CN111983020B - System and method for detecting and identifying internal defects of metal component through knocking - Google Patents
System and method for detecting and identifying internal defects of metal component through knocking Download PDFInfo
- Publication number
- CN111983020B CN111983020B CN202010862602.3A CN202010862602A CN111983020B CN 111983020 B CN111983020 B CN 111983020B CN 202010862602 A CN202010862602 A CN 202010862602A CN 111983020 B CN111983020 B CN 111983020B
- Authority
- CN
- China
- Prior art keywords
- knocking
- signal
- defect
- sound signal
- calculating
- 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
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/04—Analysing solids
- G01N29/045—Analysing solids by imparting shocks to the workpiece and detecting the vibrations or the acoustic waves caused by the shocks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4454—Signal recognition, e.g. specific values or portions, signal events, signatures
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/46—Processing the detected response signal, e.g. electronic circuits specially adapted therefor by spectral analysis, e.g. Fourier analysis or wavelet analysis
-
- 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
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- General Health & Medical Sciences (AREA)
- Pathology (AREA)
- Engineering & Computer Science (AREA)
- Acoustics & Sound (AREA)
- Mathematical Physics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
A metal member internal defect knocking detection and identification system and an identification method belong to the technical field of acoustic detection. The invention comprises a checking hammer, an information acquisition and control unit, an information processing and display unit, a knocking sound signal acquisition sensor and/or a vibration signal acquisition sensor, wherein the knocking sound signal acquisition sensor acquires a knocking sound signal generated by the checking hammer knocking a metal component, the vibration signal acquisition sensor acquires a knocking vibration signal generated by the checking hammer knocking the metal component, the information acquisition and control unit performs A/D sampling on the knocking sound signal and the knocking vibration signal, and the information processing and display unit analyzes and identifies the knocking sound signal and the knocking vibration signal. The invention simulates the process of 'hearing' defects by a inspector with abundant experience through the knocking sound, identifies the knocking process of the inspection hammer, realizes the rapid detection and intelligent identification of the defects of the pressure-bearing metal component, and improves the accuracy and objectivity of the identification of the internal defects.
Description
Technical Field
The invention belongs to the technical field of acoustic detection, and particularly relates to a knocking detection and identification system for internal defects of a metal member.
Background
The knock detection is one of nondestructive detection methods of acoustic vibration, and is a method of exciting a test piece to be detected to generate mechanical vibration and determining the mass of the object to be detected from the measurement result of the obtained acoustic wave. The method is used for judging whether the wheel is defective or not according to experience through sound generated after the small hammer strikes the wheel when a railway worker detects whether the wheel is complete or not at the earliest. The knocking test method is also commonly used in the macroscopic test of pressure-bearing special equipment, the related parts of the pressure-bearing elements of the boiler are knocked by a test hammer, and the defects are judged by auditory sense and touch sense and vision, so that the method is an important and indispensable test method in the test of the boiler, and the degradation, corrosion, scaling and the like of the metal materials of the detected objects can be primarily judged by correctly using the method. For example, when the hammering method is used for inspecting the pressure-receiving elements such as the inner wall and the outer wall of the pressure container, the boiler cylinder body, the boiler liner, the tube plate large transverse water tube and the like, the condition of the inspected pressure-receiving element is judged by listening to the sound made by hammering and the rebound degree of the inspection hammer. If the tested pressed element emits a clear and pure sound, this is an indication that the condition of the steel sheet is good; if the sound is smoldering and cloudy, the steel plate is likely to have dust, an interlayer or the metal surface is severely corroded, or the metal surface is already deposited with thick scale; such as making a salad sound, indicating that the steel sheet has developed cracks. The hammering method is used for checking the quality of the short bracing, if the bracing is broken, the resilience force of the hammer head is obviously weakened; the head of the hammering short bracing piece is tested to generate sound to be stuffy, so that the short bracing piece is indicated to have cracks or breaks.
After the hammer is detected to strike, the method for judging the defects of the metal pressure-bearing member of the special equipment by using the ear hearing and the hand feeling is simple, convenient and quick, and proved by long-term practice, the method is effective, but the method is largely established on subjective judgment of a detector, is very dependent on personal experience, and often needs to be accumulated in practice for many years.
Disclosure of Invention
The invention mainly solves the technical problems in the prior art and provides a system and a method for detecting and identifying internal defects of a metal member. The invention can simulate the process of 'hearing' defects by a inspector with abundant experience through the knocking sound, so that the determination of the knocking test result of the testing hammer is more accurate, the serious dependence of the knocking test of pressure-bearing metal components of special equipment such as boilers, pressure vessels, pressure pipelines and the like on personal experience is favorably solved, and the accuracy and the guest functionality of identifying the internal defects are improved.
The technical problems of the invention are mainly solved by the following technical proposal: the utility model provides a metal component internal defect beats detection identification system, includes inspection hammer, information acquisition and control unit, information processing and display unit, knocks sound signal acquisition sensor and/or vibration signal acquisition sensor, knocks sound signal acquisition sensor and gathers the knocks sound signal that the inspection hammer knocks metal component and produce, vibration signal acquisition sensor gathers the knocks vibration signal that the inspection hammer knocked metal component and produce, vibration signal acquisition sensor installs the periphery at the metal component is knocked the position, information acquisition and control unit set up the control and the parameter of system, carries out the AD sampling to knocks sound signal and knocks vibration signal, information processing and display unit carries out analysis and discernment to knocks sound signal and knocks vibration signal to the defect identification result is notified the user.
Preferably, the test hammer is provided with a force sensor, which records the striking force of the test hammer.
The invention also discloses a method for identifying the internal defect knocking detection and identification system of the metal component, which comprises the following steps:
(1) Collecting knocking sound signals and knocking vibration signals of the metal component, and recording the knocking force;
(2) Classifying and calibrating the collected knocking information samples according to the defect characteristics of the metal components;
(3) Preprocessing and filtering denoising are carried out on the knocked sound signal, wherein the preprocessing and filtering denoising comprises sliding filtering, standardization and abnormal point and noise removal of the signal;
(4) Calculating the zero crossing rate and short-time energy of each frame of signal, carrying out endpoint detection and audio frequency segment division on the knocking sound signal, eliminating useless silence segment signals, and dividing the continuously collected knocking sound signal into equal-length audio frequency segments corresponding to the knocking times;
(5) Performing MFCC feature extraction based on wavelet packet decomposition on the segmented audio signal in the step (4), and establishing an MFCC feature set of the knocking sound signal;
(6) Training the MFCC feature sets of the knocking sound signals and the knocking vibration signals by adopting a hidden Markov algorithm, and establishing an HMM defect model;
(7) And identifying the same or similar defects from the knocking signals by using a binary information fusion and correlation reverse evidence algorithm according to the trained HMM defect model.
Preferably, in the step (2), the method for classifying and calibrating the knocking information sample is as follows: and the inspector recognizes the defect of the knocked metal member by using an ear-hearing hand feeling mode, and adopts ultrasonic detection or digital ray technology to secondarily confirm the information sample which is suspected to be classified.
Preferably, in the step (2), the method for classifying and calibrating the knocking information sample is as follows: and manufacturing a batch of manual test pieces with prefabricated defects, simulating knocking information acquisition of a testing site, and manually changing the knocking force, changing the position of a knocking sound signal acquisition sensor and manufacturing environmental noise when the knocking information is acquired.
Preferably, in the step (3), a five-point three-time sliding filtering method is adopted to remove high-frequency random noise in the knocking sound signals, wavelet packet denoising is adopted to denoise the knocking sound signals, and the maximum value of the filtered signals is utilized to normalize the filtered knocking sound signals.
Preferably, the step (4) specifically includes:
(4.1) framing the knocked sound signal;
(4.2) calculating short-time average energy and short-time zero-crossing rate of a certain frame signal;
(4.3) dividing the knocking sound signals into a plurality of states, sequentially judging the states of the frames by using a state machine method, sequentially locating the coordinates of the beginning and the end of the knocking states by the signals, and dividing the knocking sound signals into a plurality of audio segments with equal lengths;
(4.4) monitoring abnormal noise possibly existing in the knocking sound signal by using a correlation analysis method by utilizing the homology of the knocking vibration signal and the characteristic of being not influenced by the environmental noise.
Preferably, the step (5) specifically includes:
(5.1) carrying out signal pre-emphasis processing on the knocked sound signal, and improving a high-frequency part to highlight a high-frequency formant;
(5.2) framing the pre-emphasis-processed knocking sound signals, and adding a hamming window to each frame of signals;
(5.3) performing a fast fourier transform on the knocked sound signal;
(5.4) performing multi-layer wavelet packet decomposition on the knocked sound signal;
(5.5) defining a triangular filter bank;
(5.6) calculating the logarithmic energy of each filter bank output;
(5.7) introducing the logarithmic energy signal into discrete cosine transform, and obtaining MFCC coefficients through discrete cosine transform;
(5.8) calculating a first order difference of the MFCC coefficients;
(5.9) combining the MFCC coefficients and the first order differences of the MFCC coefficients to obtain MFCC characteristic parameters of the tapping sound signal;
and (5.10), repeating the steps (5.1) - (5.9), and performing signal processing and characteristic parameter extraction on the knocking vibration signal to obtain the MFCC characteristic parameter of the knocking vibration signal.
Preferably, the step (6) specifically includes:
(6.1) dividing and classifying the knock signal samples;
(6.2) initializing HMM model parameters;
(6.3) calculating a given input signal sequence, and calculating forward and backward probabilities, state transition probabilities and mixed output probabilities of the current signal sequence moment;
(6.4) calculating the HMM parameter state probability distribution, the state transition matrix and the state emission matrix of the sample set according to the maximum expectation estimation algorithm;
(6.5) calculating the output probability of the knocking characteristic signal sequence at a certain time parameter and the previous time of the time parameter by using a Viterbi algorithm;
(6.6) repeating the steps (6.1) - (6.5) to obtain HMM model parameters of a plurality of sample sets.
Preferably, the step (7) specifically includes:
(7.1) repeating the steps (3) - (5) for the input tapping sound signal and the tapping vibration signal;
(7.2) calculating a relative fusion probability;
and (7.3) calculating cross-correlation coefficient vectors of the knocking sound signals and the knocking sound reference signals of the defect states after the knocking sound signals are respectively delayed by at least one sampling point, calculating standard deviations of the cross-correlation coefficient vectors, and carrying out inverse decision on the defect states corresponding to the current knocking sound signals through the sizes of the standard deviations.
The invention has the beneficial effects that: from the bionics perspective, the invention simulates the process of 'hearing' defects by a inspector with abundant experience through the knocking sound, identifies the knocking process of the inspection hammer, realizes the rapid detection and intelligent identification of the defects of the pressure-bearing metal member, and improves the accuracy and objectivity of the identification of the internal defects.
Drawings
FIG. 1 is a schematic view of a construction of the present invention;
FIG. 2 is a schematic representation of MFCC feature set extraction of a tap signal of the present invention;
FIG. 3 is a schematic illustration of HMM training model creation of sample defects of the present invention;
FIG. 4 is a schematic flow chart of the HMM-based "binary information fusion+correlation reverse evidence" algorithm of the present invention;
FIG. 5 is a schematic illustration of 30 consecutive beats collected in an embodiment of the invention;
FIG. 6 is a schematic diagram of a signal sliding filtered waveform in accordance with an embodiment of the present invention;
FIG. 7 is a schematic representation of the original signal (part) of a strike sound in an embodiment of the invention;
FIG. 8 is a schematic diagram showing a comparison of (part of) a knock sound signal before and after denoising a wavelet packet according to an embodiment of the present invention;
FIG. 9 is a schematic representation of a single tap sound signal without normalization in an embodiment of the present invention;
FIG. 10 is a schematic diagram of frame energy of a percussive sound signal in accordance with an embodiment of the present invention;
FIG. 11 is a schematic diagram of the frame zero crossing rate of a percussive sound signal in an embodiment of the invention;
FIG. 12 is a schematic diagram of the signal segmentation results of the acquired signals in an embodiment of the present invention;
FIG. 13 (a) is a schematic diagram of a tap sound signal at a tap in an embodiment of the invention;
FIG. 13 (b) is a schematic diagram of a tap vibration signal at a tap in accordance with an embodiment of the invention;
FIG. 14 is a schematic representation of Fourier frequency domain plots of both a tap sound signal and a tap vibration signal for a single tap in an embodiment of the invention;
FIG. 15 is a schematic diagram of correlation coefficients of a tap sound signal and a tap vibration signal at a tap in an embodiment of the invention;
FIG. 16 is a schematic diagram of correlation coefficients of a tapping sound signal and a tapping vibration signal of different types of defective test pieces according to an embodiment of the present invention;
FIG. 17 is a schematic diagram of an embodiment of the present invention in which Gaussian white noise is added to an original tapping sound signal;
FIG. 18 is a schematic diagram of correlation coefficients of a delay sequence of a tapping sound signal and a tapping vibration signal according to an embodiment of the present invention;
FIG. 19 is a schematic diagram of parameters of a tapping sound signal MFCC in accordance with an embodiment of the present invention;
FIG. 20 is a schematic diagram of parameters of a tap vibration signal MFCC in accordance with an embodiment of the invention;
FIG. 21 is a schematic illustration of a different defect sample set in an embodiment of the present invention.
In the figure: 1. checking a hammer; 2. a knock signal acquisition sensor; 3. a vibration signal acquisition sensor; 4. an information acquisition and control unit; 5. an information processing and displaying unit; 6. a force sensor.
Detailed Description
The technical scheme of the invention is further specifically described below through examples and with reference to the accompanying drawings.
The utility model provides a metal component internal defect beats detection identification system, as shown in fig. 1, including inspection hammer 1, information acquisition and control unit 4, information processing and display unit 5, beat sound signal acquisition sensor 2 and/or vibration signal acquisition sensor 3, beat sound signal that sound signal acquisition sensor 2 gathered inspection hammer 1 beat metal component produced, vibration signal acquisition sensor 3 gathers inspection hammer 1 beat metal component produced beat vibration signal, vibration signal acquisition sensor 3 installs in the periphery of being beaten the position of metal component, for example is 10cm-20cm apart from being beaten the position, vibration signal acquisition sensor 3 installs the mode can adopt to keep flat, magnet absorption, tie up the rope and tie up etc. the control and the parameter of system set up, carry out the A/D sampling to beat sound signal and beat vibration signal, sound and vibration data that gathers supply information processing and display unit 5 to handle the analysis, information processing and display unit 5 carries out analysis and discernment to beating sound signal and beat vibration signal, and discernment defect recognition result user.
In the detection and identification process, the knock signal acquisition sensor 2 or the vibration signal acquisition sensor 3 can be used independently, and the knock signal acquisition sensor 2 and the vibration signal acquisition sensor 3 can be used simultaneously, so that the accuracy of defect identification can be better when the knock signal acquisition sensor 2 and the vibration signal acquisition sensor 3 are used simultaneously.
The test hammer 1 is provided with a force sensor 6, and the force sensor 6 records the striking force of the test hammer 1. The check hammer 1 includes, but is not limited to, a hammer-like striking device, and other devices capable of performing a striking function are also included.
The knock signal acquisition sensor 2 includes, but is not limited to, an explosion-proof or non-explosion-proof external sensor, an integrated sensor of a terminal device, and the like.
The vibration signal acquisition sensor 3 includes, but is not limited to, an explosion-proof or non-explosion-proof external sensor, an integrated sensor of a terminal device, and the like.
The information processing and display unit 5 includes, but is not limited to, a portable mobile terminal having an information processing function, a notebook computer, an industrial personal computer, and other general-purpose electronic processing devices; the notification forms of the defect recognition result include, but are not limited to, screen display, sound notification, light alarm, etc.
The invention also discloses a method for identifying the internal defect knocking detection and identification system of the metal component, which comprises the following steps:
(1) Collecting knocking sound signals and knocking vibration signals of the metal component, and recording the knocking force; the knocking sound signal and the knocking vibration signal are homologous signals and complement each other, the knocking force is a reference signal, and the effectiveness of the method is not affected by the lack of the knocking force value;
(2) Classifying and calibrating the collected knocking information samples according to the defect characteristics of the metal components; the knocking information sample classification and calibration adopts two modes: firstly, a person with abundant knocking detection experience of the checking hammer 1 identifies the defect of a knocked metal component by using a traditional ear hearing hand feeling mode, and adopts ultrasonic detection (UT) or Digital Ray (DR) technology to secondarily confirm the information sample which is classified and suspected; secondly, manufacturing a batch of artificial test pieces with prefabricated defects, simulating knocking information acquisition of a special equipment inspection site, and manually changing knocking force, changing the position of a knocking sound signal acquisition sensor 2 and manufacturing environmental noise during knocking information acquisition so as to increase the diversity of information samples and the robustness of a final defect identification method;
(3) Preprocessing and filtering denoising are carried out on the knocked sound signal, wherein the preprocessing and filtering denoising comprises sliding filtering, standardization and abnormal point and noise removal of the signal; preferably, a five-point three-time sliding filtering method can be adopted to remove high-frequency random noise in the knocking sound signals, wavelet packet denoising can be adopted to denoise the knocking sound signals if necessary, and then the maximum value of the filtered signals is utilized to normalize the filtered knocking sound signals; the following formula is shown:
(4) Calculating the zero crossing rate and short-time energy of each frame of signal, carrying out endpoint detection and audio frequency segment division on the knocking sound signal, eliminating useless silence segment signals, and dividing the continuously collected knocking sound signal into equal-length audio frequency segments corresponding to the knocking times; the method comprises the following steps:
(4.1) framing the knocking sound signal X, wherein a certain overlapping section exists between two adjacent frames, and the length m and the overlapping length l of each frame are determined according to the sampling frequency fs of the sampling system and the frequency f of the knocking sound;
(4.2) calculating the short-time average energy En and the short-time zero-crossing rate Zn of the nth frame signal; the following formula is shown:
w (m) is a window function;
(4.3) dividing the knocking sound signal X into four states of silence (0), ambiguity (1), knocking (2) and ending (3), respectively representing that the knocking sound is before knocking, uncertain, knocking is occurring and knocking is ending, setting an energy threshold Et, a short zero crossing rate threshold Zt, a silence length sil and a signal minimum length threshold Lt of the knocking state, sequentially judging the states of each frame by using a state machine method, sequentially signaling the position coordinates of the beginning and ending of the knocking state, and dividing the knocking sound signal X into k audio segments with equal lengths;
and (4.4) monitoring abnormal noise (including cavity reverberation) possibly existing in the knocking sound signal by utilizing the homology of the knocking vibration signal and the characteristic of being not influenced by environmental noise and adopting a correlation analysis method, and reminding in a result display step when the correlation coefficient is smaller than a threshold corrThr.
(5) Performing MFCC (Mel cepstrum coefficient) feature extraction based on wavelet packet decomposition on the segmented audio signal in the step (4), and establishing an MFCC feature set of the knocking sound signal; the method comprises the following steps:
(5.1) carrying out signal pre-emphasis processing on the knocked sound signal, and improving a high-frequency part to highlight a high-frequency formant; the following formula is shown:
H(Z)＝1-μz -1
wherein μ has a value of between 0.9 and 1.0; preferably, to select a proper μ value, starting with μ=0.9, stepping with step=0.0005, calculating correlation coefficients of the pre-weighted signal and the original signal, respectively, and taking μ when the correlation coefficient is maximum max ；
(5.2) framing the knocking sound signals subjected to the pre-emphasis treatment in the step (5.1), wherein a section of overlapping area is formed between two adjacent frames, the number of data points in each frame is N, and the number of the overlapping area is N/4; the N value is too large to reduce the time domain resolution of the characteristic data, and too small to effectively cover the time domain change of the signal, and the calculated amount is increased; selecting N=200 as an initial value, and selecting N values with relatively smaller distance by comparing the discrete Frechet distances of the MFCC characteristics when different N values, thereby considering the time domain resolution and the time domain coverage of the characteristic data; dividing the knocking sound signal into T frames, and adding a Hamming window to each frame signal to reduce spectrum leakage;
(5.3) performing a fast fourier transform on the knocked sound signal; performing fast Fourier transform FFTS on each frame of signal subjected to Hamming window in step (5.2), and performing modulo square on the frequency spectrum of each frame of signal to obtain a Fourier power spectrum matrix P f The size is NxT;
(5.4) performing multi-layer wavelet packet decomposition on the knocked sound signal; performing J-layer wavelet packet decomposition with wavelet function of db3 to obtain J-layer 2 J The wavelet packet coefficients are sequenced according to the frequency levels corresponding to the wavelet coefficients, and the J-th layer wavelet coefficients are respectively reconstructed to obtain 2 J Time domain signals respectively corresponding to 2 from low to high J Each frequency band; pair 2 J Dividing the time domain signals into T frames according to the step (5.2), and respectively calculating the average power spectrum of each frame; the following formula is shown:
wavelet packet average power spectrum matrix P w Is of size 2 J ×T；
(5.5) defining a triangular filter bank with a center frequency f (M), m=1, 2, & M; f (m) is obtained from the Mel frequency relationship, assuming that the lowest frequency is 0 In Mel space and the highest frequency is c In (1+0.5fs/700), c is a constant, the center frequency intervals of the filters are equal In Mel frequency space, and the Mel center frequencies of the filters are:
the center frequency of the normal frequency domain can be obtained according to the definition of Mel frequency:
the spacing between f (m) in the frequency domain is widened with the increase of m value, namely the filter is relatively narrow at lower frequency band, and the filter is widened with the increase of frequency; the frequency response of the triangular filter is defined as:
in the middle of
(5.6) calculating the logarithmic energy of each filter bank as:
p may be the Fourier average power spectrum P f Or the average power spectrum P of the wavelet packet w And selecting according to actual conditions to finally obtain a total output logarithmic energy signal matrix S with the size of MxT.
(5.7) introducing the logarithmic energy signal S into Discrete Cosine Transform (DCT) to obtain MFCC coefficients:
the L-order refers to the number of MFCC coefficients, M is the number of triangular filters, and the size of the MFCC coefficient matrix C is L multiplied by T;
(5.8) calculating the first order difference of the MFCC coefficients, which can be calculated as follows:
n=1, 2,..l, i=3, 4,.. T is the number of frames of the knocking sound signals;
(5.9) the MFCC characteristic parameter consists of a first order difference of the MFCC coefficient in step (5.7) and the MFCC coefficient in step (5.8), and has a size of 2l× (T-4), denoted as AMFCC;
(5.10) performing signal processing and characteristic parameter lifting on the knocking vibration signal by adopting the same steps as the knocking sound signal to obtain MFCC characteristic parameters of the knocking vibration signal, wherein the MFCC characteristic parameters are 2L× (T-4) and marked as VMFCC;
the steps (1) - (5) achieve extraction from the continuous tap signal to the mel-frequency cepstral coefficient (MFCC) feature set, the main technical route is shown in fig. 2.
(6) Training the knocking sound signals and the MFCC feature sets of the knocking vibration signals by using a hidden Markov algorithm (HMM), and establishing a knocking identification model of typical defects, as shown in FIG. 3; the method comprises the following steps:
(6.1) tap signal sample classification and classification: the knocking signal samples are divided into two sample sets for training and testing, each sample set comprises samples of L defect categories, and when the samples with the same defects are collected, the knocking force is manually adjusted so as to increase the diversity of the samples and the robustness of the model;
(6.2), initializing HMM model parameters: initializing a probability matrix a= [1,0,0,0,0,0], wherein the length of the matrix is equal to the state number S, and in the current example, s=46; initializing a transition probability matrix of S dimension, wherein the last element on the diagonal is assigned 1, and the remaining elements are assigned 0.5; and (3) dividing each training sample into N data fragments on average, classifying the training data by adopting a kmeans cluster function algorithm, and calculating the average value m (m 1, m2, mn), variance v (v 1, v2, vn) and weight w coefficient of each type after cluster analysis to realize the initial construction of the Gaussian mixture probability output function of each state. Wherein:
denoted p=p (m, v, x);
the state number is S, and the mixed Gaussian state function is Pmix
(6.3) calculating a given input signal sequence (MFCC feature set of the tapping sound signal and the tapping vibration signal), and respectively calculating forward probability alpha and backward probability beta, state transition probability xi at the moment of the current signal sequence t (i, j) and a mixed output probability gamma t (i)。
The forward probability alpha is calculated as follows:
a t (i) The forward probability of the sequence of the tap sound signal characteristics x is x1, x2, x, a1 (i) is the state si at the initial time 1, and the probability of the tap sound signal characteristics x is x1, which represents the tap sound at time t and the state si
Step 1, initial value: alpha 1 (i)＝π i *b 1 (x 1 )，π i An initial probability for a state si;
step 2, recursion calculation: for t=1, 2..t-1, there are
Step 3, the sequence of the sound signal features x of the tap at time T is x1, x2,..:
the backward probability beta is calculated as follows:
step 1, initial value: beta T (i)＝1；
Step 2, recursive calculation, for t=t-1, T-2,..1, there is
Step 3, the forward probability P (x1|μ) of the sequence of the tap sound signal characteristics x at time 1 being x1 is:
ξ t (i, j) represents the probability (1. Ltoreq.t.ltoreq.T+1) that T is in state i and t+1 is in state j:
γ t (i) At time T, the knocking sound signal is characterized by X, and the probability 1 is more than or equal to T and less than or equal to T+1) of the state i:
from time 1 to T, the expected value of the number of transitions from any other state to state i can be expressed as:
from time 1 to time T, fromThe expected value of the number of times state i transitions out can be expressed as:
from time 1 to T, the expected value of the number of transitions from state i to state j can be expressed as:
(6.4) calculating the HMM parameter state probability distribution pi of the sample set according to the maximum expectation estimation algorithm (EM algorithm) l State transition matrix a l And state emission matrix B l (l represents the first calculation of HMM parameters, and is initialized to be the 0 th time)
Π＝[π i ] N Wherein
A＝[a ij ] N×N Wherein
B＝[b j (k)] N×N Wherein
(6.5) calculating the first-1 order parameter (pi l-1 ，A l-1 ，B l-1 ) And the first order parameter (pi l ，A l ，B l ) Output probability P at the time l-1 And P l ；
Further, when (P l -P l-1 )/P l <When epsilon, stopping calculation, outputting (pi l ，A l ，B l ) Is a model parameter;
further, when (P l -P l-1 )/P l And (3) when epsilon is not less than, repeating the step (6.4) and the step (6.5) until the output probability convergence or iteration step number is met;
(6.6) repeating the steps (6.1) - (6.5) to obtain HMM model parameters of N sample sets respectively;
(7) According to the trained HMM defect model, the same or similar defects can be identified from the knocking signals by adopting a binary information fusion and correlation reverse evidence algorithm, the success rate of identification is high, and the main identification flow is shown in figure 4; the method comprises the following steps:
(7.1) repeating the steps (3) - (5) for the input tapping sound signal RS1 and the tapping vibration signal RS2 to obtain signal feature sequences RX1 and RX2 of the RS, and calculating the hidden markov model HMM1 (n 1 i ，A1 i ，B1 i )、HMM2(Π2 i ，A2 i ，B2 i ) (i=1, 2.,; M) relative probability output POut1 (pA 1 ,PA 2 ,..PA M ) And POut2 (pV) 1 ,PV 2 ,..PV M )；
(7.2) calculating the relative fusion probability POut (p 1 ,p 2 …,p M ) =pout1+pout2×w2, where w1 and w2 are weight coefficients of two HMM models, and w1+w2=1; for p1, p2, in POut, pM is ordered by fusion probability value size, with a maximum value of (p max Seq 1), the next largest value is (p sec Seq 2), a third value of (p thi Seq 3), seq represents the defect state number corresponding to the fusion probability;
(7.3) calculating a cross-correlation coefficient vector corrT1 of the tap sound signal RS1 and the tap sound reference signal of the seq1 defect after delaying T sampling points (t=1, 2,.,. T) respectively, and calculating a standard deviation of corrT1, and recording as corrSTD1;
(7.4) if corrSTD1 is more than or equal to 0.1, the final defect identification result is pmax=pmax, and seq=seq1, namely the defect corresponding to the current knocking sound signal RS1 is a Seq1 defect;
(7.5) if corrSTD1 < 0.1, calculating the tap sound signal RS1 is delayed by T sample points (t=1, 2,., T) then the cross-correlation coefficient vector corrT2 of the striking sound reference signal with the seq2 defect, and calculating the standard deviation of corrT2, and marking as corrSTD2;
(7.6) if corrSTD2 is more than or equal to 0.1, the final defect recognition result is pmax=pmax, and seq=seq2, namely the defect corresponding to the current knocking sound signal RS1 is a Seq2 defect;
(7.7) if corrSTD2 < 0.1, the final defect recognition result is pmax=pmax, and seq=seq 3, i.e. the defect corresponding to the current tapping sound signal RS1 is a Seq3 defect.
Examples:
fig. 5 shows a certain collected Q235 flat-panel beating sound signal, the sampling frequency f=51200hz, and 30 beating sounds are collected in total.
1. And (3) filtering:
the high-frequency random noise in the signal is removed by adopting a five-point three-time sliding filtering method, and as shown in fig. 6, the signal is smoothed to remove part of tip noise.
The db3 wavelet packet basis function is adopted to carry out 5-layer wavelet packet denoising on the smoothed signal, the Shannon entropy value of the wavelet packet coefficient is calculated as a denoising basis, soft threshold denoising is carried out in the global range, the threshold value Thr is generally 0.01-0.1, fig. 7 is an original signal of a section of knocking sound, and fig. 8 is a signal comparison graph (Thr=0.05) before and after wavelet packet denoising. Fig. 9 is a diagram of a single tap sound signal without normalization processing.
2. Signal endpoint detection and segmentation:
and calculating the zero crossing rate and short-time energy of each frame of signal, carrying out endpoint detection and audio frequency division on the knocking sound signal, removing the silence section signal between two adjacent knocks, and dividing the continuously collected knocking sound signal into equal-length audio frequency sections corresponding to the knocking times. The specific treatment process is as follows:
framing the knocking sound signal X in fig. 5, where the length of each frame is lenfram=256, and a certain overlapping section is formed between two adjacent frames, so that in order to improve the endpoint detection accuracy, the overlapping length of two adjacent frames is 88, and the frame increment is 80;
as shown in fig. 10 and 11, the short-time average energy En and the short-time zero-crossing rate Zn of the nth frame signal are calculated, and in order to reduce the influence of noise when the zero-crossing rate is calculated, when the absolute value of the amplitude difference between any two adjacent signal points is smaller than 0.03, the two points are considered as noise interference, and the zero-crossing rate calculation of the frame window is not included;
classifying states of the tapping sound signal X (mute (0), blur (1), tap (2) and end (3)) by using a "state machine" algorithm, and constructing a "state machine" loop program by taking the number of frames numfram=6141 of the tapping sound signal X as a loop condition:
2.1, initializing status=0, frame zero crossing rate threshold zt=5, length threshold lt=10 of the tapping sound signal, silence length threshold sil=10 of the tapping interval; the frame energy threshold value et1=min (std (En), max (En)/4), et2=min (std (En)/8, max (En)/16), std (En) is the standard deviation of the frame energy En of the tapping sound signal X;
2.2, when status=0 or 1, status=2 (tap) if En > Et1, otherwise status=1 (blur) when En > Et2 or Zn > Zt, otherwise status=0 (silence);
2.3, when status=2, if En > Et2 or Zn > Zt, status=2, otherwise the mute length is added by 1, and if the mute length is less than lt=10, the signal is still considered to be in the range of the knocks, when the mute length is less than Sil and the recorded signal length is less than Lt, the signal at this time is considered to be noise, the signal state jumps to silence (status=0), otherwise the knocks sound signal is considered to end, the signal state jumps to end state (status=3), and the knocks sound signal segment count is added by 1;
2.4, when status=3, if the tap sound signal segment count is less than the maximum possible number of speech segments MaxSec (Length (X)/(inc×lt)), the signal state jumps to silence (status=0), otherwise the "state machine" loop step jumps out;
2.5, terminating the "state machine" loop step when the loop count is equal to NumFram.
By using the above-mentioned "state machine" loop judgment method, a certain acquisition signal containing 30 taps is subjected to signal segmentation, and the result is shown in fig. 12.
And monitoring abnormal noise (including reverberation) possibly existing in the knocking sound signal by using the homology of the knocking vibration signal and the characteristic of being not influenced by environmental noise and adopting a correlation analysis method, and reminding in a result display step when the correlation coefficient is smaller than a threshold corrThr. Fig. 13 (a) and 13 (b) are respectively a tap sound signal and a tap vibration signal at the same tap, and fig. 14 is a fourier frequency domain curve of both, and formants are substantially identical.
Assuming that the two signal sequences x (N) and y (N) are both N, the signal length is N, x (N) is kept fixed, the signal y (N) is delayed by m sampling points, and the correlation coefficient between the x (N) sequence and the y (N) delay sequence is calculated according to the following formula and is denoted as Rxy (m).
Because the knocking sound signal sequence x and the knocking vibration signal y are damped oscillation waveforms, after the knocking sound signal x and the knocking vibration signal y under different acquisition conditions are intercepted by the same length, cross-correlation analysis is carried out: (1) Cross-correlation Rxy (m) of the same tapping sound signal and tapping vibration signal of the same component; (2) Cross correlation Rxy' (m) of the tapping sound signal and the tapping vibration signal of different components. When m=3000, as shown in fig. 15 and 16, the result shows that the cross correlation coefficient between the tapping signals periodically oscillates above and below the zero point with the change of the delay time, the standard deviation of the Rxy curves of the tapping sound signal and the tapping vibration signal of the same component is 0.159, and the standard deviation of the Rxy' curves of different components is 0.02, the difference is very obvious, and this characteristic can be used to verify whether the two signals originate from the same kind of components.
The noise condition in the knocking sound signal can be monitored by calculating the correlation coefficient, as shown in fig. 17, a certain degree of gaussian white noise is added to the original knocking sound signal, the correlation coefficient of the knocking sound signal delay sequence Sn and the vibration signal V is calculated according to the method, and as shown in fig. 18, the standard deviation is 0.148, and is reduced by 7% compared with the noise before adding.
3. MFCC feature set extraction:
when the signal pre-emphasis is performed in the step (5.1), the parameter μ=0.975 is taken according to the preferred result of the parameter.
In the above step (5.2), the frame window initial values n0=200, 256, 360, 512, 1024 are taken, the MFCC characteristic parameters at the corresponding N values are calculated, the 1 st and 7 th MFCC parameters are respectively denoted as MFCC1 and MFCC7, the discrete fraiche distance (frachet distance) between the MFCC7 curves at the different N values is calculated, and the calculated results are shown in table 1, where N1, N2 represent the different frame window size values.
Table 1 shows the discrete Frechet distances between the MFCC characteristics at different values of N
| FD(200，256) | FD(200，360) | FD(200，512) | FD(200，1024) | |
| MFCC1 | 11.3 | 6.3 | 3.5 | 4.4 |
| MFCC7 | 4.36 | 4.31 | 4.32 | 5.94 |
MFCC1 and MFCC7 correspond to the low frequency and high frequency portions of the tapping sound signal, respectively, and according to the comparison result in table 1, the low frequency and high frequency portions of the signal are considered, in this case, the frame window size value n=512 of the selection signal.
According to the above steps (5.3) - (5.10), the MFCC parameter AMFCC of the tapping sound signal can be obtained as shown in fig. 19, and the MFCC parameter VMFCC of the vibration signal can be obtained as well as shown in fig. 20.
4. Tap recognition model for typical defects:
initializing the parameters of the HMM model, wherein the number of HMM states S=6, each state consists of 3 mixed Gaussian models, the initial emission probability matrix is [0,0,0,0,0,1], and the initial transition probability matrix value is:
in the above step (6.5), the iteration convergence rate epsilon=5e-6. In this case, there are 9 sample sets corresponding to different defects, as shown in fig. 21, each sample set contains different numbers of knocking sound signals and knocking vibration signals, the sample set is divided into two parts, one part is used for training the HMM model, and the other part is used for checking the recognition rate of the model.
Based on the HMM model of the knocking sound signals, the correct recognition rate of the test sample (9 classes) is 100%, 83.3%, 100%, 82.8%, 100% respectively, and the total average recognition rate is 96.3%; based on Hmm model of the knocking vibration signal, the correct recognition rate of the test sample is 100%. Assuming that the signal intensity is 0dB, respectively superposing-30 dB Gaussian white noise on the knocking sound signal and the knocking vibration signal, and respectively setting the average recognition rate of the two models to the test sample to be 57.4% and 72.1%; after the fusion algorithm is adopted, the defect recognition rate is improved to 86.7%; and by adopting a fusion algorithm and a correlation reverse evidence algorithm, the defect recognition rate is improved to 98.1%.
When-25 dB is superimposed, the average recognition rate of the two models to the test sample is 54.8% and 61.0% respectively; after the fusion algorithm is adopted, the defect recognition rate is improved to 76.3%; and by adopting a fusion algorithm and a correlation reverse evidence algorithm, the defect recognition rate is improved to 86.3%.
In summary, from the bionics perspective, the invention simulates the process of "hearing" defects by a inspector with abundant experience through the knocking sound, identifies the knocking process of the inspection hammer 1, realizes the rapid detection and intelligent identification of the defects of the pressure-bearing metal member, and improves the accuracy and objectivity of the identification of the internal defects.
Finally, it should be noted that the above embodiments are merely representative examples of the present invention. Obviously, the invention is not limited to the above-described embodiments, but many variations are possible. Any simple modification, equivalent variation and modification of the above embodiments according to the technical substance of the present invention should be considered to be within the scope of the present invention.
Claims (5)
1. A method for identifying a metal member internal defect knocking detection and identification system, the method comprising the steps of:
(1) Collecting knocking sound signals and knocking vibration signals of the metal component, and recording the knocking force;
(2) Classifying and calibrating the collected knocking information samples according to the defect characteristics of the metal components;
(3) Preprocessing and filtering denoising are carried out on the knocked sound signal, wherein the preprocessing and filtering denoising comprises sliding filtering, standardization and abnormal point and noise removal of the signal;
(4) Calculating the zero crossing rate and short-time energy of each frame of signal, carrying out endpoint detection and audio frequency segment division on the knocking sound signal, eliminating useless silence segment signals, and dividing the continuously collected knocking sound signal into equal-length audio frequency segments corresponding to the knocking times;
(5) Performing MFCC feature extraction based on wavelet packet decomposition on the segmented audio signal in the step (4), and establishing an MFCC feature set of the knocking sound signal;
(6) Training the MFCC feature sets of the knocking sound signals and the knocking vibration signals by adopting a hidden Markov algorithm, and establishing an HMM defect model;
(7) Identifying the same or similar defects from the knocking signals by utilizing a binary information fusion and correlation reverse evidence algorithm according to the trained HMM defect model;
the step (4) specifically comprises the following steps:
(4.1) framing the knocked sound signal;
(4.2) calculating short-time average energy and short-time zero-crossing rate of a certain frame signal;
(4.3) dividing the knocking sound signals into a plurality of states, sequentially judging the states of the frames by using a state machine method, sequentially locating the coordinates of the beginning and the end of the knocking states by the signals, and dividing the knocking sound signals into a plurality of audio segments with equal lengths;
(4.4) monitoring abnormal noise possibly existing in the knocking sound signal by using a correlation analysis method by utilizing the homology of the knocking vibration signal and the characteristic of being not influenced by environmental noise;
the step (5) specifically comprises the following steps:
(5.1) carrying out signal pre-emphasis processing on the knocked sound signal, and improving a high-frequency part to highlight a high-frequency formant;
(5.2) framing the pre-emphasis-processed knocking sound signals, and adding a hamming window to each frame of signals;
(5.3) performing a fast fourier transform on the knocked sound signal;
(5.4) performing multi-layer wavelet packet decomposition on the knocked sound signal;
(5.5) defining a triangular filter bank;
(5.6) calculating the logarithmic energy of each filter bank output;
(5.7) introducing the logarithmic energy signal into discrete cosine transform, and obtaining MFCC coefficients through discrete cosine transform;
(5.8) calculating a first order difference of the MFCC coefficients;
(5.9) combining the MFCC coefficients and the first order differences of the MFCC coefficients to obtain MFCC characteristic parameters of the tapping sound signal;
(5.10), repeating the steps (5.1) - (5.9), and performing signal processing and characteristic parameter extraction on the knocking vibration signal to obtain MFCC characteristic parameters of the knocking vibration signal;
the step (7) specifically comprises the following steps:
(7.1) repeating the steps (3) - (5) for the input tapping sound signal RS1 and the tapping vibration signal RS2 to obtain signal feature sequences RX1 and RX2 of the RS, and calculating the hidden markov model HMM1 (n 1 i ，A1 i ，B1 i )、HMM2(Π2 i ，A2 i ，B2 i ) (i=1, 2.,; M) relative probability output POut1 (pA 1 ,PA 2 ,..PA M ) And POut2 (pV) 1 ,PV 2 ,..PV M )；
(7.2) calculating the relative fusion probability POut (p 1 ,p 2 …,p M ) =pout1+pout2×w2, where w1 and w2 are weight coefficients of two HMM models, and w1+w2=1; for p1, p2, in POut, pM is ordered by fusion probability value size, with a maximum value of (p max Seq 1), the next largest value is (p sec Seq 2), a third value of (p thi Seq 3), seq represents the defect state number corresponding to the fusion probability;
(7.3) calculating a cross-correlation coefficient vector corrT1 of the tap sound signal RS1 and the tap sound reference signal of the seq1 defect after delaying T sampling points (t=1, 2,.,. T) respectively, and calculating a standard deviation of corrT1, and recording as corrSTD1;
(7.4) if corrSTD1 is more than or equal to 0.1, the final defect identification result is pmax=pmax, and seq=seq1, namely the defect corresponding to the current knocking sound signal RS1 is a Seq1 defect;
(7.5) if corrSTD1 < 0.1, calculating the tap sound signal RS1 is delayed by T sample points (t=1, 2,., T) then the cross-correlation coefficient vector corrT2 of the striking sound reference signal with the seq2 defect, and calculating the standard deviation of corrT2, and marking as corrSTD2;
(7.6) if corrSTD2 is more than or equal to 0.1, the final defect recognition result is pmax=pmax, and seq=seq2, namely the defect corresponding to the current knocking sound signal RS1 is a Seq2 defect;
(7.7) if corrSTD2 < 0.1, the final defect recognition result is pmax=pmax, and seq=seq 3, i.e. the defect corresponding to the current tapping sound signal RS1 is a Seq3 defect.
2. The method for identifying a metal member internal defect knock detection and identification system according to claim 1, wherein in the step (2), the knock information sample is classified and calibrated in the following manner: and the inspector recognizes the defect of the knocked metal member by using an ear-hearing hand feeling mode, and adopts ultrasonic detection or digital ray technology to secondarily confirm the information sample which is suspected to be classified.
3. The method for identifying a metal member internal defect knock detection and identification system according to claim 1, wherein in the step (2), the knock information sample is classified and calibrated in the following manner: and manufacturing a batch of manual test pieces with prefabricated defects, simulating knocking information acquisition of a testing site, and manually changing the knocking force, changing the position of a knocking sound signal acquisition sensor and manufacturing environmental noise when the knocking information is acquired.
4. The method for recognizing the internal defect knocking detection and recognition system of the metal component according to claim 1, wherein in the step (3), a five-point three-time sliding filtering method is adopted to remove high-frequency random noise in the knocking sound signals, wavelet packet denoising is adopted to denoise the knocking sound signals, and the maximum value of the filtered signals is utilized to normalize the filtered knocking sound signals.
5. The method for identifying the internal defect knocking detection and identification system for the metal components according to claim 1, wherein the step (6) is specifically:
(6.1) dividing and classifying the knock signal samples;
(6.2) initializing HMM model parameters;
(6.3) calculating a given input signal sequence, and calculating forward and backward probabilities, state transition probabilities and mixed output probabilities of the current signal sequence moment;
(6.4) calculating the HMM parameter state probability distribution, the state transition matrix and the state emission matrix of the sample set according to the maximum expectation estimation algorithm;
(6.5) calculating the output probability of the knocking characteristic signal sequence at a certain time parameter and the previous time of the time parameter by using a Viterbi algorithm;
(6.6) repeating the steps (6.1) - (6.5) to obtain HMM model parameters of a plurality of sample sets.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010862602.3A CN111983020B (en) | 2020-08-25 | 2020-08-25 | System and method for detecting and identifying internal defects of metal component through knocking |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010862602.3A CN111983020B (en) | 2020-08-25 | 2020-08-25 | System and method for detecting and identifying internal defects of metal component through knocking |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111983020A CN111983020A (en) | 2020-11-24 |
| CN111983020B true CN111983020B (en) | 2023-08-22 |
Family
ID=73444060
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010862602.3A Active CN111983020B (en) | 2020-08-25 | 2020-08-25 | System and method for detecting and identifying internal defects of metal component through knocking |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111983020B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113109011A (en) * | 2021-03-29 | 2021-07-13 | 太仓众擎机械设备有限公司 | Comprehensive detection device for steel structure frame |
| CN113075296A (en) * | 2021-04-01 | 2021-07-06 | 湖南翰坤实业有限公司 | Method and device for detecting safety of outer wall structure based on sound wave detection and BIM model |
| CN113588782A (en) * | 2021-07-01 | 2021-11-02 | 广州珠江天然气发电有限公司 | Nondestructive inspection method, device and equipment based on audio analysis |
| CN114324580A (en) * | 2021-12-03 | 2022-04-12 | 西安交通大学 | Intelligent knocking detection method and system for structural defects |
| CN114371222B (en) * | 2022-01-17 | 2023-11-03 | 上海电子信息职业技术学院 | Ceramic product structure defect acoustic wave detection method based on MODEPT coefficient square entropy and RF |
| CN114429771A (en) * | 2022-04-02 | 2022-05-03 | 武汉地震工程研究院有限公司 | Intelligent detection method and system for bonding defects of steel beam and CFRP (carbon fiber reinforced plastics) plate |
| CN115032272B (en) * | 2022-06-08 | 2023-04-07 | 广西大学 | Structural defect detection device and method integrating knocking force signal and sound signal |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH10300730A (en) * | 1997-04-24 | 1998-11-13 | Mitsubishi Electric Corp | Percussive sound judging device |
| JP2000074889A (en) * | 1998-09-02 | 2000-03-14 | Mitsubishi Electric Corp | Hammering determining device |
| CN101701935A (en) * | 2009-11-24 | 2010-05-05 | 东北大学 | Refractory brick inside quality online detection device |
| CN106645435A (en) * | 2016-12-09 | 2017-05-10 | 北京工业大学 | Method for detecting inclusion defects in metal component |
| JP2017083330A (en) * | 2015-10-29 | 2017-05-18 | 古河機械金属株式会社 | Hammering inspection system and hammering inspection method |
| CN106706119A (en) * | 2016-12-15 | 2017-05-24 | 北方工业大学 | Vibration source identification method and system based on signal frequency domain characteristics |
| CN108956770A (en) * | 2018-07-04 | 2018-12-07 | 北京林业大学 | A kind of intelligence hammer and screening method for the preliminary screening of historic building fire prevention internal flaw |
| CN109387565A (en) * | 2018-10-12 | 2019-02-26 | 山东理工大学 | A method of brake block internal flaw is detected by analysis voice signal |
| CN110038813A (en) * | 2019-04-01 | 2019-07-23 | 山东理工大学 | A kind of disc brake pad Inner Defect Testing equipment based on sound |
| CN110045015A (en) * | 2019-04-18 | 2019-07-23 | 河海大学 | A kind of concrete structure Inner Defect Testing method based on deep learning |
| CN111289611A (en) * | 2020-01-19 | 2020-06-16 | 河南科技大学 | Composite material defect type detection method and device |
-
2020
- 2020-08-25 CN CN202010862602.3A patent/CN111983020B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH10300730A (en) * | 1997-04-24 | 1998-11-13 | Mitsubishi Electric Corp | Percussive sound judging device |
| JP2000074889A (en) * | 1998-09-02 | 2000-03-14 | Mitsubishi Electric Corp | Hammering determining device |
| CN101701935A (en) * | 2009-11-24 | 2010-05-05 | 东北大学 | Refractory brick inside quality online detection device |
| JP2017083330A (en) * | 2015-10-29 | 2017-05-18 | 古河機械金属株式会社 | Hammering inspection system and hammering inspection method |
| CN106645435A (en) * | 2016-12-09 | 2017-05-10 | 北京工业大学 | Method for detecting inclusion defects in metal component |
| CN106706119A (en) * | 2016-12-15 | 2017-05-24 | 北方工业大学 | Vibration source identification method and system based on signal frequency domain characteristics |
| CN108956770A (en) * | 2018-07-04 | 2018-12-07 | 北京林业大学 | A kind of intelligence hammer and screening method for the preliminary screening of historic building fire prevention internal flaw |
| CN109387565A (en) * | 2018-10-12 | 2019-02-26 | 山东理工大学 | A method of brake block internal flaw is detected by analysis voice signal |
| CN110038813A (en) * | 2019-04-01 | 2019-07-23 | 山东理工大学 | A kind of disc brake pad Inner Defect Testing equipment based on sound |
| CN110045015A (en) * | 2019-04-18 | 2019-07-23 | 河海大学 | A kind of concrete structure Inner Defect Testing method based on deep learning |
| CN111289611A (en) * | 2020-01-19 | 2020-06-16 | 河南科技大学 | Composite material defect type detection method and device |
Non-Patent Citations (1)
| Title |
|---|
| 曾向阳 等.《声信号处理基础》.西北工业大学出版社,2015,第136-138页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111983020A (en) | 2020-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111983020B (en) | System and method for detecting and identifying internal defects of metal component through knocking | |
| Wu et al. | Robust endpoint detection algorithm based on the adaptive band-partitioning spectral entropy in adverse environments | |
| CN110045016B (en) | Tunnel lining nondestructive testing method based on audio frequency analysis | |
| CN109782274B (en) | Water damage identification method based on time-frequency statistical characteristics of ground penetrating radar signals | |
| CN102097095A (en) | Speech endpoint detecting method and device | |
| CN101477801A (en) | Method for detecting and eliminating pulse noise in digital audio signal | |
| CN111487315A (en) | Audio frequency nondestructive testing method for tunnel lining thickness and void | |
| CN112157368B (en) | Laser non-penetration welding seam penetration nondestructive testing method | |
| CN110568073B (en) | Method for picking up impact signal in noise environment | |
| CN110412129A (en) | A kind of microscopic defect ultrasound detection signal processing method considering surface roughness | |
| WO2016004757A1 (en) | Noise detection method and apparatus | |
| CN113345399A (en) | Method for monitoring sound of machine equipment in strong noise environment | |
| JP2002517752A5 (en) | ||
| CN110455923B (en) | Rapid evaluation method for anchor rod anchoring quality grade | |
| CN116401571A (en) | Concrete filled steel tube abnormal shape void identification method based on knocking sound wave and MiniRoccket | |
| CN109884483A (en) | Insulating tube type busbar shelf depreciation acoustics on-line monitoring method and device | |
| Wu et al. | A pitch-based method for the estimation of short reverberation time | |
| CN117191956A (en) | Acoustic emission-based titanium alloy stress corrosion damage classification method and apparatus | |
| CN104200815B (en) | A kind of audio-frequency noise real-time detection method based on correlation analysis | |
| JP2006038478A (en) | Quality determination method and device for specimen | |
| Wu et al. | Acoustic tunnel lining cavity detection using cepstral coefficients with optimized filter bank | |
| JPH11304773A (en) | Method for discriminating peeling of external wall tile | |
| Baek et al. | In-process noise inspection system for product fault detection in a loud shop-floor environment | |
| Yang et al. | Research on Voiceprint recognition method of buried drainage pipe based on MFCC and GMM-HMM | |
| CN117783294B (en) | Acoustic emission energy entropy-based wood damage dynamic detection 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 |