CN114088817A - Deep learning flat ceramic membrane ultrasonic defect detection method based on deep features - Google Patents

Deep learning flat ceramic membrane ultrasonic defect detection method based on deep features Download PDF

Info

Publication number
CN114088817A
CN114088817A CN202111259179.9A CN202111259179A CN114088817A CN 114088817 A CN114088817 A CN 114088817A CN 202111259179 A CN202111259179 A CN 202111259179A CN 114088817 A CN114088817 A CN 114088817A
Authority
CN
China
Prior art keywords
ultrasonic
ceramic membrane
defect
signals
neural network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202111259179.9A
Other languages
Chinese (zh)
Other versions
CN114088817B (en
Inventor
孙进
雷震霆
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yangzhou University
Original Assignee
Yangzhou University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Yangzhou University filed Critical Yangzhou University
Priority to CN202111259179.9A priority Critical patent/CN114088817B/en
Publication of CN114088817A publication Critical patent/CN114088817A/en
Application granted granted Critical
Publication of CN114088817B publication Critical patent/CN114088817B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
    • G01N29/4409—Processing the detected response signal, e.g. electronic circuits specially adapted therefor by comparison
    • G01N29/4418—Processing the detected response signal, e.g. electronic circuits specially adapted therefor by comparison with a model, e.g. best-fit, regression analysis
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
    • G01N29/4481—Neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Analytical Chemistry (AREA)
  • General Engineering & Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Signal Processing (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)

Abstract

The invention belongs to the field of ultrasonic defect detection, and particularly relates to a deep learning flat ceramic membrane ultrasonic defect detection method based on deep features, which comprises the following steps: step 1: building a device, collecting ultrasonic signals and building a detection model; step 2: extracting deep features of the ultrasonic defect signal; and step 3: training a cavity convolutional neural network; and 4, step 4: defining the defect type and classifying the defects. The invention carries out noise reduction processing through continuous wavelet transformation, converts signals into time spectrogram, realizes the extraction of ultrasonic signal deep level, takes the time spectrogram gray level image as the input of a cavity convolution neural network, analyzes the time spectrogram gray level image through the cavity convolution neural network and judges the defects, and can accurately, quickly and objectively detect the internal defects of the flat ceramic membrane.

Description

Deep learning flat ceramic membrane ultrasonic defect detection method based on deep features
Technical Field
The invention belongs to the technical field of ultrasonic defect detection, and particularly relates to a deep learning flat ceramic membrane ultrasonic defect detection method based on deep features.
Background
The flat ceramic membrane is prepared from inorganic materials such as alumina and the like through a high-temperature sintering process, and is widely applied to the fields of seawater desalination, sewage purification and the like due to the advantages of high water permeability, strong pollution resistance, easy flushing of surface attachments, small working pressure and the like. The defects of pores, slag falling and the like of the flat ceramic membrane in the high-temperature calcination stage need to be screened and removed in the production process. The traditional detection method for the internal defects of the flat ceramic membrane is lamp detection, the internal defects of the flat ceramic membrane are detected in a water detection mode, the detection result is inaccurate, the error is large, and the long-term detection of detection personnel is not facilitated.
In 2020, Qinping et al decomposed the noise-added signal by an empirical mode decomposition method in an invention patent of a reconstruction extraction method (publication number CN111795931A) for laser ultrasonic defect detection diffraction echo signals to generate a plurality of Intrinsic Mode Function (IMF) components, ordered the IMF components according to frequency, calculated the discrete degree of each IMF component, and selected the corresponding IMF component according to the discrete degree distribution and IMF component relationship for reconstruction to obtain a reconstruction signal. The identification of the signal finally depends on the identification of a detection person, and the identification detection has an error of artificial detection. In 2020, Weishifei et al used an ultrasonic flaw detector to scan the interior of a casting blank in the invention patent of a casting blank internal flaw detection method (publication number CN 112147223A); the sensitivity gain of the ultrasonic flaw detector is adjusted by using the sensitivity gain value at the central thickness of the casting blank, so that the influence of excessive grass-shaped waves on the accuracy of the detection result in the detection process is restrained. The method simply processes the collected signals through the functions in the ultrasonic flaw detector, and cannot deeply provide useful information of the detected signals. In 2021, Zunou et al invented a stainless steel weld ultrasonic defect detection method (publication No. CN112232400A) based on depth feature fusion by fusing manually extracted shallow features and CNN extracted deep features and using support vector machine to classify, thereby realizing classification and identification of stainless steel defects. The time resolution and the frequency resolution cannot be obtained by using short-time Fourier transform, so that the signal extraction is lacked in the deep feature extraction, and the identification accuracy is reduced.
In summary, although the conventional method can detect the internal defects of the flat ceramic membrane, the conventional method has the problems of low detection accuracy, low detection rate, incapability of deep characterization, and the like.
Disclosure of Invention
In order to overcome the defects of the prior art and the method, the invention provides a deep learning flat ceramic membrane ultrasonic defect detection method based on deep features, and the method can effectively improve the effect of flat ceramic membrane defect detection.
The technical solution for realizing the purpose of the invention is as follows:
a deep learning flat ceramic membrane ultrasonic defect detection method based on deep features comprises the following steps:
step 1: building a device, collecting ultrasonic signals and building a detection model;
step 2: extracting deep features of the ultrasonic defect signal;
and step 3: training a cavity convolutional neural network;
and 4, step 4: defining the defect type and classifying the defects.
Further, the step 1 specifically includes:
the device comprises an ultrasonic detection moving device and an ultrasonic probe clamping device, a ceramic membrane to be detected is placed on the ultrasonic detection moving device, the ultrasonic probe is clamped by the ultrasonic probe clamping device, the ceramic membrane is adjusted until the ultrasonic probe is in the optimal position, so that the ultrasonic probe is spread over the surface of the whole ceramic membrane, a plurality of flat ceramic membranes with scratches, fractures and holes are replaced, the flat ceramic membranes with scratches, fractures and holes are detected, and the defect detection signals of the flat ceramic membranes at each time are collected,
establishing a detection model, wherein the scattering effect generated when the ultrasonic waves impact the defect depends on the relation between the wavelength lambda and the defect diameter d, and Rayleigh scattering occurs when the following conditions are met:
Figure BDA0003324907220000021
wherein < represents less than one tenth of wavelength λ, α is the ultrasonic attenuation, β is the average grain volume, s is the optical propagation rate of the material, f the frequency of the incident light, and → represents the inverse ratio.
Establishing a mathematical model h (t) of the monopulse ultrasonic echo, as shown in formula (2):
h(t)＝s(t)+δe(t) (2)
Figure BDA0003324907220000022
where t is 0,1, …, s (t) denotes a defect detection signal, δ e (t) denotes a noise signal, which is represented by a distribution σ2(0,1) Gaussian random variable composition,
Figure BDA0003324907220000031
is a reflecting surface, takes a value of zero, tau is the thickness of the ceramic membrane to be detected, beta is the amplitude of the echo peak value, fcPsi is the center frequency of the echo, and psi is the bandwidth coefficient;
according to the Thompson-Gray measurement model, the transducer occupies VFThe amplitude of the echo signal received by the defect of the region is modeled as follows:
Figure BDA0003324907220000032
where Φ is the amplitude of the echo signal, P is the electrical power of the propagating cable, i represents the wave velocity of the medium, ω represents the angular frequency of the ultrasonic wave, δpFor detecting density differences between media and defects, UiRepresenting the particle polarization, superscript o representing the defect-free field variable, δcIs the similarity difference of elastic constants, UijExpressed as scatterer density.
Neglecting scatterer density fluctuations, then:
Figure BDA0003324907220000033
as can be seen from equation (5), if a defect occurs in the propagation path of the received echo, P > 0, the amplitude Φ of the echo signal decreases as P increases.
Further, the ultrasonic detection moving device comprises a supporting plate, a ceramic membrane placing table, four supporting columns, a stabilizing plate, a Y-direction advancing device, an X-direction advancing device, four supporting nuts and a moving platform, wherein the supporting plate is connected with the four supporting columns through threads, the ceramic membrane placing table is matched with the four supporting columns through apertures, the ceramic membrane placing table is supported by the four supporting nuts and the four supporting columns through threads, the supporting nuts are connected with the corresponding supporting columns through threads, the supporting nuts are positioned between the ceramic membrane placing table and the supporting plate and can drive the ceramic membrane placing table to move up and down through the left-right rotation of the supporting nuts, the Y-direction advancing device can drive the moving platform to advance and retreat in the Y direction, so that the probe can advance and retreat in the Y direction, and the X-direction advancing device can drive the moving platform to advance and retreat in the X direction, and then realize advancing and retreating of the X direction of probe, the top of four support columns passes the steadying plate and makes through the support of pressing from both sides tight piece the ultrasonic testing mobile device is more stable, ultrasonic probe clamping device includes probe centre gripping backup pad, grip block, guide rail, twist grip, probe locating hole and presss from both sides tight piece, gathers ultrasonic signal and specifically includes:
when carrying out defect detection to dull and stereotyped ceramic membrane, put the probe locating hole with ultrasonic probe, make the grip block add through the twist grip and hold ultrasonic probe, install probe centre gripping backup pad to X direction advancing device again, the ceramic membrane that will detect is put the ceramic membrane and is placed the platform, through rotatory support nut, adjust ceramic membrane to ultrasonic probe and be the best position, let ultrasonic probe distribute all over whole ceramic membrane surface through using X direction advancing device and Y direction advancing device, it has the mar to change, the fracture, a plurality of dull and stereotyped ceramic membranes of hole defect, to having the mar, the fracture, the dull and stereotyped ceramic membrane of hole detects, and collect every time dull and stereotyped ceramic membrane defect detected signal.
Further, step 2 specifically comprises:
the shallow layer characteristic is that after the collected signals are directly subjected to simple noise reduction processing, the signals are used as the input of a cavity convolution neural network, and the existing signals are subjected to characteristic extraction and classification by the cavity convolution neural network. The accuracy of network identification can be greatly reduced only by means of extracting features from signals by the void convolutional neural network; compared with the extraction of shallow features, the extraction of deep features requires a plurality of times of image transformation, including a series of image processing such as graying, ROI segmentation, Canny operator extraction and the like, and more deep useful information in the acquired signals is highlighted. These features are not available through the network itself for the hole convolutional neural network. The deep features are combined with a series of image transformation and extraction features of the cavity convolution neural network to increase the accuracy of model identification. Under the same cavity convolution neural network model, the image extracted by the deep layer features is easier to identify than the image extracted by the shallow layer features.
The acquired ultrasonic signals are one-dimensional time series signals, deep features of the signals acquired in the step 1 are extracted, the signals are denoised by adopting a continuous wavelet transform method, useful defect signals are removed from the signals, then the denoised time series signals are converted into a two-dimensional scale map, all the two-dimensional scale maps are collected, and the features of the signals in a time-frequency domain can be seen from the two-dimensional scale map. Graying and segmenting the collected two-dimensional scale map into ROI, filtering the image, extracting edges through Canny operator, extracting the characteristic edges of the two-dimensional scale map through the Canny operator, highlighting the defect edges, and improving the accuracy of subsequent model identification.
The preprocessing is to perform graying processing on the acquired picture, reduce the picture memory and accelerate the training and recognition speed of the hole convolution neural network. And dividing the ROI to determine the positions of the defect features of the two-dimensional scale map. And filtering the image to reduce the filtering of the image. The data after each A scanning is converted into a two-dimensional scale map, so that useful defect signals can be provided deeply, and the two-dimensional scale map can be used as the input of the cavity convolutional neural network, so that the identification of the cavity convolutional neural network can be better in an optimal state.
Further, step 3 specifically comprises:
the cavity convolution neural network comprises a convolution layer, a pooling layer, an activation function, a batch normalization layer and an attention mechanism, wherein the cavity convolution is to expand the field of the conventional 3 x 3 convolution kernel to 5 x 5 by changing the cavity, and other expanded cavities are filled with zeros. Under the condition of the same calculation amount, compared with the traditional convolution kernel, the traditional hole convolution kernel can obtain a larger receptive field and obtain more useful information.
Further, step 4 specifically includes:
dividing the types of the ceramic membrane defects into scratches, fractures and holes, establishing a universal defect classification formula model, setting a classification task as N, expressing the embedding of the classification task as x, and evaluating the probability of the y classification by using a Softmax function, wherein the formula (6) is as follows:
Figure BDA0003324907220000051
if x belongs to the ζ -th class, the corresponding Softmax loss function is expressed as:
Figure BDA0003324907220000052
wherein (y)1,…,yk,…,yN) Is a label of classification, ykAnd (5) taking the value as 0 or 1, and outputting the probability through the last Softmax to obtain a detection result.
Compared with the prior art, the invention has the remarkable advantages that:
the invention provides a deep learning flat ceramic membrane ultrasonic defect detection method based on deep features, which is characterized in that noise reduction is carried out through continuous wavelet transformation, signals are converted into time spectrogram, the extraction of the deep level of ultrasonic signals is realized, the time spectrogram gray graph is used as the input of a cavity convolution neural network, the time spectrogram gray graph is analyzed through the cavity convolution neural network and the defects are judged, and the internal defects of the flat ceramic membrane can be accurately, quickly and objectively detected.
Drawings
FIG. 1 is a schematic view of a positive three-axis view of an ultrasonic testing mobile device.
Fig. 2 is a schematic front view of an ultrasonic testing mobile device.
Fig. 3 is a top view of an ultrasonic probe holder.
Fig. 4 is a flow chart of ultrasonic testing.
Fig. 5 is a schematic diagram of the structure of a convolutional neural network.
FIG. 6 is a variation of the convolution kernel of the hole convolution neural network.
FIG. 7 is a flow chart of a method for detecting ultrasonic defects of a flat ceramic membrane based on a void convolutional neural network fused with deep features.
FIG. 8 is a flow chart of shallow and deep extraction features.
The device comprises a support plate 1, a ceramic membrane placing table 2, a support column 3, a stabilizing plate 4, a moving device in the direction 5-Y, a moving device in the direction 6-X, a support nut 7, a probe clamping support plate 8, a clamping block 9, a guide rail 10, a rotating handle 11, a probe positioning hole 12, a clamping block 13 and a moving platform 14.
Detailed Description
The following will further describe the specific implementation of the present invention based on the detection of the defect of the planar ceramic membrane of the void convolutional neural network fused with deep features with reference to fig. 1, fig. 2, fig. 3 and fig. 7.
As shown in FIG. 1, the defect detection of the cavity convolution neural network flat ceramic membrane based on the fusion deep layer characteristics comprises the following steps:
step 1: and (3) setting up a device, acquiring an ultrasonic signal and establishing a detection model.
With reference to fig. 1-2, the supporting plate 1 is in threaded connection with the four supporting columns 3, the ceramic membrane placing table 2 is matched with the four supporting columns 3 through the apertures, the supporting screw caps 7 are matched with the four supporting columns 3 through the threads to support the ceramic membrane placing table 2, and the ceramic membrane placing table 2 is driven to move up and down through the left and right rotation of the supporting screw caps 7. The Y-direction advancing device 5 drives the screw rod to rotate forward and backward through the forward and backward rotation of the motor, and the forward and backward rotation of the screw rod drives the moving platform 14 to advance and retreat in the Y direction, so that the probe can advance and retreat in the Y direction; similarly, the X-direction advancing device 6 is arranged in the same manner as the Y-direction advancing device 5. The two stabilising plates 4 are supported by the clamping blocks 13 to further stabilise the overall arrangement.
Referring to fig. 3, when defect detection is performed on a flat ceramic membrane, an ultrasonic probe is placed in a probe positioning hole 12, the ultrasonic probe is clamped by a clamping block 9 through a rotating handle 11, a probe clamping support plate 9 is installed on an X-direction advancing device 6, the detected ceramic membrane is placed on a ceramic membrane placing table 2, a support nut 7 is rotated to adjust the ceramic membrane until the ultrasonic probe is in the optimal position, the ultrasonic probe is spread on the surface of the whole ceramic membrane through the X-direction advancing device 6 and the Y-direction advancing device 5, a plurality of flat ceramic membranes with scratches, fractures and hole defects are replaced, the flat ceramic membranes with the scratches, the fractures and the holes are detected, and defect detection signals of the flat ceramic membranes at each time are collected.
And establishing a detection model, wherein the scattering effect generated when the ultrasonic waves strike the defect depends on the relation between the wavelength lambda and the diameter d of the defect. Rayleigh scattering occurs when the following conditions are met:
Figure BDA0003324907220000071
wherein < represents less than one tenth of wavelength λ, α is the ultrasonic attenuation, β is the average grain volume, s is the optical propagation rate of the material, f the frequency of the incident light, and → represents the inverse ratio.
Establishing a mathematical model h (t) of the monopulse ultrasonic echo, as shown in formula (2):
h(t)＝s(t)+δe(t) (2)
Figure BDA0003324907220000072
where t is 0,1, …, s (t) denotes a defect detection signal, δ e (t) denotes a noise signal, which is represented by a distribution σ2(0,1) Gaussian random variable composition,
Figure BDA0003324907220000073
is a reflecting surface, and takes values generallyZero, τ is the thickness of the sample of the plate ceramic membrane, β is the amplitude of the echo peak, fcPsi is the bandwidth factor for the center frequency of the echo.
According to the Thompson-Gray measurement model, the transducer occupies VFThe amplitude of the echo signal received by a defect of a region can be modeled as follows:
Figure BDA0003324907220000074
where Φ is the amplitude of the echo signal, P is the electrical power of the propagating cable, i represents the wave velocity of the medium, ω represents the angular frequency of the ultrasonic wave, δpFor detecting density differences between media and defects, UiRepresenting the particle polarization, superscript o representing the defect-free field variable, δcIs the similarity difference of elastic constants, UijExpressed as scatterer density.
If scatterer density fluctuations are negligible:
Figure BDA0003324907220000075
as can be seen from equation (5), if a defect occurs in the propagation path of the received echo, P > 0, the amplitude Φ of the echo signal decreases as P increases.
Step 2: and extracting deep features of the ultrasonic defect signal.
As shown in fig. 8, the shallow feature is obtained by directly passing the acquired signal through simple noise reduction processing, and then taking the signal as an input of a hole convolutional neural network, and extracting and classifying the existing signal by means of the hole convolutional neural network. The accuracy of network identification can be greatly reduced only by means of extracting features from signals by the void convolutional neural network; compared with the extraction of shallow features, the extraction of deep features requires a plurality of times of image transformation, including a series of image processing such as graying, ROI segmentation, Canny operator extraction and the like, and more deep useful information in the acquired signals is highlighted. These features are not available through the network itself for the hole convolutional neural network. The deep features are combined with a series of image transformation and extraction features of the cavity convolution neural network to increase the accuracy of model identification. Under the same cavity convolution neural network model, the image extracted by the deep layer features is easier to identify than the image extracted by the shallow layer features.
The acquired ultrasonic signals are one-dimensional time series signals, deep features of the signals acquired in the step 1 are extracted, the signals are denoised by adopting a continuous wavelet transform method, useful defect signals are removed from the signals, then the denoised time series signals are converted into a two-dimensional scale map, all the two-dimensional scale maps are collected, and the features of the signals in a time-frequency domain can be seen from the two-dimensional scale map. Graying and segmenting the collected two-dimensional scale map into ROI, filtering the image, extracting edges through Canny operator, extracting the characteristic edges of the two-dimensional scale map through the Canny operator, highlighting the defect edges, and improving the accuracy of subsequent model identification.
The preprocessing is to perform graying processing on the acquired picture, reduce the picture memory and accelerate the training and recognition speed of the hole convolution neural network. And dividing the ROI to determine the positions of the defect features of the two-dimensional scale map. And filtering the image to reduce the filtering of the image. The data after each A scanning is converted into a two-dimensional scale map, so that useful defect signals can be provided deeply, and the two-dimensional scale map can be used as the input of the cavity convolutional neural network, so that the identification of the cavity convolutional neural network can be better in an optimal state.
And step 3: and training a hole convolutional neural network.
As shown in FIG. 5, the hole convolutional neural network includes convolutional layers, pooling layers, activation functions, Batchnorm, attention mechanism. The hole convolutional neural network includes convolutional layers, pooling layers, activation functions, batcnorm, attention mechanisms, and the like. As shown in fig. 6, the hole convolution is to expand the field of the conventional 3 × 3 convolution kernel to the field of the 5 × 5 convolution kernel by changing the hole from the original 3 × 3 convolution kernel, and fill the other expanded holes with zeros. Under the condition of the same calculation amount, compared with the traditional convolution kernel, the traditional hole convolution kernel can obtain a larger receptive field and obtain more useful information.
And taking the gray-scale images of the scratches, the fractures and the holes as input of the cavity convolution neural network, extracting the characteristics of the gray-scale images through the convolution layer, and reducing the dimensionality and the size of the images through the pooling layer. The model of the cavity convolution neural network can be trained after at least 50 iterations.
And 4, step 4: defining the defect type and classifying the defects.
The invention divides the defect types of the ceramic membrane into scratch, fracture and hole. Establishing a general defect classification formula model, setting a classification task as N, wherein the embedding of the classification task is represented as x, and the probability of the y-th classification can be evaluated by a Softmax function, as shown in a formula (6):
Figure BDA0003324907220000091
if x belongs to the ζ -th class, the corresponding Softmax loss function can be expressed as:
Figure BDA0003324907220000092
wherein (y)1,…,yk,…,yN) Is a label of classification, ykAnd (5) taking the value as 0 or 1, and outputting the probability through the last Softmax to obtain a detection result.
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (6)

1. A deep learning flat ceramic membrane ultrasonic defect detection method based on deep features is characterized by comprising the following steps:
step 1: building a device, collecting ultrasonic signals and building a detection model;
step 2: extracting deep features of the ultrasonic defect signal;
and step 3: training a cavity convolutional neural network;
and 4, step 4: defining the defect type and classifying the defects.
2. The method according to claim 1, wherein step 1 specifically comprises:
the device comprises an ultrasonic detection moving device and an ultrasonic probe clamping device, a ceramic membrane to be detected is placed on the ultrasonic detection moving device, the ultrasonic probe is clamped by the ultrasonic probe clamping device, the ceramic membrane is adjusted until the ultrasonic probe is in the optimal position, so that the ultrasonic probe is spread over the surface of the whole ceramic membrane, a plurality of flat ceramic membranes with scratches, fractures and holes are replaced, the flat ceramic membranes with scratches, fractures and holes are detected, and the defect detection signals of the flat ceramic membranes at each time are collected,
establishing a detection model, wherein the scattering effect generated when the ultrasonic waves impact the defect depends on the relation between the wavelength lambda and the defect diameter d, and Rayleigh scattering occurs when the following conditions are met:
Figure FDA0003324907210000011
wherein < represents less than one tenth of wavelength lambda, alpha is ultrasonic attenuation, beta is average crystal grain volume, s is material optical propagation rate, f frequency of incident ray, → represents inverse ratio,
establishing a mathematical model h (t) of the monopulse ultrasonic echo, as shown in formula (2):
h(t)＝s(t)+δe(t) (2)
Figure FDA0003324907210000012
where t is 0,1, …, s (t) denotes a defect detection signal, δ e (t) denotes a noise signal, which is represented by a distribution σ2(0,1) Gaussian random variable composition,
Figure FDA0003324907210000013
is a reflecting surface, takes a value of zero, tau is the thickness of the ceramic membrane to be detected, beta is the amplitude of the echo peak value, fcPsi is the center frequency of the echo, and psi is the bandwidth coefficient;
according to the Thompson-Gray measurement model, the transducer occupies VFThe amplitude of the echo signal received by the defect of the region is modeled as follows:
Figure FDA0003324907210000021
where Φ is the amplitude of the echo signal, P is the electrical power of the propagating cable, i represents the wave velocity of the medium, ω represents the angular frequency of the ultrasonic wave, δpFor detecting density differences between media and defects, UiRepresenting the particle polarization, superscript o representing the defect-free field variable, δcIs the similarity difference of elastic constants, UijExpressed as a density of scatterers,
neglecting scatterer density fluctuations, then:
Figure FDA0003324907210000022
as can be seen from equation (5), if a defect occurs in the propagation path of the received echo, P > 0, the amplitude Φ of the echo signal decreases as P increases.
3. The method according to claim 2, wherein the ultrasonic detection moving device comprises a support plate (1), a ceramic membrane placing table (2), four support columns (3), a stabilizing plate (4), a Y-direction advancing device (5), an X-direction advancing device (6), four support nuts (7) and a moving platform (14), the support plate (1) is connected with the four support columns (3) through threads, the ceramic membrane placing table (2) is matched with the four support columns (3) through apertures, the ceramic membrane placing table (2) is supported by the four support nuts (7) and the four support columns (3) through threads, the support nuts (7) are connected with the corresponding support columns (3) through threads, the support nuts (7) are positioned between the ceramic membrane placing table (2) and the support plate (1), and the ceramic membrane placing table (2) can be driven to move up and down through the left and right rotation of the support nuts (7), y direction advancing device (5) can drive moving platform (14) and advance and retreat in the Y direction, and then realize advancing and retreating of the Y direction of probe, and X direction advancing device (6) can drive moving platform (14) and advance and retreat in the X direction, and then realize advancing and retreating of the X direction of probe, and the top of four support columns (3) passes stabilizer plate (4) and makes through the support of pressing from both sides tight piece (13) ultrasonic testing moving device is more stable, ultrasonic probe clamping device includes probe centre gripping backup pad (8), grip block (9), guide rail (10), twist grip (11), probe locating hole (12) and presss from both sides tight piece (13), gathers ultrasonic signal and specifically includes:
when carrying out defect detection to dull and stereotyped ceramic membrane, put probe locating hole (12) with ultrasonic probe, make grip block (9) add through rotatory handle (11) and hold ultrasonic probe, install X direction advancing device (6) with probe centre gripping backup pad (9) again, the ceramic membrane that will detect is put ceramic membrane and is placed on platform (2), through rotatory support nut (7), adjust ceramic membrane to ultrasonic probe is the best position till, let ultrasonic probe distribute over whole ceramic membrane surface through using X direction advancing device (6) and Y direction advancing device (5), the change has the mar, the fracture, a plurality of dull and stereotyped ceramic membranes of hole defect, to having the mar, the fracture, the dull and stereotyped ceramic membrane of hole detects, and collect dull and stereotyped ceramic membrane defect detected signal every time.
4. The method according to claim 3, wherein step 2 is specifically:
the acquired ultrasonic signals are one-dimensional time series signals, deep features of the signals acquired in the step 1 are extracted, the signals are denoised by adopting a continuous wavelet transform method, useful defect signals are removed from the signals, then the denoised time series signals are converted into a two-dimensional scale map, all the two-dimensional scale maps are collected, and the features of the signals in a time-frequency domain can be seen from the two-dimensional scale map. Graying and dividing ROI of the collected two-dimensional scale map, filtering the image, extracting edges by Canny operator, extracting the characteristic edges of the two-dimensional scale map by Canny operator, highlighting the defect edges, improving the accuracy of subsequent model identification,
the preprocessing is to perform graying processing on the acquired picture, reduce the picture memory, accelerate the training and recognition speed of the cavity convolution neural network, segment the ROI to determine the defect characteristic position of the two-dimensional scale map, filter the image, reduce the filtering of the image, convert the data after each A scanning into the two-dimensional scale map, not only can deeply provide useful defect signals, but also can better enable the recognition of the cavity convolution neural network to reach the optimal state as the input of the cavity convolution neural network.
5. The method according to claim 4, wherein step 3 is specifically:
the cavity convolution neural network comprises a convolution layer, a pooling layer, an activation function, a batch normalization layer and an attention mechanism, the cavity convolution is to expand the receptive field of the conventional 3 x 3 convolution kernel to the receptive field of the 5 x 5 convolution kernel through the change of the cavity, other expanded cavities are filled with zeros, and the cavity convolution neural network is increased to enable the same calculated amount to obtain a larger receptive field, so that more useful information is obtained.
6. The method according to claim 5, wherein step 4 is specifically:
dividing the types of the ceramic membrane defects into scratches, fractures and holes, establishing a universal defect classification formula model, setting a classification task as N, expressing the embedding of the classification task as x, and evaluating the probability of the y classification by using a Softmax function, wherein the formula (6) is as follows:
Figure FDA0003324907210000031
if x belongs to the ζ -th class, the corresponding Softmax loss function is expressed as:
Figure FDA0003324907210000041
wherein (y)1,…,yk,…,yN) Is a label of classification, ykAnd (5) taking the value as 0 or 1, and outputting the probability through the last Softmax to obtain a detection result.
CN202111259179.9A 2021-10-28 2021-10-28 Deep learning flat ceramic membrane ultrasonic defect detection method based on deep features Active CN114088817B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111259179.9A CN114088817B (en) 2021-10-28 2021-10-28 Deep learning flat ceramic membrane ultrasonic defect detection method based on deep features

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111259179.9A CN114088817B (en) 2021-10-28 2021-10-28 Deep learning flat ceramic membrane ultrasonic defect detection method based on deep features

Publications (2)

Publication Number Publication Date
CN114088817A true CN114088817A (en) 2022-02-25
CN114088817B CN114088817B (en) 2023-10-24

Family

ID=80298090

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111259179.9A Active CN114088817B (en) 2021-10-28 2021-10-28 Deep learning flat ceramic membrane ultrasonic defect detection method based on deep features

Country Status (1)

Country Link
CN (1) CN114088817B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117428291A (en) * 2023-12-18 2024-01-23 南京理工大学 Weld bead fusion width quantification method based on sonogram characteristic analysis

Citations (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030081215A1 (en) * 2001-01-09 2003-05-01 Ajay Kumar Defect detection system for quality assurance using automated visual inspection
RU2246724C1 (en) * 2003-08-25 2005-02-20 Кубланов Владимир Семенович Method of ultrasonic testing of material quality
US20080004527A1 (en) * 2006-04-05 2008-01-03 Coleman D Jackson High-resolution ultrasound spectral and wavelet analysis of vascular tissue
WO2017063355A1 (en) * 2015-10-15 2017-04-20 浙江大学 Method for automatically identifying defect type of polyethylene electrofusion joint by means of ultrasonic phased array inspection
CN107037131A (en) * 2017-05-04 2017-08-11 中南大学 A kind of tiny flaw supersonic detection method theoretical based on the extreme value distribution
CN108694711A (en) * 2018-02-19 2018-10-23 江苏新时高温材料股份有限公司 Ceramic membrane surface defect two-dimensional detecting method is realized based on machine vision technique
CN108693251A (en) * 2018-02-19 2018-10-23 江苏新时高温材料股份有限公司 The 3 D detection method of hollow plate type ceramic film deep zone defect is realized based on ultrasonic technique
CN108764006A (en) * 2018-02-05 2018-11-06 北京航空航天大学 A kind of SAR image object detection method based on deeply study
CN108896660A (en) * 2018-07-09 2018-11-27 中南大学 A kind of hexagonal crystal material near surface tiny flaw detection method based on shear wave back scattering
CN108961203A (en) * 2018-02-19 2018-12-07 江苏新时高温材料股份有限公司 A kind of three-dimensional rebuilding method of fusion ultrasound and the hollow plate type ceramic film defect of machine vision technique
CN109444863A (en) * 2018-10-23 2019-03-08 广西民族大学 A kind of estimation method of the narrowband ultrasonic echo number based on convolutional neural networks
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
CN110243934A (en) * 2019-05-30 2019-09-17 中国计量大学上虞高等研究院有限公司 A kind of ultrasonic weld seam detection method based on wavelet convolution neural network
WO2019200753A1 (en) * 2018-04-17 2019-10-24 平安科技（深圳）有限公司 Lesion detection method, device, computer apparatus and storage medium
EP3581961A1 (en) * 2018-06-13 2019-12-18 Technische Universität München Method and apparatus for ultrasound imaging with improved beamforming
CN111795931A (en) * 2020-07-06 2020-10-20 武汉理工大学 Reconstruction extraction method for laser ultrasonic defect detection diffraction echo signal
CN112147223A (en) * 2020-10-14 2020-12-29 首钢京唐钢铁联合有限责任公司 Method for detecting internal defects of casting blank
WO2021004174A1 (en) * 2019-07-11 2021-01-14 暨南大学 Ultrasound image-based automatic measuring method and device for intrapartum cephalopelvic relationship
CN112232400A (en) * 2020-10-12 2021-01-15 太原科技大学 Stainless steel weld ultrasonic defect detection method based on depth feature fusion
CN112614091A (en) * 2020-12-10 2021-04-06 清华大学 Ultrasonic multi-section data detection method for congenital heart disease
CN113298757A (en) * 2021-04-29 2021-08-24 同济大学 Metal surface defect detection method based on U-NET convolutional neural network

Patent Citations (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030081215A1 (en) * 2001-01-09 2003-05-01 Ajay Kumar Defect detection system for quality assurance using automated visual inspection
RU2246724C1 (en) * 2003-08-25 2005-02-20 Кубланов Владимир Семенович Method of ultrasonic testing of material quality
US20080004527A1 (en) * 2006-04-05 2008-01-03 Coleman D Jackson High-resolution ultrasound spectral and wavelet analysis of vascular tissue
WO2017063355A1 (en) * 2015-10-15 2017-04-20 浙江大学 Method for automatically identifying defect type of polyethylene electrofusion joint by means of ultrasonic phased array inspection
CN107037131A (en) * 2017-05-04 2017-08-11 中南大学 A kind of tiny flaw supersonic detection method theoretical based on the extreme value distribution
CN108764006A (en) * 2018-02-05 2018-11-06 北京航空航天大学 A kind of SAR image object detection method based on deeply study
CN108694711A (en) * 2018-02-19 2018-10-23 江苏新时高温材料股份有限公司 Ceramic membrane surface defect two-dimensional detecting method is realized based on machine vision technique
CN108693251A (en) * 2018-02-19 2018-10-23 江苏新时高温材料股份有限公司 The 3 D detection method of hollow plate type ceramic film deep zone defect is realized based on ultrasonic technique
CN108961203A (en) * 2018-02-19 2018-12-07 江苏新时高温材料股份有限公司 A kind of three-dimensional rebuilding method of fusion ultrasound and the hollow plate type ceramic film defect of machine vision technique
WO2019200753A1 (en) * 2018-04-17 2019-10-24 平安科技（深圳）有限公司 Lesion detection method, device, computer apparatus and storage medium
EP3581961A1 (en) * 2018-06-13 2019-12-18 Technische Universität München Method and apparatus for ultrasound imaging with improved beamforming
CN108896660A (en) * 2018-07-09 2018-11-27 中南大学 A kind of hexagonal crystal material near surface tiny flaw detection method based on shear wave back scattering
CN109444863A (en) * 2018-10-23 2019-03-08 广西民族大学 A kind of estimation method of the narrowband ultrasonic echo number based on convolutional neural networks
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
CN110243934A (en) * 2019-05-30 2019-09-17 中国计量大学上虞高等研究院有限公司 A kind of ultrasonic weld seam detection method based on wavelet convolution neural network
WO2021004174A1 (en) * 2019-07-11 2021-01-14 暨南大学 Ultrasound image-based automatic measuring method and device for intrapartum cephalopelvic relationship
CN111795931A (en) * 2020-07-06 2020-10-20 武汉理工大学 Reconstruction extraction method for laser ultrasonic defect detection diffraction echo signal
CN112232400A (en) * 2020-10-12 2021-01-15 太原科技大学 Stainless steel weld ultrasonic defect detection method based on depth feature fusion
CN112147223A (en) * 2020-10-14 2020-12-29 首钢京唐钢铁联合有限责任公司 Method for detecting internal defects of casting blank
CN112614091A (en) * 2020-12-10 2021-04-06 清华大学 Ultrasonic multi-section data detection method for congenital heart disease
CN113298757A (en) * 2021-04-29 2021-08-24 同济大学 Metal surface defect detection method based on U-NET convolutional neural network

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
刘旭等: "超声检测缺陷分类的降噪及特征提取问题研究", 《中国矿业大学学报》, pages 248 - 251 *
孙进等: "陶瓷膜表面缺陷的表征与分类研究", 《徐州工程学院学报》, pages 76 - 79 *
廖均梅等: "基于频域特征的图像深度信息提取方法", 《自动化与仪器仪表》, pages 167 - 168 *
李力等: "超声 A 扫描信号建模及其缺陷识别方法研究", 《中国测试》, pages 14 - 16 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117428291A (en) * 2023-12-18 2024-01-23 南京理工大学 Weld bead fusion width quantification method based on sonogram characteristic analysis

Also Published As

Publication number Publication date
CN114088817B (en) 2023-10-24

Similar Documents

Publication Publication Date Title
Jansen et al. Immersion tomography using Rayleigh and Lamb waves
CN1844908A (en) Ultrasonic wave detection method and detection apparatus therefor
Yang et al. Comparative study of ultrasonic techniques for reconstructing the multilayer structure of composites
CN109165617A (en) A kind of ultrasonic signal sparse decomposition method and its signal de-noising and defect inspection method
CN114088817A (en) Deep learning flat ceramic membrane ultrasonic defect detection method based on deep features
CN108693251B (en) Three-dimensional detection method for realizing deep-layer defects of hollow plate-type ceramic membrane based on ultrasonic technology
CN100495018C (en) Ultrasonic method and device for testing macroscopic cleanness of continuous casting billet
CN101344505A (en) On-line monitoring method and device for pollution of hollow fiber film
CN115420806A (en) Nondestructive ultrasonic detection method based on neural network and image fusion
CN107727749A (en) A kind of ultrasonic quantitative detection method based on wavelet packet fusion feature extraction algorithm
CN113126099A (en) Single-beam substrate classification method based on audio feature extraction
CN116698986B (en) Sound absorbing material quality control system and method based on tuned band-pass filter
CN117686589A (en) Composite material delamination defect detection method based on ultrasonic guided wave
CN106680375B (en) For determining the Air Coupling supersonic detection method of the elasticity modulus of material
Jarno-Druaux et al. Dynamical evolution of ripples in a wave channel
Ibrahim et al. A review on ultrasonic process tomography system
CN112818762B (en) Large-size composite material and rapid nondestructive testing method for sandwich structure thereof
CN115184276A (en) Towed plankton polarization imaging recorder
CN109060611A (en) A kind of suspended sediment data processing method based on laser particle analyzer
CN113624710A (en) Terahertz information identification method for liquid substance component change and application thereof
CN115078190A (en) Suspension body on-site laser granularity data processing method and device
CN105651863A (en) Detection method of bore hole in corn seed
Okafor et al. Non-destructive evaluation of concrete with ultrasonic C-scan and digital image enhancement techniques
Zhang et al. Eliminating backwall effects in the phased array imaging of near backwall defects
CN114487115B (en) High-resolution defect nondestructive testing method based on combination of Canny operator and ultrasonic plane wave imaging

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