signal is an abnormal sound signal; FP is the number of normal signals classified as the recognition result in the case where the true signal is the abnormal signal; FN is the number of signals identified as normal in the case where the true signal is an abnormal signal.
In one embodiment, the method for detecting motor noise in step S05 by using the trained random forest network model includes:
s51: obtaining a feature matrix of the motor noise signal after dimension reduction, setting the number of frames of the matrix after window separation as T, inputting the matrix into a trained random forest neural network model, obtaining output values of T random forest neural network models, and marking the output values as r;
s52: when r is greater than or equal to a set threshold, the noise detection mark of the motor to be detected is 1, namely noise exists and the motor detection result is unqualified; and when r is smaller than a set threshold value, the motor noise detection mark to be detected is set to be 0, namely no noise exists, and the motor detection result is qualified.
In another embodiment, a computer storage medium has a computer program stored thereon, which when executed implements the motor sound abnormality fault diagnosis method described above.
In still another embodiment, as shown in fig. 3, a motor sound abnormality fault diagnosis system includes:
the feature extraction module 10 performs window-separated time domain feature extraction on the acquired motor noise signals;
the dimension reduction module 20 performs dimension reduction on the extracted features by using a principal component analysis method;
the training data set generating module 30 generates a random forest neural network training data set according to the feature matrix after dimension reduction;
training the random forest network model through a training set by a training evaluation module 40, and evaluating classification performance of the classifier by using the kappa coefficient;
the detection module 50 detects motor noise using a trained random forest network model.
Specifically, the following describes a working flow of a motor sound abnormality fault diagnosis system by taking a preferred embodiment as an example, and the motor takes an automobile intelligent seat back motor as an example, specifically as follows:
s1: and acquiring a motor vibration signal (noise signal) of the intelligent automobile seat backrest.
Vibration signals of the intelligent seat back motor of the automobile are collected, the sampling frequency is 20000Hz, and the collected data are stored in a TDMS format. The data acquisition environment and the conditions are consistent. Through signal acquisition, 92 groups of vibration signals of the intelligent seat back motor of the automobile are acquired, wherein the normal signals are 68 groups, and the signals with abnormal noise are 24 groups.
S2: and extracting window separated characteristics of the vibration signal, (time domain characteristic value and time domain characteristic value).
And extracting the characteristics of window separation of the acquired signals, wherein the length of the window is set to be one fifth of the sampling frequency. The use of window separation can increase the number of samples interchangeably. And 11 features are extracted for each window-separated inter-cell, which are respectively: peak-to-peak, average, mean square, standard deviation, effective, peak factor, pulse factor, waveform factor, margin factor, skewness factor, and kurtosis factor. Through the above-mentioned feature extraction of window separation, feature vector space features is finally obtained, the size is 11× 14238, the number of rows represents the features, and the number of columns represents the number of feature samples. The manual method is adopted for calibration, the normal section is 0, the abnormal sound section is 1, the label vector Classes is formed, the size is 1 multiplied by 14238, and 14238 is the label number.
S3: and (5) reducing the dimension of the extracted features by using a principal component analysis method.
The principal component analysis method is used for reducing the feature dimension of the extracted 11 motor abnormal sound signal features. Under the condition that the classification accuracy of the normal vibration signals and the abnormal sound signals of the motor can be distinguished by the reserved characteristics, the dimension of network input is reduced, the data redundancy is reduced, and the training efficiency and speed of the model are improved. The specific method comprises the following steps:
(1) Calculating a feature matrix c of the original data x 。
(2) Calculating covariance matrix Σ x 。
(3) Calculating the characteristic equation det (lambda i E- Σx) =0, thereby obtaining a characteristic value λ i Where E represents an identity matrix. det is a square matrix function for solving a determinant of a square matrix.
(4) Solving (lambda) i E-∑x)ω i =0, resulting in a corresponding eigenvalue λ i Feature vector omega of (2) i 。
(5) The feature vector omega i Constructed as a W matrix.
(6) By calculating c y ＝c x W T To obtain a data representation of the data in the new feature space.
(7) And arranging the characteristic values from large to small, wherein the large characteristic values indicate that the contribution degree of the corresponding characteristic vectors is large in signal reconstruction, and the small characteristic values indicate that the contribution degree of the corresponding characteristic vectors is small in signal reconstruction. Thus, the first m principal components of the characteristic value ranging from big to small are set as y in the orthogonal space 1 ,y 2 ,Λ,y m Then its cumulative variance contribution is found as:
where i=1, 2,..n, n is the number of feature values before dimension reduction, k=1, 2,..m;
after calculating the cumulative variance contribution ratio, the cumulative variance contribution ratio of several principal components is required to have a sufficiently large specific gravity, and is set to 95% here. I.e. when
When the matrix after the dimension reduction is displayed, the information contained in the matrix accounts for 95% of the original information. That is, only the first m principal components are needed to represent the information contained in the original signal. The size of the finally obtained dimension-reduced matrix is 6× 14238. The matrix needs to be normalized before it is input into the random forest neural network.
S4: and generating a random forest neural network training data set according to the feature matrix after the dimension reduction.
According to step S1, 92 seat motor noise signals are collected, wherein the signals comprise 68 normal motors and 24 motors with abnormal noise, and 92 sections of audios are obtained. According to steps S2 and S3, 11 eigenvalues and principal component analysis are respectively extracted from the 100 sections of audio, and a matrix of 6 rows 14238 columns can be obtained and marked as Input. Each column represents 6 feature data sets for a certain motor at a certain moment.
When the normal motor is marked as 0 and the motor with abnormal noise is marked as 1, a row vector with the dimension 14238 corresponding to Input can be obtained and is marked as Output.
The Input and Output data are in one-to-one correspondence. If the first column of Input corresponds to a normal motor, the first column of Output should have a value of 0; and vice versa.
Input is an Input matrix of the training data set of the random forest neural network, and Output is an Output matrix of the training data set of the random forest neural network.
S5: and training the random forest network model through the training set.
Random forests refer to a classifier that uses multiple decision trees to train and predict samples. And using the normalized feature matrix after dimension reduction as a random forest neural network training data set, randomly selecting 80% of samples as a training set, and the remaining 20% of samples as a test set, and performing training of random forest neural network classifier classification.
The specific flow is as follows:
(1) Setting N samples in an original training set, and randomly sampling the N samples by using a bootstrap method to form the training set;
(2) D characteristics of a sample are set, D (D is less than D) characteristics are randomly extracted at each node of each tree, one variable with the most classification capacity is selected from the D characteristics, and the threshold value of variable classification is determined by checking each classification point;
(3) Constructing a random forest model by using the extracted features;
(4) And inputting data, judging and classifying the new data set by the random forest classifier, wherein the classification result is determined according to the number of votes of the tree classifier.
The number of decision trees is set to 30 on the parameters. After training, a model is obtained, denoted model_RF.
S6: and detecting abnormal noise of the seat motor by using the trained random forest network model and evaluating the classification performance of the classifier by using the Kappa coefficient.
According to steps S1 to S5, a trained model_RF is obtained, and the remaining samples divided into test sets are marked as I and input into a trained random forest model. Each decision tree in the random forest judges and votes each sample of the test set, and finally outputs the result with a large number of votes as the final test result of the random forest through the principle of 'minority compliance with majority'. The output result obtained by the test is marked as O. The classifier classification performance is then evaluated for the input sample I and the output result O using specificity (Sp), sensitivity (Se), accuracy (Ac) and Kappa Coefficient (KC). In the process of identifying the vibration signal of the motor of the intelligent seat back of the automobile, the specificity (Sp) represents the proportion of the original signal which belongs to the abnormal sound signal and is correctly identified:
sensitivity (Se) then represents the proportion of the original signal that belongs to the normal signal and is correctly identified:
accuracy (Ac) refers to the proportion of classifier predictions that are correct. The expression is the ratio of the number of predicted correct samples to the total samples output by the classifier:
wherein TP, TN, FP, FN is a parameter of the confusion matrix, which represents true positive, true negative, false positive, and false negative, respectively. Namely: TP is the number of normal signals classified as the recognition result in the case that the real signal is the normal signal; TN is the number of abnormal sound signals identified in the case where the real signal is an abnormal sound signal; FP is the number of normal signals classified as the recognition result in the case where the true signal is the abnormal signal; FN is the number of signals identified as normal in the case where the true signal is an abnormal signal.
Table 1 shows the final kappa coefficient evaluation results of the random forest model trained in an embodiment of the present application:
TABLE 1
S7: automobile intelligent seat backrest motor noise detection.
And (3) obtaining a feature matrix of the vibration data of the intelligent seat back motor of the automobile to be detected after dimension reduction according to the steps S1 to S4, and setting the number of frames of the matrix after window separation as T. The matrix is input into a trained random forest neural network, and output values of T random forest neural networks can be obtained and recorded as r. When r is more than or equal to 5, the noise detection mark of the seat motor to be detected is 1, namely noise exists and the motor detection result is unqualified. When r is less than or equal to 5, the noise detection mark of the seat motor to be detected is set to be 0, namely no noise exists, and the motor detection result is qualified.
Generally, steps S3 to S6 are used in the training phase, and only need to be performed once, so as to establish a random forest neural network detection model of the abnormal sound signal of the intelligent seat back motor of the automobile. These steps are not required during the detection phase.
Particularly, when a new unqualified intelligent automobile seat back motor is found to exist and the vibration signal of the motor is not similar to the trained vibration signal, the intelligent automobile seat back motor is added into an original motor training sample, and the random forest neural network is trained again.
The foregoing examples are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the foregoing examples, and any other changes, modifications, substitutions, combinations, and simplifications that do not depart from the spirit and principles of the present invention should be made therein and are intended to be equivalent substitutes within the scope of the present invention.
Claims (10)
1. A motor sound abnormality fault diagnosis method, characterized by comprising the steps of:
s01: window-separated time domain feature extraction is carried out on the acquired motor noise signals;
s02: performing dimension reduction on the extracted features by using a principal component analysis method;
s03: generating a random forest neural network training data set according to the feature matrix after dimension reduction;
s04: training the random forest network model through a training set, and evaluating the classification performance of the classifier by using the Kappa coefficient;
s05: and detecting motor noise by using the trained random forest network model.
2. The motor sound abnormality fault diagnosis method according to claim 1, characterized in that the time domain features include: peak-to-peak, average, mean square, standard deviation, effective, peak factor, pulse factor, waveform factor, margin factor, skewness factor, and kurtosis factor.
3. The motor sound abnormality fault diagnosis method according to claim 1, characterized in that the method of reducing the dimension of the extracted features using the principal component analysis method in step S02 includes:
s21: calculating the characteristic equation det (lambda i E- Σx) =0, resulting in a eigenvalue λ i Wherein E represents an identity matrix, det is a square matrix function for determining a determinant of a square matrix, c x Is the characteristic matrix of the original data, sigma x Is covariance matrix;
s22: solving (lambda) i E-∑x)ω i =0, resulting in a corresponding eigenvalue λ i Feature vector omega of (2) i The feature vector omega i Constructed as a W matrix;
s23: through meterCalculation c y ＝c x W T To obtain a data representation of the data in the new feature space;
s24: the characteristic values are arranged from big to small to obtain first m principal components as y 1 ,y 2 ,Λ,y m Then the cumulative variance contribution is found to be:
where i=1, 2,..n, n is the number of feature values before dimension reduction, k=1, 2,..m;
s25: after the cumulative variance contribution rate is calculated, when phi (m) is larger than a set value, the first m principal components are obtained to represent information contained in the original signal.
4. The motor sound abnormality fault diagnosis method according to claim 1, wherein the training method of the random forest network model in step S04 includes:
s41: setting N samples in an original training set, and randomly sampling the N samples by using a bootstrap method to form the training set;
s42: setting D characteristics of a sample, randomly extracting D characteristics at each node of each tree, wherein D is less than D, selecting a variable with the most classification capability from the D characteristics, and determining a variable classification threshold by checking each classification point;
s43: constructing a random forest network model by using the extracted features;
s44: and inputting data, judging and classifying the new data set by the random forest network model, wherein the classification result is determined according to the number of votes of the tree classifier.
5. The motor sound abnormality fault diagnosis method according to claim 1, characterized in that the method of evaluating the classification performance of the classifier using the kappa coefficient in step S04 includes:
s041: marking the rest samples divided into test sets as I and inputting the I into a trained random forest model;
s042: each decision tree in the random forest judges and votes each sample of the test set, and finally, the result with more votes is output as the final test result of the random forest through the principle of 'minority compliance with majority';
s043: for an input sample I and an output result O obtained by testing, evaluating the classification performance of the classifier by using specificity, sensitivity, accuracy and kappa coefficient;
the specificity Sp represents the proportion of the original signal that belongs to the abnormal signal and is correctly recognized:
sensitivity Se represents the proportion of the original signal that belongs to the normal signal and is correctly identified:
accuracy Ac refers to the correct proportion of classifier predictions:
wherein TP, TN, FP, FN is a parameter of the confusion matrix, which represents true positive, true negative, false positive, false negative, respectively, namely: TP is the number of normal signals classified as the recognition result in the case that the real signal is the normal signal; TN is the number of abnormal sound signals identified in the case where the real signal is an abnormal sound signal; FP is the number of normal signals classified as the recognition result in the case where the true signal is the abnormal signal; FN is the number of signals identified as normal in the case where the true signal is an abnormal signal.
6. The motor sound abnormality fault diagnosis method according to claim 1, wherein the method of detecting motor noise using the trained random forest network model in step S05 includes:
s51: obtaining a feature matrix of the motor noise signal after dimension reduction, setting the number of frames of the matrix after window separation as T, inputting the matrix into a trained random forest neural network model, obtaining output values of T random forest neural network models, and marking the output values as r;
s52: when r is greater than or equal to a set threshold, the noise detection mark of the motor to be detected is 1, namely noise exists and the motor detection result is unqualified; and when r is smaller than a set threshold value, the motor noise detection mark to be detected is set to be 0, namely no noise exists, and the motor detection result is qualified.
7. A motor sound abnormality fault diagnosis system, characterized by comprising:
the feature extraction module is used for extracting time domain features of window separation of the acquired motor noise signals;
the dimension reduction module is used for reducing dimension of the extracted features by using a principal component analysis method;
the training data set generation module is used for generating a random forest neural network training data set according to the feature matrix after dimension reduction;
the training evaluation module is used for training the random forest network model through the training set and evaluating the classification performance of the classifier by using the kappa coefficient;
and the detection module is used for detecting motor noise by using the trained random forest network model.
8. The motor sound abnormality fault diagnosis system according to claim 7, wherein the method of dimension reduction of the extracted features using a principal component analysis method in the dimension reduction module includes:
s21: calculating the characteristic equation det (lambda i E- Σx) =0, resulting in a eigenvalue λ i Wherein E represents an identity matrix, det is a square matrix function for determining a determinant of a square matrix, c x Is the characteristic matrix of the original data, sigma x Is covariance matrix;
s22: solving (lambda) i E-∑x)ω i =0, resulting in a corresponding eigenvalue λ i Feature vector omega of (2) i The feature vector omega i Constructed as a W matrix;
s23: by calculating c y ＝c x W T To obtain a data representation of the data in the new feature space;
s24: the characteristic values are arranged from big to small to obtain first m principal components as y 1 ,y 2 ,Λ,y m Then the cumulative variance contribution is found to be:
where i=1, 2,..n, n is the number of feature values before dimension reduction, k=1, 2,..m;
s25: after the cumulative variance contribution rate is calculated, when phi (m) is larger than a set value, the first m principal components are obtained to represent information contained in the original signal.
9. The motor sound abnormality fault diagnosis system according to claim 7, wherein the training method of the random forest network model in the training evaluation module includes:
s41: setting N samples in an original training set, and randomly sampling the N samples by using a bootstrap method to form the training set;
s42: setting D characteristics of a sample, randomly extracting D characteristics at each node of each tree, wherein D is less than D, selecting a variable with the most classification capability from the D characteristics, and determining a variable classification threshold by checking each classification point;
s43: constructing a random forest network model by using the extracted features;
s44: and inputting data, judging and classifying the new data set by the random forest network model, wherein the classification result is determined according to the number of votes of the tree classifier.
10. A computer storage medium having a computer program stored thereon, characterized in that the computer program, when executed, implements the motor sound abnormality fault diagnosis method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310187876.0A CN116168720A (en) | 2023-02-28 | 2023-02-28 | Motor sound abnormality fault diagnosis method, system and storable medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310187876.0A CN116168720A (en) | 2023-02-28 | 2023-02-28 | Motor sound abnormality fault diagnosis method, system and storable medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116168720A true CN116168720A (en) | 2023-05-26 |
Family
ID=86413110
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310187876.0A Pending CN116168720A (en) | 2023-02-28 | 2023-02-28 | Motor sound abnormality fault diagnosis method, system and storable medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116168720A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117292709A (en) * | 2023-11-23 | 2023-12-26 | 中瑞恒（北京）科技有限公司 | Abnormal audio identification method and device for heating ventilation machine room |
| CN117607683A (en) * | 2023-11-24 | 2024-02-27 | 河南华东工控技术有限公司 | Intelligent abnormality detection method for motor |
-
2023
- 2023-02-28 CN CN202310187876.0A patent/CN116168720A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117292709A (en) * | 2023-11-23 | 2023-12-26 | 中瑞恒（北京）科技有限公司 | Abnormal audio identification method and device for heating ventilation machine room |
| CN117292709B (en) * | 2023-11-23 | 2024-02-09 | 中瑞恒（北京）科技有限公司 | Abnormal audio identification method and device for heating ventilation machine room |
| CN117607683A (en) * | 2023-11-24 | 2024-02-27 | 河南华东工控技术有限公司 | Intelligent abnormality detection method for motor |
| CN117607683B (en) * | 2023-11-24 | 2024-05-10 | 河南华东工控技术有限公司 | Intelligent abnormality detection method for motor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110132598B (en) | Fault noise diagnosis algorithm for rolling bearing of rotating equipment | |
| CN108760266B (en) | The virtual degeneration index building method of mechanical key component based on learning distance metric | |
| Yang et al. | Bearing fault automatic classification based on deep learning | |
| CN110110768B (en) | Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers | |
| CN116168720A (en) | Motor sound abnormality fault diagnosis method, system and storable medium | |
| CN104849050B (en) | A kind of Fault Diagnosis of Roller Bearings based on compound multiple dimensioned arrangement entropy | |
| CN107144430B (en) | A kind of Method for Bearing Fault Diagnosis based on incremental learning | |
| CN103868690B (en) | Rolling bearing state automatic early warning method based on extraction and selection of multiple characteristics | |
| CN113111820B (en) | Rotary part fault diagnosis method and device based on improved CNN and relation module | |
| CN109186973A (en) | A kind of mechanical failure diagnostic method of unsupervised deep learning network | |
| CN104596780B (en) | Diagnosis method for sensor faults of motor train unit braking system | |
| CN111964909A (en) | Rolling bearing operation state detection method, fault diagnosis method and system | |
| CN114742115B (en) | Method for constructing fault diagnosis model of rolling bearing and diagnosis method | |
| CN110991471A (en) | Fault diagnosis method for high-speed train traction system | |
| CN107655690A (en) | Motor bearings method for diagnosing faults based on artificial neural network | |
| CN112146882A (en) | Bearing fault diagnosis method based on transfer learning vibration signal image recognition | |
| CN112465124A (en) | Twin depth space-time neural network model acquisition/fault diagnosis method and device | |
| CN114705432B (en) | Method and system for evaluating health state of explosion-proof motor bearing | |
| CN117628005A (en) | Signal-fused hydraulic motor fault diagnosis method and system | |
| Sadoughi et al. | A deep learning approach for failure prognostics of rolling element bearings | |
| CN115392782A (en) | Method and system for monitoring and diagnosing health state of process system of nuclear power plant | |
| CN116595465A (en) | High-dimensional sparse data outlier detection method and system based on self-encoder and data enhancement | |
| CN114048787A (en) | Method and system for intelligently diagnosing bearing fault in real time based on Attention CNN model | |
| CN112069621B (en) | Method for predicting residual service life of rolling bearing based on linear reliability index | |
| CN112560674A (en) | Method and system for detecting quality of sound signal |
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 |