CN113405825B - Belt conveyor fault diagnosis method based on sound signals - Google Patents
Belt conveyor fault diagnosis method based on sound signals Download PDFInfo
- Publication number
- CN113405825B CN113405825B CN202110651133.5A CN202110651133A CN113405825B CN 113405825 B CN113405825 B CN 113405825B CN 202110651133 A CN202110651133 A CN 202110651133A CN 113405825 B CN113405825 B CN 113405825B
- Authority
- CN
- China
- Prior art keywords
- belt conveyor
- sound
- signal
- sound signal
- sound signals
- 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
Links
- 230000005236 sound signal Effects 0.000 title claims abstract description 89
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000003745 diagnosis Methods 0.000 title claims abstract description 24
- 238000012706 support-vector machine Methods 0.000 claims abstract description 41
- 230000004927 fusion Effects 0.000 claims abstract description 26
- 238000013135 deep learning Methods 0.000 claims abstract description 25
- 238000000605 extraction Methods 0.000 claims abstract description 16
- 238000013145 classification model Methods 0.000 claims abstract description 9
- 238000012545 processing Methods 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 49
- 238000013527 convolutional neural network Methods 0.000 claims description 15
- 238000011176 pooling Methods 0.000 claims description 15
- 239000013598 vector Substances 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 12
- 238000001228 spectrum Methods 0.000 claims description 12
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000000354 decomposition reaction Methods 0.000 claims description 10
- 238000001914 filtration Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 9
- 238000009432 framing Methods 0.000 claims description 8
- 230000004044 response Effects 0.000 claims description 7
- 230000009977 dual effect Effects 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 4
- 239000000654 additive Substances 0.000 claims description 3
- 230000000996 additive effect Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 238000003384 imaging method Methods 0.000 claims description 3
- 238000002156 mixing Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000003595 spectral effect Effects 0.000 claims description 3
- 230000007257 malfunction Effects 0.000 claims 2
- 238000001514 detection method Methods 0.000 abstract description 7
- 238000007689 inspection Methods 0.000 abstract description 6
- 238000004519 manufacturing process Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011049 filling Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- OYPRJOBELJOOCE-UHFFFAOYSA-N Calcium Chemical compound [Ca] OYPRJOBELJOOCE-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M99/00—Subject matter not provided for in other groups of this subclass
- G01M99/005—Testing of complete machines, e.g. washing-machines or mobile phones
-
- 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
- G01M99/00—Subject matter not provided for in other groups of this subclass
- G01M99/008—Subject matter not provided for in other groups of this subclass by doing functionality tests
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention provides a belt conveyor fault diagnosis method based on sound signals, which can reduce the labor intensity of inspection personnel and has the characteristics of high detection speed, high real-time performance, high safety and the like; which comprises the following steps: s1, collecting sound signals of the belt conveyor; s2, carrying out improved wavelet threshold denoising processing on the collected sound signals; s3, performing MFCC and deep learning feature extraction on the noise-reduced sound signal of the belt conveyor; s4, establishing a support vector machine classification model and forming a trained SVM model; and S5, putting the extracted characteristic information data into the trained SVM model to obtain posterior probability, then carrying out decision-level fusion by using a D-S evidence theory, finally matching the fusion output result with the running state of the belt conveyor in the known running state of the SVM, and corresponding to the current running state of the belt conveyor if the matching degree of the fusion output result is the highest, thereby completing the fault diagnosis of the belt conveyor.
Description
Technical Field
The invention relates to the technical field of belt conveyor fault diagnosis, in particular to a belt conveyor fault diagnosis method based on sound signals.
Background
With the rapid development of science and technology and the increasingly modern industrial production, various high-intelligence and high-integration large-scale mechanical equipment gradually appears. In port, mine, coal and other industries, the production operation of the belt conveyor has the characteristics of large usage amount, difficult inspection, difficult failure prediction and the like, and through on-site investigation on port production, the belt conveyor needs to operate under high load for a long time due to large production and transportation throughput, so that a failure event which cannot be found in time by manual inspection often occurs, and the research on a belt conveyor failure diagnosis technology is promoted based on the production pain point.
At present belt conveyor adopts traditional manual work mode of patrolling and examining, and the personnel of patrolling and examining need carry tedious instrument of patrolling and examining and shuttle work at the scene, greatly increased the personnel's of patrolling and examining work risk, and belt conveyor fault detection point is many moreover, and the fault detection precision requires highly, and this mode of patrolling and examining makes belt conveyor's the work of patrolling and examining be difficult to accomplish fault detection characteristics such as detection speed is fast, the real-time is high, the security is strong.
Disclosure of Invention
Aiming at the problems, the invention provides a belt conveyor fault diagnosis method based on sound signals, which can reduce the labor intensity of inspection personnel and has the characteristics of high detection speed, high real-time performance, high safety and the like.
In order to achieve the purpose, the invention adopts the following technical scheme:
a belt conveyor fault diagnosis method based on sound signals comprises the following steps:
s1, collecting sound signals of the belt conveyor;
s2, carrying out improved wavelet threshold denoising processing on the collected sound signals;
s3, performing MFCC and deep learning feature extraction on the noise-reduced sound signal of the belt conveyor;
s4, establishing a support vector machine classification model and forming a trained SVM model;
and S5, putting the extracted characteristic information data into the trained SVM model to obtain posterior probability, then carrying out decision-level fusion by using a D-S evidence theory, finally matching the fusion output result with the running state of the belt conveyor in the known running state of the SVM, and corresponding to the current running state of the belt conveyor if the matching degree of the fusion output result is the highest, thereby completing the fault diagnosis of the belt conveyor.
Further, in step S1, sound signals of the respective operation states of the belt conveyor during operation are collected by a sound collection device at a sampling frequency of 48kHz and a number of sampling points of 4096;
further, in the step S5, the operation state of the belt conveyor includes a normal state and three fault states of a carrier roller fault, a belt tearing fault and a roller fault;
further, in the step S2, the denoising process includes the following steps:
s2.1, selecting a wavelet base db6 with attenuation;
s2.2, performing 3-layer wavelet decomposition by using a wavelet base db 6;
s2.3, selecting a fixed threshold lambda, setting a wavelet coefficient after wavelet decomposition as w, and obtaining a wavelet coefficient by a formula
further, in the step S3, the MFCC and the deep learning feature extraction of the noise-reduced belt conveyor sound signal includes the steps of:
s3.1, pre-emphasis, framing and windowing are carried out on the sound signals subjected to noise reduction in the step S2, then the frequency spectrum of each frame of sound signals is obtained through fast Fourier change, a plurality of log energies are generated through filtering of a Mel filter bank, and discrete cosine transform is carried out on the log energies to obtain MFCC characteristics;
s3.2, imaging the sound signals by using a speech spectrogram algorithm, constructing an improved VGG16 convolutional neural network structure, putting the images imaged by the sound signals into the improved VGG16 convolutional neural network structure, and obtaining the deep learning characteristics of the belt conveyor through convolution operation, pooling operation and full connection;
further, in said step S3.1, the extraction of MFCC features comprises the steps of:
s3.1.1, passing the noise-reduced sound signal through a high-pass filter, and calculating the low-frequency signal and the high-frequency signal under the same signal-to-noise ratio, wherein the high-pass filter has the calculation formula: (Z) -1-. mu.z-1Wherein, Z is the pre-emphasized sound signal, Z is a section of sound signal, and mu is a high-pass filter coefficient;
s3.1.2, then framing the sound signal to make the adjacent sound frames set a section of overlap, the overlap area is set to 1/2 or 1/3 of the frame length, the time length of each frame signal is 20 ms-30 ms;
s3.1.3, windowing the sound signal after framing, the function expression is:
wherein D is the window length, and n belongs to [0, D-1 ]; e is a Hanning window adjustment coefficient;
s3.1.4, performing fast Fourier transform on the sound signal, wherein the fast Fourier transform expression is as follows:
where x (N) is the voice input signal, N is the number of points of Fourier transform,
s3.1.5, filtering the transformed sound signal by a Mel filter, wherein the filtering expression is as follows:
wherein m is the number of filters, and f (m) is the mth filter pairCenter frequency of response, Hm(k) Is the frequency response; then an additive expression is calculated according to the frequency response:
then, the logarithmic energy of a frame signal is calculated by using the single logarithmic energy, and the formula is as follows:
where S (m) is the log energy of a filter output,
the logarithmic energy of a frame of signal calculated by single logarithmic energy, M is the maximum number of filters;
s3.1.6 performing discrete cosine transform on the obtained logarithmic energy, setting C (n) as Mel frequency cepstrum coefficient, and calculating
Wherein, L is the order of the Mel frequency cepstrum coefficient;
forming a vector by the obtained plurality of Mel frequency cepstrum coefficients according to the number of the filters, and recording the vector as
Obtaining the MFCC characteristics under the frame of sound signals;
further, in the step S3.2, the extraction of the deep learning feature of the belt conveyor comprises the following steps:
s3.2.1, setting the short-time amplitude spectrum estimation of the sound input signal X (n) as Xn(k) The spectral energy density function is P (n, k), and the sound signal spectrum estimation expression is as follows: p (n, k) ═ Xn(k)|2，
Drawing an image by taking n as an x axis and k as a y axis, and carrying out logarithmic operation on a spectrum energy density function to obtain a spectrogram by taking decibel db as a unit;
s3.2.2, building an improved VGG16 convolutional neural network structure with 8 convolutional layers, 2 pooling layers, 256 full-connection units and 64 full-connection units, putting the sample imaged in the step S3.2.1 into an improved VGG16 convolutional neural network, and setting the size of an input image to be Width Height, the size of a convolutional kernel to be g, the step size to be stride and the filling value to be padding, so that the size f output by the convolutional layersheigh1、fwidth1Comprises the following steps:
size f of the pooling layer outputheigh2、fwidth2Comprises the following steps:
after passing through the convolutional layer and the pooling layer, the first fully-connected layer converts the convolution kernel into a global convolution with the feature map size output by the last convolutional layer, and performs weighted sum on all local features converted, the last fully-connected layer converts the convolution kernel into a convolution with the convolution kernel size of 1 × 1, that is, converts the convolution kernel into a specific numerical value, and all the numerical values are fully connected through the last fully-connected layer, so that the deep learning feature vector of a frame of sound signal is obtained
Further, in the step S4, the establishing a support vector machine classification model includes the following steps:
s4.1, determining an optimal hyperplane:
let given sample Q ═ xi,yi)，yiE { -1, +1}, then there is a classification hyperplane ωTx + b is 0, where ω is a vector, T is the transposed symbol, and b is any real number;
s4.2, finding the maximum linear separable distance:
under the constraint of a classification hyperplane formula, the problem of hyperplane optimization is converted into the problem of function minimum, namely quadratic programming operation:
s4.3, designing a kernel function:
the expression is as follows:
wherein i, j belongs to R, and sigma is a nuclear parameter;
s4.4, constructing a Lagrangian function:
and replacing the inner product of the sample in the high-dimensional space by using the kernel function, wherein the final objective function is expressed as:
wherein C is a penalty factor and l is a constant;
by introducing lagrange multipliers, the formula of the dual form is:
further, in the step S5, obtaining a fusion output result includes the following steps:
s5.1, mixing
And
input variable M as mass function1And M2Inputting the two outputs into the SVM which finishes training to respectively obtain two outputs f (c) without threshold valuesi) And f(s)i) Then, a sigmoid-fitting method is utilized to process a standard SVM non-threshold output result, and the result is converted into a posterior probability:
after substituting two non-threshold outputs f (c)i) And f(s)i) Then, the posterior probability P (c) can be obtained respectivelyi) And P(s)i)；
Wherein f is the non-threshold output of the sample, and p and q are the parameters to be fitted;
s5.2, then for each hypothesis a:
wherein K is a normalization constant, A1And A2Denoted as hypothesis 1 and hypothesis 2, respectively, i.e. the corresponding two posterior probabilities, the formula translates to:
after substituting the posterior probability P (c)i) And P(s)i) Then obtain
The method has the advantages that the collected sound signal of the belt conveyor to be diagnosed can be processed by improved wavelet threshold denoising to obtain the reconstructed sound signal, and then the MFCC characteristic and the deep learning characteristic are obtained after the characteristic extraction, each MFCC feature and each deep learning feature have a belt conveyor state corresponding to the state, the MFCC features and the deep learning features are input into the SVM which completes training to obtain a fusion output result, the obtained fusion output result is utilized to be matched with four types of running states of the SVM under the known running state, the result with the highest matching degree with the fused output result is selected from the four types of running states, which corresponds to the current running state of the belt conveyor, and then the accurate diagnosis result can be output, therefore, the labor intensity of inspection personnel is reduced, and the device has the characteristics of high detection speed, high real-time performance, high safety and the like.
Drawings
FIG. 1 is a schematic block flow diagram of the present invention;
FIG. 2 is a schematic block diagram of the wavelet threshold denoising process of the sound signal of the belt conveyor according to the present invention;
FIG. 3 is a schematic block diagram of the process of feature extraction for an acoustic signal MFCC of a belt conveyor in accordance with the present invention;
FIG. 4 is a block diagram of an improved VGG16 convolutional neural network of the present invention;
FIG. 5 is a schematic block diagram of a decision-level fusion based SVM fault diagnosis process of the present invention;
FIG. 6 is a graph showing the results of the identification of the experimental test values and the actual values according to the present invention.
Detailed Description
As shown in fig. 1, the invention relates to a belt conveyor fault diagnosis method based on sound signals, which comprises the following steps:
s1, collecting sound signals of the belt conveyor:
specifically, in step S1, sound signals of the respective operating states of the belt conveyor during operation are collected by the sound collection device at a sampling frequency of 48kHz and a sampling point number of 4096; the running state of the belt conveyor comprises a normal state and three fault states of roller fault, belt tearing fault and roller fault;
s2, carrying out improved wavelet threshold denoising processing on the collected sound signals, wherein the denoising processing process comprises wavelet basis selection, decomposition layer number selection, threshold function improvement and signal reconstruction;
specifically, in step S2, the noise removal process includes the steps of:
s2.1, selecting a wavelet base db6 with the support length of 5-9 and with symmetry, orthogonality and attenuation;
s2.2, the larger the wavelet decomposition layer number is, the easier the signal component characteristics and the noise component characteristics in the sound can be distinguished, the easier the two can be separated, and the distortion occurs in signal reconstruction if the wavelet decomposition layer number is too high; performing 3-layer wavelet decomposition using wavelet basis db6 according to a sampling frequency of the sound signal;
s2.3, selecting a fixed threshold lambda, and generally selecting 0.6774; the wavelet coefficient after wavelet decomposition is set as w, and the improvement of the threshold function is shown as the following formula:
s3, performing MFCC and deep learning feature extraction on the noise-reduced sound signal of the belt conveyor;
specifically, in step S3, the MFCC and the deep learning feature extraction on the noise-reduced belt conveyor sound signal includes the steps of:
s3.1, pre-emphasis, framing and windowing are carried out on the sound signals subjected to noise reduction in the step S2, then the frequency spectrum of each frame of sound signals is obtained through fast Fourier change, m log energies are generated through filtering of a Mel filter bank, and discrete cosine transform is carried out on the log energies to obtain MFCC characteristics;
s3.2, imaging the sound signals by using a speech spectrogram algorithm, constructing an improved VGG16 convolutional neural network structure, putting the images imaged by the sound signals into a VGG16 convolutional neural network, and obtaining deep learning characteristics of the belt conveyor through convolution operation, pooling operation and full connection;
specifically, in step S3.1, the extraction of MFCC features comprises the steps of:
s3.1.1, passing the noise-reduced sound signal through a high-pass filter, and calculating the low-frequency signal and the high-frequency signal under the same signal-to-noise ratioThe high-pass filter has the calculation formula as follows: (Z) -1-. mu.z-1Wherein, Z is the pre-emphasized sound signal, and Z is a section of sound signal; the value of the high-pass filter coefficient mu is 0.9-1.0, and is usually 0.97;
s3.1.2, pre-emphasizing and then framing the sound signal to make adjacent sound frames overlap, wherein the overlapping area is 1/2 or 1/3 of frame length, and the time length of one frame of signal is 20 ms-30 ms to prevent the amplitude of the front and back frames from being too different;
s3.1.3, dividing the frame, windowing the sound signal, multiplying the window function with each frame signal, making the two ends of the frame signal more continuous, preferably, using a hanning window as the window function, the function expression of the hanning window is:
wherein D is the window length, and n belongs to [0, D-1 ]; e is a Hanning window adjusting coefficient, e belongs to R and is generally 0.46, and R is a real number;
s3.1.4, performing fast Fourier transform on the sound signal, wherein the expression of the fast Fourier transform is as follows:
where x (N) is the voice input signal, N is the number of points of Fourier transform,
s3.1.5, filtering the transformed sound signal by a Mel filter, wherein the filtering expression is as follows:
wherein m is the number of filters, f (m) is the center frequency corresponding to the mth filter, Hm(k) Is the frequency response;
then an additive expression is calculated according to the frequency response:
then, the logarithmic energy of a frame signal is calculated by using the single logarithmic energy, and the formula is as follows:
where S (m) is the log energy of a filter output,
the logarithmic energy of a frame of signal is calculated by using single logarithmic energy, M is the maximum number of filters, and M is generally 22-26;
s3.1.6, performing discrete cosine transform on the obtained logarithmic energy to obtain MFCC, and setting C (n) as a Mel frequency cepstrum coefficient, wherein the expression is as follows:
wherein L is the order of the mel-frequency cepstrum coefficient, and preferably, L is 12-16;
if n equals to i according to the number of the filters, i Mel frequency cepstrum coefficients are obtained, and the i Mel frequency cepstrum coefficients are combined into a vector and recorded as
Obtaining the MFCC characteristics under the frame of sound signals;
specifically, in step S3.2, the extraction of the belt conveyor deep learning feature comprises the steps of:
s3.2.1, because the input is necessarily an image when the convolutional neural network processes, aiming at the problem, the voice signal is imaged by using a speech spectrogram algorithm:
let the short-time amplitude spectrum estimate of the sound input signal X (n) be Xn(k) Function of spectral energy densityFor P (n, k), the expression is estimated for the sound signal spectrum: p (n, k) ═ Xn(k)|2，
Drawing an image by taking n as an x axis and k as a y axis, and carrying out logarithmic operation on a spectrum energy density function to obtain a spectrogram by taking decibel db as a unit;
s3.2.2, constructing an improved VGG16 convolutional neural network, wherein 4 groups of data sets are used based on collected samples, the number of each group of data is 200, the number of the samples is 800, the number of the samples is small, and the data is not suitable for a deep structure of VGG16, so that the traditional model is improved, preferably, 13 convolutional layers are reduced to 8 convolutional layers, the number of pooling layers is reduced to 2, the number of units of a full connection layer is modified, and 4096 of the number of full connection units of classical VGG16 are reduced to 256 and 64; putting the spectrogram sample imaged in the step S3.2.1 into an improved VGG16 convolutional neural network, and setting the size of an input image as Width Height, the size of a convolutional kernel as g, the step size as stride and the filling value as padding, wherein the size f of the output of the convolutional layer isheigh1、fwidth1Comprises the following steps:
size f of the pooling layer outputheigh2、fwidth2Comprises the following steps:
after the convolutional layer and the pooling layer, the first fully-connected layer will convert the convolutional kernel into a global convolution of the output feature map size output by the last convolutional layer, and will convert all of the convolutionThe local features are weighted and summed, the last layer of fully-connected layer is converted into convolution with convolution kernel of 1 × 1 size, namely, the convolution kernel is converted into a specific numerical value, all the numerical values are fully connected through the last layer of fully-connected layer, and the deep learning feature vector of a frame of sound signal is obtained
S4, establishing a Support Vector machine classification model, and forming a trained SVM model, namely a Support Vector Machine (SVM) is a supervised learning algorithm which is provided by Vapnik and is established on the basis of statistical learning, and is widely applied to the field of pattern recognition, wherein the SVM basic principle is that a hyperplane is searched for, so that the hyperplane can just distribute two types of samples on two sides of the hyperplane, the core idea is to map low-dimensional inseparable data to a high-dimensional space by using a kernel function, and then establish the hyperplane, so that the hyperplane can realize the function of optimally classifying the data, and the problem to be solved is to convert linear inseparable into linear separable substantially;
the establishing of the classification model of the support vector machine specifically comprises the following steps: determining an optimal hyperplane, searching for a maximum linear separable distance, designing a kernel function, and constructing a Lagrangian function; the SVM training is implemented by taking the MFCC characteristics and the deep learning characteristics of the sound signal of the belt conveyor obtained in the step S3 as a training set, so that the training is completed, wherein the training method is the existing method;
specifically, in step S4, the establishing the support vector machine classification model includes the following steps:
s4.1, determining an optimal hyperplane:
let given sample Q ═ xi,yi)，yiE { -1, +1}, then there is a classification hyperplane ωTx + b is 0, where ω is a vector, T is the transposed symbol, and b is any real number;
s4.2, searching the maximum linear separable distance:
under the constraint of a classification hyperplane formula, the problem of hyperplane optimization is converted into the problem of function minimum, namely quadratic programming operation:
s4.3, designing a kernel function: radial basis kernel function (RBF) is compared with other kernel functions, RBF parameters are few, function fitting performance is good, and the space dimension can be extended to infinity, and the expression is as follows:
the method comprises the following steps that i and j are integers, RBF is used as a kernel function of a support vector machine, the adjustment of the RBF mainly depends on a kernel parameter sigma, if the kernel parameter sigma is set too large, a feature weight value is attenuated quickly, a higher-dimensional mapping space cannot be created for the kernel function, if the kernel parameter sigma is set too small, an SVM can generate a serious fitting problem, and preferably the kernel parameter sigma is set to be 4;
s4.4, converting the nonlinear problem into a linear problem by utilizing a kernel function and constructing a Lagrangian function:
and replacing the inner product of the sample in the high-dimensional space by using the kernel function, wherein the final objective function is expressed as:
the C is a punishment factor, the size of the punishment factor C also determines the classification effect of the SVM, if the C is set too large, overfitting can be caused in training, if the C is set too small, the sample can not be fully learned, and the accuracy rate of the trained sample is reduced, preferably, the punishment factor C is set to be 11.8; l is a constant;
by introducing lagrange multipliers, the formula of the dual form is:
by constructing the dual, a simpler one of dual problems can be selected to solve, and the algorithm is simplified.
And the algorithm for establishing the support vector machine is the existing algorithm.
S5, putting the extracted MFCC features and deep learning features into a trained SVM model to obtain posterior probability, converting the obtained posterior probability into a standard SVM output result, then performing decision-level fusion by using a D-S evidence theory, finally matching the fusion output result with four types of running states (namely normal running state, carrier roller fault, belt tearing fault and roller fault) under the known running state of the SVM by using the fusion output result, selecting a value with the highest matching degree with the fusion result from the four types of running states to correspond to the current running state of the belt conveyor, obtaining the probability which is most consistent with the current four types of running states, and then outputting the current running state of the belt conveyor, thereby completing the fault diagnosis of the belt conveyor;
specifically, in step S5, obtaining the fusion output result includes the following steps:
s5.1, mixing
And
input variable M as a function of mass1And M2That is, the support vector machine 1 and the support vector machine 2 in fig. 5, input into the trained SVM model to obtain two non-threshold outputs f (c) respectivelyi) And f(s)i) Then, a sigmoid-fitting method is utilized to process a standard SVM non-threshold output result, and the standard SVM non-threshold output result is converted into a posterior probability:
after substituting two non-threshold outputs f (c)i) And f(s)i) Then, the corresponding posterior probability P (c) can be obtained respectivelyi) And P(s)i)；
Wherein f is output as the non-threshold of the sample, and p and q are parameters to be fitted;
s5.2, performing decision-level fusion according to a D-S evidence theory, wherein the D-S evidence theory is proposed in the 60 th 20 th century, is a popularization of Bayes theory, and is a mathematical method for processing uncertainty reasoning problems by introducing a trust function, namely, if an identified universe framework U is set, for each hypothesis A:
wherein K is a normalization constant, A1And A2Respectively marked as hypothesis 1 and hypothesis 2, that is, the corresponding two posterior probabilities, the formula is converted into:
after substituting the posterior probability P (c)i) And P(s)i) Then obtain
In the invention, the sound signal of the belt conveyor to be diagnosed is collected, and the collected sound signal of the belt conveyor to be diagnosed is denoised by the improved wavelet threshold denoising processing method of the step S2; extracting MFCC features and deep learning features of the reconstructed signal after noise reduction through step S3, and respectively recording the MFCC features and the deep learning features as vectors
And
for each vector
And
all have a belt conveyor running state corresponding to the running state of the belt conveyor
And
the process of determining the current running state of the belt conveyor comprises the following steps: will be provided with
And
input variable M as mass function1And M2Inputting the two outputs into the SVM which finishes training to obtain two outputs without threshold values, and respectively recording the outputs as f (c)i) And f(s)i) Converting the non-threshold output into posterior probability by using sigmoid-fitting method to obtain P (c)i) And P(s)i) Finally, two posterior probabilities P (c) can be obtained according to the D-S evidence theoryi) And P(s)i) And fusing to obtain a fusion result, matching the fusion result with four types of running states (namely normal running state, carrier roller fault, belt tearing fault and roller fault) of the SVM under the known running state, selecting the value with the highest matching degree with the fusion result in the four types of running states to correspond to the current running state of the belt conveyor, so that the probability which is most consistent with the current four types of running states is obtained, and then outputting the current running state of the belt conveyor to perform fault diagnosis.
The invention is illustrated by the following application cases:
1. belt conveyor acoustic signal data set preparation
The collection of the sound signal data set is that a noise sensor is used for collecting the sound signal data set in a certain harbor belt conveyor production field, the sampling frequency is 48kHz, the collected data cover sound signals of four states of normal operation of the belt conveyor, carrier roller fault, belt tearing fault and roller fault, the signal time of each group is 17s, and each group is cut into 200 segments, and each segment is 85 ms.
2. Experimental environment configuration
The experimental environment is 64-bit Windows 10 operating system, the CPU is Intel (R) core (TM) i5-6500@3.20Hz, the display card is NVIDIA GeForce GT 710, the memory is 8.00GB, and the MATLAB version is 2018 b.
3. Parameter design for sound signal processing algorithm
In the wavelet threshold denoising, the wavelet basis is selected to be db6, the decomposition layer number is selected to be 3, the threshold is selected to be a heuristic threshold, and the threshold function adopts an improved threshold function; when MFCC characteristics are extracted, in pretreatment, the high-pass filter coefficient is 0.97, the number of sampling points is 4096, a Hamming window is selected as a window function, the window length is 50, the step length is 100, and the number of Mel filter groups is set to 20; during deep learning feature extraction, the improved VGG16 convolutional neural network structure is as follows: 8conv +2maxpool +2fc, spectrogram input size 80 x 80, convolution kernel size set to 3 x 3, pooling level select maximum pooling method, pooling level convolution kernel size 2 x 2, and choose to use the ReLU activation function; the kernel parameter and penalty factor C are set to 4 and 11.8 for SVM.
4. Analysis of Experimental results
The belt conveyor is subjected to fault diagnosis by using the configured sound signal processing algorithm, fig. 6 shows the recognition result of the SVM test value and the true value based on decision-level fusion, and the categories 1, 2, 3 and 4 respectively refer to carrier roller fault, normal signal, belt tearing fault and roller fault, that is, the numbers 1, 2, 3 and 4 marked on the ordinate in fig. 6 respectively correspond to the categories 1, 2, 3 and 4;
in fig. 6, SVM test values are represented by circles, real values of sounds are represented by dots, and if the circles and the dots are all overlapped, the recognition rate is 100%.
TABLE 1 recognition rate of SVM based on decision-level fusion for four types of running states of belt conveyor
According to experimental result analysis, four types of state signals of the belt conveyor are respectively input into 200, wherein 198 normal signals are correctly identified, 2 error identifications are obtained, 194 carrier roller fault signals are correctly identified, 6 error identifications are obtained, 198 belt tearing fault signals are correctly identified, 2 error identifications are obtained, 188 roller fault signals are correctly identified, and 12 error identifications are obtained. The total number of correctly identified samples is 778, the false identification is 22, the total accuracy is 97.25%, and the identification effect is good.
Compared with the prior art, the invention has the beneficial effects that: by collecting the sound signal of the belt conveyor and carrying out improved wavelet threshold denoising processing on the collected sound signal, extracting MFCC characteristics and deep learning characteristics of the noise-reduced sound signals, putting the extracted characteristics into a Support Vector Machine (SVM) classification model to obtain posterior probability, performing decision-level fusion on the posterior probability by using a D-S evidence theory, collecting the sound signals of the belt conveyor to be diagnosed, performing an experiment, outputting a fault diagnosis result, designing by using an acoustic fault diagnosis algorithm, can monitor the working state of the belt conveyor in real time, effectively ensure the production safety of ports, reduce the working intensity of inspection personnel, meanwhile, the abnormal downtime caused by the failure of the belt conveyor is reduced, the expansion of port accidents is avoided, therefore, the invention has great application value and economic benefit for fault diagnosis of large-scale equipment such as belt conveyors and the like.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.
Claims (6)
1. A belt conveyor fault diagnosis method based on sound signals is characterized in that: which comprises the following steps:
s1, collecting sound signals of the belt conveyor;
s2, carrying out improved wavelet threshold denoising processing on the collected sound signals;
s3, performing MFCC and deep learning feature extraction on the noise-reduced sound signal of the belt conveyor;
s4, establishing a support vector machine classification model and forming a trained SVM model;
s5, putting the extracted characteristic information data into the trained SVM model to obtain posterior probability, then carrying out decision-level fusion by using a D-S evidence theory, finally matching the fusion output result with the running state of the belt conveyor under the known running state of the SVM, and corresponding to the current running state of the belt conveyor if the matching degree of the fusion output result is highest, thereby completing the fault diagnosis of the belt conveyor;
in the step S5, the operation states of the belt conveyor include a normal state and three fault states of a roller fault, a belt tearing fault and a roller fault;
in step S3, the MFCC and the deep learning feature extraction of the noise-reduced belt conveyor sound signal includes the steps of:
s3.1, pre-emphasis, framing and windowing are carried out on the sound signals subjected to noise reduction in the step S2, then the frequency spectrum of each frame of sound signals is obtained through fast Fourier change, a plurality of log energies are generated through filtering of a Mel filter bank, and discrete cosine transform is carried out on the log energies to obtain MFCC characteristics;
s3.2, imaging the sound signals by using a speech spectrogram algorithm, constructing an improved VGG16 convolutional neural network structure, putting the images imaged by the sound signals into the improved VGG16 convolutional neural network structure, and obtaining the deep learning characteristics of the belt conveyor through convolution operation, pooling operation and full connection;
in the step S5, obtaining a fusion output result includes the following steps:
s5.1, mixing
And
input variable M as mass function1And M2Inputting the two outputs into the SVM which finishes training to respectively obtain two outputs f (c) without threshold valuesi) And f(s)i) Then, a sigmoid-fitting method is utilized to process a standard SVM non-threshold output result, and the standard SVM non-threshold output result is converted into a posterior probability:
after substituting two non-threshold outputs f (c)i) And f(s)i) Then, the posterior probabilities P (c) can be obtained respectivelyi) And P(s)i)；
Wherein f is the non-threshold output of the sample, and p and q are the parameters to be fitted;
s5.2, then for each hypothesis a:
wherein K is a normalization constant, A1And A2Denoted as hypothesis 1 and hypothesis 2, respectively, i.e. the corresponding two posterior probabilities, the formula translates to:
after substituting the posterior probability P (c)i) And P(s)i) Then obtain
Wherein,
feature vectors are learned for the depth of a frame of sound signals,
is the MFCC feature vector under a frame of sound signals.
2. The belt conveyor fault diagnosis method based on the acoustic signal according to claim 1, characterized in that: in step S1, the sound signal of each operation state of the belt conveyor during operation is collected by the sound collection device at a sampling frequency of 48kHz and a sampling point number of 4096.
3. The belt conveyor fault diagnosis method based on the acoustic signal according to claim 1, characterized in that: in step S2, the denoising process includes the steps of:
s2.1, selecting a wavelet base db6 with attenuation;
s2.2, performing 3-layer wavelet decomposition by using a wavelet base db 6;
s2.3, selecting a fixed threshold lambda, setting a wavelet coefficient after wavelet decomposition as w, and obtaining a wavelet coefficient by a formula
Modifying a threshold function, and then reconstructing by using wavelet coefficients processed by the threshold function to obtain a noise-reduced belt conveyor sound signal, wherein j and k are integers, and a>0。
4. The method for diagnosing the malfunction of the belt conveyor based on the acoustic signal according to claim 1, characterized in that: in said step S3.1, the extraction of MFCC features comprises the steps of:
s3.1.1, passing the noise-reduced sound signal through a high-pass filter, and calculating the low-frequency signal and the high-frequency signal under the same signal-to-noise ratio, wherein the high-pass filter has the calculation formula: (Z) -1-. mu.z-1Wherein, Z is the pre-emphasized sound signal, Z is a section of sound signal, and mu is a high-pass filter coefficient;
s3.1.2, then framing the sound signal to make the adjacent sound frames set a section of overlap, the overlap area is set to 1/2 or 1/3 of the frame length, the time length of each frame signal is 20 ms-30 ms;
s3.1.3, windowing the sound signal after framing, the function expression is:
wherein D is the window length, n belongs to [0, D-1 ]; e is a Hanning window adjustment coefficient;
s3.1.4, performing fast Fourier transform on the sound signal, wherein the fast Fourier transform expression is as follows:
wherein x (N) is a voice input signal, N is the number of Fourier transform points,
s3.1.5, filtering the transformed sound signal by a Mel filter, wherein the filtering expression is as follows:
wherein m is the number of filters, f (m) is the center frequency corresponding to the mth filter, Hm(k) Is the frequency response;
then an additive expression is calculated according to the frequency response:
then, the logarithmic energy of a frame signal is calculated by using the single logarithmic energy, and the formula is as follows:
where S (m) is the log energy of a filter output,
the logarithmic energy of a frame of signal calculated by single logarithmic energy, M is the maximum number of filters;
s3.1.6 performing discrete cosine transform on the obtained logarithmic energy, setting C (n) as Mel frequency cepstrum coefficient, and calculating
Wherein, L is the order of the Mel frequency cepstrum coefficient;
forming a vector by the obtained plurality of Mel frequency cepstrum coefficients according to the number of the filters, and recording the vector as
The MFCC characteristic under this frame of sound signal is obtained.
5. The belt conveyor fault diagnosis method based on the acoustic signal according to claim 4, characterized in that: in said step S3.2, the extraction of the belt conveyor deep learning feature comprises the steps of:
s3.2.1, let the short-time amplitude spectrum estimation of the sound input signal X (n) be Xn(k) The spectral energy density function is P (n, k), and the sound signal spectrum estimation expression is as follows: p (n, k) ═ Xn(k)|2，
Drawing an image by taking n as an x axis and k as a y axis, and carrying out logarithmic operation on a spectrum energy density function to obtain a spectrogram by taking decibel db as a unit;
s3.2.2, building an improved VGG16 convolutional neural network structure with 8 convolutional layers, 2 pooling layers and 256 and 64 full-connection unit numbers, putting the sample imaged in the step S3.2.1 into an improved VGG16 convolutional neural network, and setting the size of an input image as Width Height, the size of a convolutional kernel as g, and the step size as stEdge, padding value is padding, size of convolutional layer output fheigh1、fwidth1Comprises the following steps:
size f of the pooling layer outputheigh2、fwidth2Comprises the following steps:
after passing through the convolutional layer and the pooling layer, the first fully-connected layer converts the convolution kernel into a global convolution with the feature map size output by the last convolutional layer, and performs weighted sum on all local features converted, the last fully-connected layer converts the convolution kernel into a convolution with the convolution kernel size of 1 × 1, that is, converts the convolution kernel into a specific numerical value, and all the numerical values are fully connected through the last fully-connected layer, so that the deep learning feature vector of a frame of sound signal is obtained
6. The method for diagnosing the malfunction of the belt conveyor based on the acoustic signal according to claim 1, characterized in that: in step S4, the establishing a classification model of a support vector machine includes the following steps:
s4.1, determining an optimal hyperplane:
let given sample Q ═ xi,yi)，yiE { -1, +1}, then there is a classification hyperplane ωTx + b is 0, where ω is a vector, T is the transposed symbol, and b is any real number;
s4.2, finding the maximum linear separable distance:
under the constraint of a classification hyperplane formula, the problem of hyperplane optimization is converted into the problem of function minimum, namely quadratic programming operation:
s4.3, designing a kernel function:
the expression is as follows:
wherein i, j belongs to R, and sigma is a nuclear parameter;
s4.4, constructing a Lagrangian function:
and replacing the inner product of the sample in the high-dimensional space by using the kernel function, wherein the final objective function is expressed as:
wherein C is a penalty factor and l is a constant;
by introducing lagrange multipliers, the formula of the dual form is:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110651133.5A CN113405825B (en) | 2021-06-11 | 2021-06-11 | Belt conveyor fault diagnosis method based on sound signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110651133.5A CN113405825B (en) | 2021-06-11 | 2021-06-11 | Belt conveyor fault diagnosis method based on sound signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113405825A CN113405825A (en) | 2021-09-17 |
| CN113405825B true CN113405825B (en) | 2022-06-17 |
Family
ID=77683590
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110651133.5A Active CN113405825B (en) | 2021-06-11 | 2021-06-11 | Belt conveyor fault diagnosis method based on sound signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113405825B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114013957B (en) * | 2021-11-29 | 2023-03-17 | 天津电子信息职业技术学院 | Conveying belt longitudinal tearing detection method based on sound signals and related equipment |
| CN115266914B (en) * | 2022-03-28 | 2024-03-29 | 华南理工大学 | Pile sinking quality monitoring system and method based on acoustic signal processing |
| CN114972349B (en) * | 2022-08-01 | 2022-10-25 | 山东西曼克技术有限公司 | Carrier roller running state detection method and system based on image processing |
| CN115440242A (en) * | 2022-09-02 | 2022-12-06 | 天津市恒一机电科技有限公司 | Method for detecting longitudinal tearing of conveying belt and related equipment |
| CN115424635B (en) * | 2022-11-03 | 2023-02-10 | 南京凯盛国际工程有限公司 | Cement plant equipment fault diagnosis method based on sound characteristics |
| CN116359642B (en) * | 2023-03-10 | 2024-01-23 | 湖南金烽信息科技有限公司 | Transformer running state 5G intelligent monitoring system and method |
| CN115892923B (en) * | 2023-03-10 | 2023-06-16 | 四川东林重工科技股份有限公司 | Intelligent inspection robot for belt conveyor |
| CN116990004A (en) * | 2023-08-04 | 2023-11-03 | 无锡斯帝尔科技有限公司 | GIS mechanical fault diagnosis method and system based on acoustic emission signals |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105510729A (en) * | 2014-10-11 | 2016-04-20 | 国家电网公司 | Overheating fault diagnosis method of transformer |
| CN108332970A (en) * | 2017-11-17 | 2018-07-27 | 中国铁路总公司 | A kind of Method for Bearing Fault Diagnosis based on LS-SVM and D-S evidence theory |
| CN109850519A (en) * | 2018-11-30 | 2019-06-07 | 太原理工大学 | A kind of coal mine conveyer abnormal detector and method based on sound positioning |
| CN110991363A (en) * | 2019-12-09 | 2020-04-10 | 天地（常州）自动化股份有限公司 | Method for extracting CO emission characteristics of coal mine safety monitoring system under different coal mining processes |
| CN111259921A (en) * | 2019-12-19 | 2020-06-09 | 杭州安脉盛智能技术有限公司 | Transformer sound anomaly detection method based on improved wavelet packet and deep learning |
| JP2020183307A (en) * | 2019-05-08 | 2020-11-12 | 三菱電機株式会社 | Belt conveyor monitoring system |
-
2021
- 2021-06-11 CN CN202110651133.5A patent/CN113405825B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105510729A (en) * | 2014-10-11 | 2016-04-20 | 国家电网公司 | Overheating fault diagnosis method of transformer |
| CN108332970A (en) * | 2017-11-17 | 2018-07-27 | 中国铁路总公司 | A kind of Method for Bearing Fault Diagnosis based on LS-SVM and D-S evidence theory |
| CN109850519A (en) * | 2018-11-30 | 2019-06-07 | 太原理工大学 | A kind of coal mine conveyer abnormal detector and method based on sound positioning |
| JP2020183307A (en) * | 2019-05-08 | 2020-11-12 | 三菱電機株式会社 | Belt conveyor monitoring system |
| CN110991363A (en) * | 2019-12-09 | 2020-04-10 | 天地（常州）自动化股份有限公司 | Method for extracting CO emission characteristics of coal mine safety monitoring system under different coal mining processes |
| CN111259921A (en) * | 2019-12-19 | 2020-06-09 | 杭州安脉盛智能技术有限公司 | Transformer sound anomaly detection method based on improved wavelet packet and deep learning |
Non-Patent Citations (4)
| Title |
|---|
| Belt Conveyor Roller Fault Audio Detection Based On The Wavelet Neural Network;Xiao-ping Jiang等;《2015 lith International Conference on Natural Computation (lCNC)》;20151231;954-958 * |
| D-S证据理论在带式输送机故障诊断中的应用;王莉等;《江南大学学报(自然科学版)》;20130228(第01期);34-37 * |
| 基于多源异构信息迁移学习的融合故障诊断方法;陈丹敏等;《信息工程大学学报》;20200415(第02期);29-34 * |
| 基于深度学习模型的电力变压器故障声音诊断方法研究;吴帆等;《电声技术》;20200105(第01期);81-85 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113405825A (en) | 2021-09-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113405825B (en) | Belt conveyor fault diagnosis method based on sound signals | |
| CN111238814B (en) | Rolling bearing fault diagnosis method based on short-time Hilbert transform | |
| CN110390952B (en) | City sound event classification method based on dual-feature 2-DenseNet parallel connection | |
| Wang et al. | ia-PNCC: Noise Processing Method for Underwater Target Recognition Convolutional Neural Network. | |
| CN112885368B (en) | Multi-band spectral subtraction vibration signal denoising method based on improved capsule network | |
| Shuuji et al. | Low-speed bearing fault diagnosis based on improved statistical filtering and convolutional neural network | |
| Yao et al. | A recursive denoising learning for gear fault diagnosis based on acoustic signal in real industrial noise condition | |
| CN112052712B (en) | Power equipment state monitoring and fault identification method and system | |
| CN112908344A (en) | Intelligent recognition method, device, equipment and medium for bird song | |
| CN113990303B (en) | Environmental sound identification method based on multi-resolution cavity depth separable convolution network | |
| CN114234061B (en) | Intelligent discrimination method for water leakage sound of pressurized operation water supply pipeline based on neural network | |
| CN115641871A (en) | Fan blade abnormity detection method based on voiceprint | |
| CN113707175B (en) | Acoustic event detection system based on feature decomposition classifier and adaptive post-processing | |
| CN118016055A (en) | Heart sound classifying method based on two-way long-short period memory network and multi-head attention mechanism | |
| CN111968669B (en) | Multi-element mixed sound signal separation method and device | |
| CN113724731A (en) | Method and device for audio discrimination by using audio discrimination model | |
| CN117419915A (en) | Motor fault diagnosis method for multi-source information fusion | |
| CN111785262A (en) | Speaker age and gender classification method based on residual error network and fusion characteristics | |
| CN116524273A (en) | Method, device, equipment and storage medium for detecting draft tube of power station | |
| CN116484258A (en) | Elevator traction machine bearing fault diagnosis method | |
| CN115901265A (en) | Rolling bearing fault diagnosis method based on MFCC-FcaNet | |
| CN114898766A (en) | Distributed optical fiber voice enhancement method based on GAN network and tunnel rescue system | |
| CN114638266A (en) | VMD-WT-CNN-based multi-fault coupling signal processing and diagnosis method for gas turbine rotor | |
| CN112560674B (en) | Method and system for detecting sound signal quality | |
| CN112201226B (en) | Sound production mode judging method and system |
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 |