CN113988142A - Tunnel lining cavity acoustic identification method based on convolutional neural network - Google Patents
Tunnel lining cavity acoustic identification method based on convolutional neural network Download PDFInfo
- Publication number
- CN113988142A CN113988142A CN202111607420.2A CN202111607420A CN113988142A CN 113988142 A CN113988142 A CN 113988142A CN 202111607420 A CN202111607420 A CN 202111607420A CN 113988142 A CN113988142 A CN 113988142A
- Authority
- CN
- China
- Prior art keywords
- sound pressure
- tunnel lining
- neural network
- convolutional neural
- time domain
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- 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
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Abstract
The invention discloses a tunnel lining cavity acoustic identification method based on a convolutional neural network. And then, standardizing the time-domain sound pressure sample data of the two working conditions by using a data standardization program. And then, performing time-frequency analysis on the sound pressure data under the two working conditions by using a wavelet analysis technology, and converting the sound pressure data into a two-dimensional time-frequency characteristic spectrogram. And finally, constructing a deep learning model based on the convolutional neural network, and training the model by using a two-dimensional time-frequency characteristic map of sound pressure information to obtain a tunnel cavity recognition model. And finally, identifying and classifying new sample data by using the tunnel void identification model, and judging whether a tunnel region corresponding to the sample data has a void. The method has higher accuracy, reliability, robustness and applicability to the identification of the cavity of the tunnel lining structure.
Description
Technical Field
The invention relates to the field of detection of cavities in tunnel lining structures, in particular to a tunnel lining cavity acoustic identification method based on a convolutional neural network.
Background
In the construction and service stage, the tunnel lining structure can generate various damage types such as cracks, cavities and the like under the action of factors such as gravity, load, settlement, environment and the like, wherein the cavities behind the lining have the most serious influence on the normal service of the tunnel.
The hollow holes of the lining structure are mainly caused by the following points:
in the tunnel construction excavation stage, if a mining method is used, when smooth blasting control is not proper, overbreak is caused, a construction unit does not fill according to relevant regulations, filling between a support and surrounding rocks is not tight, and further a cavity is formed.
Secondly, in the construction process of secondary lining of the tunnel, the concrete is in an unsaturated state due to insufficient power of pumping concrete, poor fluidity of the concrete, too early pumping pipes and the like, and further the tunnel lining is hollow.
And thirdly, the stability of the bottom of the formwork support is insufficient, particularly at the vault position of the tunnel, the downward displacement of the formwork is large, and then concrete at the vault position of the tunnel sinks to be empty and form a cavity.
Fourthly, after the secondary lining is poured, the bottom of the vertical wall is not poured in time at the position of the vault, the vertical wall may generate relative displacement to cause the vault to sink, and the vault lining concrete sinks and is emptied, so that a cavity is formed.
The raw materials used in the tunnel construction process are not strictly controlled, the shrinkage of concrete is too large due to improper sand-stone gradation, excessive cement consumption, excessive concrete water-cement ratio, improper temperature difference and ventilation control and the like, so that the lining concrete sinks and voids to form cavities.
Sixthly, in the long-term operation process of the tunnel, underground water erodes or scours surrounding rocks behind the tunnel, so that the supporting structure and the surrounding rocks are separated.
When the cavity appears in the tunnel lining structure, the stress of lining structure and the stress state of country rock all will change, and the fracture takes place easily at the lining cutting top edge, and then forms the passageway for the circulation of groundwater, leads to the emergence of percolating water, and percolating water can get into the lining cutting along cavity and crack, and then leads to aggravating of seepage phenomenon, and then leads to freeze injury and reinforcing bar corrosion. The occurrence of the cavity can also cause the surrounding rock to loose and deform due to the loss of the support of the surrounding rock, so that the tunnel structure is unstable, falls into blocks and falls off, and sudden collapse can also occur in serious cases, thereby causing serious influence on the driving safety.
The existence of the cavity brings great potential safety hazard to the safe operation of the tunnel, and the timely discovery and identification of the position and the range of the tunnel lining cavity have important significance for guaranteeing the safe and stable service of the tunnel structure.
The current common methods for detecting the cavity of the tunnel lining comprise the following steps: knocking echo detection method based on acoustics, geological radar method and ultrasonic echo synthesis method.
The geological radar method and the ultrasonic echo comprehensive method have the advantages of automation, rapidness, no damage, low cost and the like in the tunnel cavity detection process, but the automatic detection method is sensitive to material properties, has high requirement on the homogeneity of a detected target and has more limitations on the test environment. But the service environment of the tunnel is complex, and the tunnel construction quality cannot be guaranteed. Due to the existence of the contradiction, the geological radar method and the ultrasonic echo comprehensive method have high missing rate in the tunnel cavity detection process, and the tunnel cavity condition of the whole line cannot be effectively checked.
At present, for the actual detection work of the tunnel hole, a knocking echo detection method based on acoustics is still largely used. The knocking echo detection method based on acoustics can be divided into two stages of initial detection and re-detection, wherein in the initial detection process, maintainers knock and detect tunnel sections point by point, and the maintainers judge whether the tunnel lining structure is abnormal or not through knocking echoes and make corresponding marks. And in the rechecking process, a maintainer visually detects whether the tunnel has a cavity by adopting a core drilling method. The knocking echo method has the advantages of intuition, high precision, high coverage rate, high reliability, capability of truly reflecting the internal condition of the tunnel lining structure and the like. The knocking echo detection method based on acoustics can effectively finish effective investigation of the hole damage of the whole section of the tunnel, and is a tunnel hole detection mode which is most trusted by field maintenance personnel, but the method has the advantages of low detection efficiency, high labor cost, low initial detection accuracy, poor detection process safety, and higher requirements of the initial detection process on the experience of engineering personnel.
Disclosure of Invention
The invention provides a tunnel lining cavity acoustic recognition method based on a convolutional neural network, which is characterized in that a convolutional neural network model is used for recognizing whether a cavity appears in a tunnel, firstly, a wavelet analysis technology (a leading-edge technology in the field of signal analysis) is used for carrying out time-frequency analysis on sound pressure data to obtain the proportion characteristics of instantaneous frequency energy of sound pressure under different working conditions, and the proportion characteristics are presented in a two-dimensional map form to finish primary characteristic extraction work; and then, a convolutional neural network model (a leading-edge technology in the field of deep learning) is utilized, a time-frequency map is used as a material, and the purpose of finally identifying the hole is achieved.
The technical scheme provided by the invention is as follows:
on one hand, the tunnel lining cavity acoustic identification method based on the convolutional neural network comprises the following steps:
s1: acquiring sound pressure time domain signal data of the tunnel lining structure under the external excitation action;
s2: carrying out standard pretreatment on the sound pressure time domain signal data;
s3: carrying out time-frequency analysis on the sound pressure time-domain signal data subjected to the standardized preprocessing to obtain a two-dimensional characteristic map reflecting the sound pressure energy distribution;
s4: setting and training a tunnel lining cavity recognition model structure and hyper-parameters based on a convolutional neural network;
the tunnel lining cavity recognition model structure based on the convolutional neural network comprises the convolutional neural network and a classifier which are sequentially connected; processing the historical samples according to S1-S3 to obtain training samples, and performing model training by taking the two-dimensional characteristic images and the identification labels of the training samples as input information and output information of a convolutional neural network-based tunnel lining cavity identification model respectively;
s5: and (3) carrying out hole recognition on the two-dimensional characteristic map obtained by processing the tunnel lining according to S1-S3 by using the trained tunnel lining hole recognition model based on the convolutional neural network.
Compared with the prior art, the technical scheme of the invention combines a wavelet analysis technology and a convolution neural network technology, and is originated from tunnel hole identification; in the design of a data acquisition scheme, firstly, the specific structure of a tunnel needs to be considered, secondly, the reason and the distribution rule of the formation of the cavity need to be considered, and finally, the coverage surface of the data needs to meet the training requirement of a convolutional neural network, and is embodied in the number of samples and the coverage surface; the knowledge needs to be deeply known about civil engineering knowledge and deep learning knowledge and can be obtained through a large amount of research and analysis, which is not thought and achieved by professionals in a single field; and determining a wavelet analysis method adaptive to the specific working condition of the cavity identification. Wavelet analysis is a very wide concept, and the difference in use is very large, which is mainly reflected in the selection of wavelet basis functions, different signal samples and different wavelet basis functions, and the final obtained effects are different. The time-frequency map presentation effect influences the identification precision of the convolutional neural network. The invention compares the analysis and presentation effects of various wavelet basis functions on sound pressure signals, and finds that the MLP wavelet basis functions are most suitable for the working condition.
Furthermore, the convolutional neural network of the tunnel lining cavity identification model based on the convolutional neural network is provided with two convolutional layers, two pooling layers and two full-connection layers; the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the first full-connection layer and the second full-connection layer are sequentially connected;
the convolution layer adopts convolution kernel with size of 2Step size is set to 2, and the activation function used is a simgioid function.
Further, the classifier adopts a softmax classifier.
Further, sound pressure time domain signal data of the tunnel lining structure under the external excitation action are acquired by selecting a knocking point every 10 degrees in an area of 0-30 degrees on one side of the vault and selecting a knocking point at 60 degrees to acquire the sound pressure time domain signal data.
Further, performing time-frequency analysis on the sound pressure time domain signal data subjected to the standardized preprocessing refers to performing continuous wavelet transform on the sound pressure time domain signal data subjected to the standardized preprocessing. :
wherein the content of the first and second substances,is Fourier transform; t is a time independent variable; a. b is selected time starting point and selected time ending point respectively;is a frequency independent variable.
In the time-frequency map, the calculation mode of the energy ratio of the sound pressure energy under each instantaneous frequency is derived; further, in the training process of the tunnel lining cavity recognition model based on the convolutional neural network, the training round is set to be 80 rounds, the initial learning rate is set to be 0.01, the learning rate is updated by adopting an exponential decay method, and the learning rate is attenuated to be 0.9 times of the last learning rate after each training round is completed.
Further, the step of performing normalization preprocessing on the sound pressure time domain signal data specifically includes:
and taking the peak point of the sound pressure information as a capturing target point, capturing 4800 sampling points before the peak point, 57600 sampling points after the peak point, totaling 62400 sampling points, and totaling the duration for 1.3 s.
Further, a microphone acoustic pressure sensor (sensitivity) is utilized1.5 dB) acquiring structural sound pressure time domain data information sent by a tunnel lining cavity region and a non-cavity region under the action of external knocking excitation, wherein the sampling frequency is set to be 48kHz, and the sampling time of each knocking test is set to be 2.5 seconds;
the sensitivity of the microphone sound pressure sensor is-1.5 dB to +1.5 dB.
In another aspect, a convolutional neural network-based tunnel lining cavity acoustic identification system includes:
sound pressure time domain signal acquisition unit: the system is used for acquiring sound pressure time domain signal data of the tunnel lining structure under the external excitation action;
a signal preprocessing unit: carrying out standard pretreatment on the sound pressure time domain signal data;
a signal time-frequency analysis unit: carrying out time-frequency analysis on the sound pressure time-domain signal data subjected to the standardized preprocessing to obtain a two-dimensional characteristic map reflecting the sound pressure energy distribution;
a recognition model construction and training unit: setting and training a tunnel lining cavity recognition model structure and hyper-parameters based on a convolutional neural network;
the tunnel lining cavity recognition model structure based on the convolutional neural network comprises the convolutional neural network and a classifier which are sequentially connected;
calling a sound pressure time domain signal acquisition unit, a signal preprocessing unit and a signal time-frequency analysis unit by using a historical sample for processing to obtain a training sample, and performing model training by respectively taking a two-dimensional characteristic image and an identification label of the training sample as input information and output information of a tunnel lining cavity identification model based on a convolutional neural network;
an identification unit: and carrying out hole recognition on a two-dimensional characteristic map obtained by processing the tunnel lining calling sound pressure time domain signal acquisition unit, the signal preprocessing unit and the signal time frequency analysis unit by using the trained tunnel lining hole recognition model based on the convolutional neural network.
Advantageous effects
Compared with the prior art, the invention has the advantages that:
(1) the method comprehensively utilizes the advantages of wavelet analysis feature extraction and the advantages of convolutional neural network image identification, converts the sound pressure signal from a disordered one-dimensional signal into a time-frequency spectrogram which can clearly present the energy distribution features of the sound pressure signal, then utilizes the convolutional neural network to analyze, identify and classify the time-frequency spectrogram, and finally realizes the judgment of whether the tunnel has the cavity or not with higher precision.
(2) The method combines two steps of feature extraction, hole identification and the like in the identification process of the tunnel lining hole, ensures higher identification precision and has stronger applicability.
(3) The method gets rid of the dependence on expert diagnosis experience and complex signal processing in the tunnel lining cavity identification process, and has better universality.
Drawings
FIG. 1 is a flow chart of a method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of an arrangement of sound pressure time domain signal acquisition tapping points;
fig. 3 is original sound pressure time domain data, wherein (a) is vibration sound pressure time domain data of the measuring point # 1 of the cavity region, and (b) is vibration sound pressure time domain data of the measuring point # 1 of the non-cavity region;
FIG. 4 is normalized data, wherein (a) is a normalized time domain diagram of sound pressure at the number # 1 measuring point of the hollow region, and (b) is a normalized time domain diagram of sound pressure at the number # 1 measuring point of the normal region;
FIG. 5 is a diagram of a time-frequency feature spectrum;
FIG. 6 is a schematic diagram of a tunnel lining cavity recognition model structure and training based on a convolutional neural network.
Detailed Description
The invention will be further described with reference to the following figures and examples.
As shown in fig. 1, a tunnel lining cavity acoustic identification method based on a convolutional neural network includes the following steps:
s1: acquiring sound pressure time domain signal data of the tunnel lining structure under the external excitation action;
the sound pressure time domain signal data of the tunnel lining structure under the external excitation action are acquired by selecting a knocking point every 10 degrees in a 0-30-degree area on one side of the vault and selecting a knocking point at 60 degrees.
Using microphone sound pressure sensors (sensitivity)1.5 dB) acquiring structural sound pressure time domain data information sent by a tunnel lining cavity region and a non-cavity region under the action of external knocking excitation, wherein the sampling frequency is set to be 48kHz, and the sampling time of each knocking test is set to be 2.5 seconds;
the sensitivity of the microphone sound pressure sensor is-1.5 dB to +1.5 dB.
In S1, an external stimulus is applied to the tunnel, the lining structure is vibrated to sound, and sound pressure time domain information is collected using a microphone sound pressure sensor. In this step, sample data of two types of regions are collected: a void region and a non-void region. The process can be completed in the centralized maintenance stage of the tunnel lining cavities, and after the tunnel cavities are rechecked by using a core drilling method, maintainers clearly know which areas of the tunnel lining have the cavities and which areas do not have the cavities, so that the correctness of the collected sample data can be effectively ensured. In the sampling process of the sample data, the set sampling frequency in the step is 48kHz, the sampling time of each knocking test is set to be 1s, and 500 groups of sound pressure time domain data under two working conditions are obtained. The tunnel lining cavity is mainly formed due to insufficient concrete pumping pressure in the construction stage, gravity influence and underground water erosion in the service period. The hollow holes are generally arranged in the area above the arch waist of the tunnel and are mainly distributed in the area from 0 degrees of the arch top to 30 degrees of the arch top at both sides. In order to ensure that the sound data set can effectively reflect the characteristic frequency of the knocking echo of each area of the tunnel, the position of a knocking point is selected according to the distribution density of the holes, one knocking point is selected at intervals of 10 degrees in an area of 0-30 degrees on one side of the vault, and one knocking point is selected at an area of 60 degrees. The numbers of the shot points are respectively #1, #2, #3, #4, #5, and the shot point distribution chart is shown in FIG. 2.
In a concrete engineering practice, five measuring points meeting requirements are difficult to find on a tunnel section, measuring points meeting angle requirements are selected on different sections according to a cavity working condition, a knocking test is completed on one section under a normal working condition, and each measuring point is subjected to a knocking test for 100 times. FIG. 3 shows the sound pressure data of vibration generated by the cavity region and the non-cavity region of No. # 1 measuring point under the external excitation;
s2: carrying out standard pretreatment on the sound pressure time domain signal data;
the step of carrying out standardization preprocessing on the sound pressure time domain signal data specifically includes:
and taking the peak point of the sound pressure information as a capturing target point, capturing 4800 sampling points before the peak point, 57600 sampling points after the peak point, totaling 62400 sampling points, and totaling the duration for 1.3 s.
The sampling frequency was 48000Hz, i.e. the number of acquisitions per second was 48000, in order to normalize the initial sound pressure data. The initial signal data is intercepted, processed and standardized through corresponding preprocessing setting. The resulting normalized data was 1.3s long and contained 62400 samples (48000 × 1.3= 64200).
After the data shown in fig. 3 is normalized, as shown in fig. 4;
s3: carrying out time-frequency analysis on the sound pressure time-domain signal data subjected to the standardized preprocessing to obtain a two-dimensional characteristic map reflecting the sound pressure energy distribution;
the time-frequency analysis of the sound pressure time-domain signal data after the standardized preprocessing refers to the adoption of continuous wavelet transformation on the sound pressure time-domain signal data after the standardized preprocessing. :
wherein the content of the first and second substances,is Fourier transform; t is a time independent variable; a. b is selected time starting point and selected time ending point respectively;is a frequency independent variable.
In the time-frequency map, the calculation mode of the energy ratio of the sound pressure energy under each instantaneous frequency is derived;
and performing corresponding analysis on the sound pressure time domain signal characteristic extraction, namely the sound pressure signal. The existing analysis methods include: fourier transform, short-time fourier transform, Wigner-Ville distribution (WVD), wavelet transform, and the like. The invention uses wavelet analysis technique to analyze the sound pressure time domain data in two dimensions of time domain and frequency domain. The disordered one-dimensional time sequence data is converted into a time-frequency characteristic map which can clearly and obviously highlight the energy distribution of the sound pressure in two dimensions of a time domain and a frequency domain, and the preliminary characteristic extraction work of the sound pressure information is completed through the process.
Wavelet analysis is the development and continuation of the idea of Fourier analysis, and has been closely related to Fourier analysis since its generation, and wavelet base construction and time-frequency analysis both depend on Fourier analysis, and both are complementary, and wavelet analysis has the following advantages compared with Fourier transform:
the essence of the Fourier transform is to use energy limited signalsf(t) Decomposed into { exp (j) }t) } is a space of an orthogonal base, so that a sound pressure signal is converted into a frequency form from a time sequence form to obtain frequency spectrum information (one dimension); the essence of wavelet transform is to transmit energy limited signalf(t) Decompose to W-jAnd V-jThe sound pressure signal characteristics are displayed in two dimensions of time domain and frequency.
② the basic function used in Fourier transform is only sin: (t),cos(t), having uniqueness; the functions (wavelet functions) used for wavelet analysis have diversity and can be adjusted according to different signal types and characteristics.
The Fourier transform is to perform spectrum analysis of the whole time course in the whole time period, and the obtained spectrum result cannot reflect the local characteristics of the discrete signal; if a signal changes only at a certain moment and in the neighborhood, the whole frequency spectrum of the signal is affected, the time position of sudden change and the intensity of the sudden change cannot be calibrated for the change of the frequency spectrum, and Fourier transform is insensitive to the singularity of the signal. In many engineering applications, however, singularities are just features in the local range of the signal we are interested in. The wavelet analysis adopts a moving window setting, the analysis method is a time-frequency localized analysis method with a fixed analysis window, but a time window and a frequency window can be automatically adjusted along with frequency change, the automatic adjustment of the time window enables the wavelet transformation to have the self-adaptability to signals, ensures higher frequency resolution and lower time resolution in a low frequency band and higher time resolution and lower frequency resolution in a high frequency band, and therefore, the synchronous analysis of a signal time domain and a signal frequency domain is realized.
Wavelet analysis
The wavelet analysis has multi-scale characteristics, signals can be gradually observed from coarse to fine, a telescopic window can be obtained by selecting a scale factor and a translation factor, and the wavelet transformation can have the capacity of representing local characteristics of the signals in both a time domain and a frequency domain by combining a proper wavelet base.
Definition of wavelet transform:
in the formula:a wavelet basis function; t is a time independent variable; a. b is selected time starting point and selected time ending point respectively;
transient signalSum wavelet basis functionIs a function of the square integrable real number, andsatisfying the fourier transform condition. Will be provided withPerforming expansion and translation to obtain wavelet base, and recording as
In the formula: a is a scale factor, a >0, and b is a translation factor.
Wavelet ridge and instantaneous frequency
In time-frequency analysis, a set consisting of points where the wavelet coefficients modulus takes a maximum value (i.e., wavelet ridge points) at each time instant is called a "wavelet ridge. The wavelet ridge points are energy concentration points of the signals in the frequency components, and the wavelet ridge lines can more clearly reflect the time-frequency characteristics of the signals.
In the formula: h (a), (b)) Hilbert transform for the signal;is the instantaneous amplitude;is the instantaneous phase; () Is composed ofThe canonical transform pair of (1).
Wherein the content of the first and second substances,
phase stagnation pointSatisfy the requirement of() Is not less than 0, and is obtained by the formula (9)
From the formula (11)Is a function of a and b, the wavelet ridge line is satisfied on the phase planeSet of all points (a, b) of (a, b) = b, these points are called wavelet ridge points. Thus, it can be obtained from the formula (11)
The wavelet ridge points are energy concentration points of the signals in the frequency components, and all wavelet ridges can more clearly reflect real-time characteristics of the signals.
Wavelet energy ratio
The wavelet coefficients given by equation (8) form a small spectrum on the time-frequency plane. In wavelet spectrum, for any frequency componentThe corresponding wavelet coefficient is integrated along the time axis to obtain the informationThe frequency division energy of the frequency component in the number in the train passing time period. Further, the energy corresponding to all frequency components is along the frequency axisfThe total energy of the signal can be obtained by integration. From which it is possible to define the frequency of the signalAt a wavelet energy ratio of
In the invention, wavelet analysis is used for carrying out time-frequency analysis on initial sound pressure data, on one hand, the time-frequency analysis is carried out on sound pressure data under two working conditions, preliminary feature extraction is completed, and energy distribution features of knocking echoes under different working conditions and at different time and frequencies are obtained; and on the other hand, the disordered one-dimensional time domain data is converted into two-dimensional spectrogram data capable of reflecting the working condition characteristics.
The time-frequency analysis of the time-domain signal by using the wavelet analysis technology is an advanced signal processing means at present, and the energy distribution characteristics of the time-domain signal can be deeply known from two angles of time domain, frequency domain and the like by using the technology.
According to the method, the knocking echoes under two working conditions of the tunnel cavity and the normal tunnel need to be subjected to time-frequency analysis, and the aim of identifying and distinguishing the two working conditions is fulfilled by observing and analyzing the energy distribution time-frequency spectrum of the knocking echoes under the two working conditions. However, for a specific time domain signal of the tunnel knocking echo, in order to make the time frequency spectrum have a stronger distinction degree and further achieve a better distinction effect, a wavelet basis function matched with the time frequency spectrum needs to be selected, which is one of core works of the technical scheme of the invention.
For a complete tunnel structure, the knocking echo characteristic is crisp, the sound pressure energy is mainly distributed at high frequency, the energy distribution is concentrated, and the sound pressure energy is mainly distributed at about 2000Hz in the reverberation stage. For a tunnel region with a cavity, knocking echo is characterized by obvious clunk and depression, wide energy distribution frequency band and stable reverberation sound pressure energy mainly distributed in 500Hz-1500 Hz.
In view of the above, this patent chooses to emphasize the mesoscopic feature of the knocking echo in the void region. On one hand, because of the tunnel with a complete structure, the knocking echo time-frequency characteristics are uniform, and the significance of highlighting the sound pressure energy distribution of a high-frequency section is not large; on the other hand, when the tunnel is in a cavity or other damages, the structural integrity of the tunnel is damaged, the natural frequency of the tunnel is necessarily changed, the knocking rising main frequency is shifted downwards, but the shifting quantity of different damage types is different from the energy distribution characteristics. In the patent, the energy distribution details of knocking echoes under the condition that a tunnel cavity is highlighted as much as possible are sought, and the subsequent CNN deep learning model is served for accurately identifying the cavity diseases in a plurality of diseases.
Based on the purpose, the invention compares the time-frequency analysis effect of various wavelet basis functions on the knocking echo, and finally finds that the Modified Littlewood-Pattern (MLP) wavelet basis functions have the best analysis effect.
Currently, in the industry, the most frequently used wavelet basis functions for vibration and noise analysis are the Morlet wavelet basis functions (=cos(5t)) And Mexihat wavelet (=) The present invention tries to use the two wavelet basis functions for analyzing the sound pressure time domain signal, but the two wavelet basis functions have a better effect on the energy distribution of the sound pressure below 1000Hz, but are not suitable for the technique of the present inventionThe operation is required. Finally, the invention selects the Modified Littlewood-Paley (MLP) wavelet basis function (MLP)=,21/4) The MLP wavelet is mainly applied to analysis of high-frequency signals, has a good analysis effect on signal components below 2000Hz, and can effectively meet the technical requirements of the patent, and the obtained time-frequency spectrum is shown in figure 5.
S4: setting and training a tunnel lining cavity recognition model structure and hyper-parameters based on a convolutional neural network, wherein the model structure and a training schematic diagram are shown in FIG. 6;
the tunnel lining cavity recognition model structure based on the convolutional neural network comprises the convolutional neural network and a classifier which are sequentially connected; processing the historical samples according to S1-S3 to obtain training samples, and performing model training by taking the two-dimensional characteristic images and the identification labels of the training samples as input information and output information of a convolutional neural network-based tunnel lining cavity identification model respectively;
the convolutional neural network of the tunnel lining cavity identification model based on the convolutional neural network is provided with two convolutional layers, two pooling layers and two full-connection layers; the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the first full-connection layer and the second full-connection layer are sequentially connected;
the convolution layer adopts convolution kernel with size of 2Step size is set to 2, and the activation function used is a simgioid function.
The classifier adopts a softmax classifier.
In the training process of the tunnel lining cavity recognition model based on the convolutional neural network, the training round is set to 80 rounds, the initial learning rate is set to 0.01, the learning rate is updated by adopting an exponential decay method, and the learning rate is attenuated to 0.9 times of the last learning rate every time one training round is completed.
In S5, the time-frequency map reflecting the tunnel lining knocking echo energy distribution under two working conditions is used for training the model, so that on one hand, the excavation of depth characteristic information is completed, and on the other hand, the adjustment of model parameters is completed. And taking 80% of sample data as a training set for training the model, taking 20% of sample data as a test set for verifying the model, and finally obtaining a mature tunnel void recognition model.
In the method, two types of different sample data are used, and the Softmax classifier can classify the input sample data into two types: no cavity appears in lining and a cavity appears in lining. The mature model can accurately judge whether the tunnel lining structure has a cavity or not by analyzing, identifying and classifying the new sound pressure data samples.
S5: and (3) carrying out hole recognition on the two-dimensional characteristic map obtained by processing the tunnel lining according to S1-S3 by using the trained tunnel lining hole recognition model based on the convolutional neural network. A tunnel lining cavity acoustic identification system based on a convolutional neural network comprises:
sound pressure time domain signal acquisition unit: the system is used for acquiring sound pressure time domain signal data of the tunnel lining structure under the external excitation action;
a signal preprocessing unit: carrying out standard pretreatment on the sound pressure time domain signal data;
a signal time-frequency analysis unit: carrying out time-frequency analysis on the sound pressure time-domain signal data subjected to the standardized preprocessing to obtain a two-dimensional characteristic map reflecting the sound pressure energy distribution;
a recognition model construction and training unit: setting and training a tunnel lining cavity recognition model structure and hyper-parameters based on a convolutional neural network;
the tunnel lining cavity recognition model structure based on the convolutional neural network comprises the convolutional neural network and a classifier which are sequentially connected;
calling a sound pressure time domain signal acquisition unit, a signal preprocessing unit and a signal time-frequency analysis unit by using a historical sample for processing to obtain a training sample, and performing model training by respectively taking a two-dimensional characteristic image and an identification label of the training sample as input information and output information of a tunnel lining cavity identification model based on a convolutional neural network;
an identification unit: and carrying out hole recognition on a two-dimensional characteristic map obtained by processing the tunnel lining calling sound pressure time domain signal acquisition unit, the signal preprocessing unit and the signal time frequency analysis unit by using the trained tunnel lining hole recognition model based on the convolutional neural network.
It should be understood that the functional unit modules in the embodiments of the present invention may be integrated into one processing unit, or each unit module may exist alone physically, or two or more unit modules are integrated into one unit module, and may be implemented in the form of hardware or software.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.
Claims (9)
1. A tunnel lining cavity acoustic identification method based on a convolutional neural network is characterized by comprising the following steps:
s1: acquiring sound pressure time domain signal data of the tunnel lining structure under the external excitation action;
s2: carrying out standard pretreatment on the sound pressure time domain signal data;
s3: carrying out time-frequency analysis on the sound pressure time-domain signal data subjected to the standardized preprocessing to obtain a two-dimensional characteristic map reflecting the sound pressure energy distribution;
s4: setting and training a tunnel lining cavity recognition model structure and hyper-parameters based on a convolutional neural network;
the tunnel lining cavity recognition model structure based on the convolutional neural network comprises the convolutional neural network and a classifier which are sequentially connected; processing the historical samples according to S1-S3 to obtain training samples, and performing model training by taking the two-dimensional characteristic images and the identification labels of the training samples as input information and output information of a convolutional neural network-based tunnel lining cavity identification model respectively;
s5: and (3) carrying out hole recognition on the two-dimensional characteristic map obtained by processing the tunnel lining according to S1-S3 by using the trained tunnel lining hole recognition model based on the convolutional neural network.
2. The method according to claim 1, wherein the convolutional neural network of the tunnel lining cavity recognition model based on the convolutional neural network is provided with two convolutional layers, two pooling layers and two fully-connected layers; the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the first full-connection layer and the second full-connection layer are sequentially connected;
3. The method of claim 1, wherein the classifier employs a softmax classifier.
4. The method of claim 1, wherein the step of acquiring the sound pressure time domain signal data of the tunnel lining structure under the external excitation action is to select one knocking point every 10 degrees in the area of 0-30 degrees on one side of the vault, and select one knocking point at 60 degrees for sound pressure time domain signal data acquisition.
5. The method of claim 1, wherein performing the time-frequency analysis on the normalized pre-processed sound pressure time domain signal data is performed by applying a continuous wavelet transform to the normalized pre-processed sound pressure time domain signal data.
6. The method as claimed in claim 1, wherein in the training process of the convolutional neural network-based tunnel lining void recognition model, the training round is set to 80 rounds, the initial learning rate is set to 0.01, and the learning rate is updated by an exponential decay method, wherein the learning rate decays to 0.9 times of the last learning rate every time one training round is completed.
7. The method according to claim 1, wherein the normalizing the sound pressure time domain signal data specifically comprises:
and taking the peak point of the sound pressure information as a capturing target point, capturing 4800 sampling points before the peak point, 57600 sampling points after the peak point, totaling 62400 sampling points, and totaling the duration for 1.3 s.
8. Method according to claim 1, characterized in that a microphone sound pressure sensor (sensitivity) is used1.5 dB) acquiring structural sound pressure time domain data information sent by a tunnel lining cavity region and a non-cavity region under the action of external knocking excitation, wherein the sampling frequency is set to be 48kHz, and the sampling time of each knocking test is set to be 2.5 seconds;
the sensitivity of the microphone sound pressure sensor is-1.5 dB to +1.5 dB.
9. A tunnel lining cavity acoustic identification system based on a convolutional neural network is characterized by comprising the following components:
sound pressure time domain signal acquisition unit: the system is used for acquiring sound pressure time domain signal data of the tunnel lining structure under the external excitation action;
a signal preprocessing unit: carrying out standard pretreatment on the sound pressure time domain signal data;
a signal time-frequency analysis unit: carrying out time-frequency analysis on the sound pressure time-domain signal data subjected to the standardized preprocessing to obtain a two-dimensional characteristic map reflecting the sound pressure energy distribution;
a recognition model construction and training unit: setting and training a tunnel lining cavity recognition model structure and hyper-parameters based on a convolutional neural network;
the tunnel lining cavity recognition model structure based on the convolutional neural network comprises the convolutional neural network and a classifier which are sequentially connected;
calling a sound pressure time domain signal acquisition unit, a signal preprocessing unit and a signal time-frequency analysis unit by using a historical sample for processing to obtain a training sample, and performing model training by respectively taking a two-dimensional characteristic image and an identification label of the training sample as input information and output information of a tunnel lining cavity identification model based on a convolutional neural network;
an identification unit: and carrying out hole recognition on a two-dimensional characteristic map obtained by processing the tunnel lining calling sound pressure time domain signal acquisition unit, the signal preprocessing unit and the signal time frequency analysis unit by using the trained tunnel lining hole recognition model based on the convolutional neural network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111607420.2A CN113988142B (en) | 2021-12-27 | 2021-12-27 | Tunnel lining cavity acoustic identification method based on convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111607420.2A CN113988142B (en) | 2021-12-27 | 2021-12-27 | Tunnel lining cavity acoustic identification method based on convolutional neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113988142A true CN113988142A (en) | 2022-01-28 |
CN113988142B CN113988142B (en) | 2022-04-29 |
Family
ID=79734374
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111607420.2A Active CN113988142B (en) | 2021-12-27 | 2021-12-27 | Tunnel lining cavity acoustic identification method based on convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113988142B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115015268A (en) * | 2022-05-24 | 2022-09-06 | 山东大学 | Tunnel lining defect disease diagnosis method based on laser technology |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100157737A1 (en) * | 2007-12-21 | 2010-06-24 | Schlumberger Technology Corporation | Microhydraulic fracturing with downhole acoustic measurement |
CN102395903A (en) * | 2009-04-16 | 2012-03-28 | 雪佛龙美国公司 | System and method to estimate compressional to shear velocity (vp/vs) ratio in a region remote from a borehole |
US20120155702A1 (en) * | 2005-01-13 | 2012-06-21 | Kraft Clifford H | System and Method for Detecting Nuclear Material in Shipping Containers |
CN110068816A (en) * | 2019-05-10 | 2019-07-30 | 长沙理工大学 | It is a kind of based on the Gpr Signal of Via Lifting Scheme wavelet basis building method |
CN110956164A (en) * | 2019-12-31 | 2020-04-03 | 南京工大桥隧与轨道交通研究院有限公司 | Train wheel damage identification method based on noise signals |
CN112036425A (en) * | 2020-05-09 | 2020-12-04 | 中铁四局集团有限公司 | Tunnel cavity state radar spectrum image recognition model construction method and tunnel cavity state radar spectrum image recognition method |
CN112415467A (en) * | 2020-11-06 | 2021-02-26 | 中国海洋大学 | Single-vector subsurface buoy target positioning implementation method based on neural network |
CN112557510A (en) * | 2020-12-11 | 2021-03-26 | 广西交科集团有限公司 | Concrete pavement void intelligent detection device and detection method thereof |
CN112561903A (en) * | 2020-12-24 | 2021-03-26 | 中铁建设集团基础设施建设有限公司 | Temperature shrinkage crack resistance method suitable for asphalt pavement in cold region |
CN113343329A (en) * | 2021-08-05 | 2021-09-03 | 中南大学 | Method for establishing parameterized railway tunnel model |
-
2021
- 2021-12-27 CN CN202111607420.2A patent/CN113988142B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120155702A1 (en) * | 2005-01-13 | 2012-06-21 | Kraft Clifford H | System and Method for Detecting Nuclear Material in Shipping Containers |
US20100157737A1 (en) * | 2007-12-21 | 2010-06-24 | Schlumberger Technology Corporation | Microhydraulic fracturing with downhole acoustic measurement |
CN102395903A (en) * | 2009-04-16 | 2012-03-28 | 雪佛龙美国公司 | System and method to estimate compressional to shear velocity (vp/vs) ratio in a region remote from a borehole |
CN110068816A (en) * | 2019-05-10 | 2019-07-30 | 长沙理工大学 | It is a kind of based on the Gpr Signal of Via Lifting Scheme wavelet basis building method |
CN110956164A (en) * | 2019-12-31 | 2020-04-03 | 南京工大桥隧与轨道交通研究院有限公司 | Train wheel damage identification method based on noise signals |
CN112036425A (en) * | 2020-05-09 | 2020-12-04 | 中铁四局集团有限公司 | Tunnel cavity state radar spectrum image recognition model construction method and tunnel cavity state radar spectrum image recognition method |
CN112415467A (en) * | 2020-11-06 | 2021-02-26 | 中国海洋大学 | Single-vector subsurface buoy target positioning implementation method based on neural network |
CN112557510A (en) * | 2020-12-11 | 2021-03-26 | 广西交科集团有限公司 | Concrete pavement void intelligent detection device and detection method thereof |
CN112561903A (en) * | 2020-12-24 | 2021-03-26 | 中铁建设集团基础设施建设有限公司 | Temperature shrinkage crack resistance method suitable for asphalt pavement in cold region |
CN113343329A (en) * | 2021-08-05 | 2021-09-03 | 中南大学 | Method for establishing parameterized railway tunnel model |
Non-Patent Citations (5)
Title |
---|
PENG WU 等: "Spatiotemporal Monitoring and Evaluation Method for Sand-Filling of Immersed Tube Tunnel Foundation", 《APPLIED SCIENCES》 * |
SHUAI HAN 等: "A Deep Learning Based Method for the Non-Destructive Measuring of Rock Strength through Hammering Sound", 《APPLIED SCIENCES》 * |
WEIDONGWANG 等: "Automated crack severity level detection and classification for ballastless track slab using deep convolutional neural network", 《AUTOMATION IN CONSTRUCTION》 * |
张鸿飞 等: "隧道衬砌空洞的探地雷达图谱特征研究", 《河南科学》 * |
彭永恒 等: "声振法水泥混凝土路面地基脱空判定", 《黑龙江大学自然科学学报》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115015268A (en) * | 2022-05-24 | 2022-09-06 | 山东大学 | Tunnel lining defect disease diagnosis method based on laser technology |
CN115015268B (en) * | 2022-05-24 | 2023-10-20 | 山东大学 | Tunnel lining defect disease diagnosis method based on laser technology |
Also Published As
Publication number | Publication date |
---|---|
CN113988142B (en) | 2022-04-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10995606B2 (en) | Well integrity analysis using sonic measurements over depth interval | |
CN100416269C (en) | Non destructive detection mothod used for anchor rod anchored system | |
CN106556647B (en) | A kind of impact echo data processing method | |
CN108519596A (en) | A kind of section of jurisdiction Hidden Fracture recognition methods based on match tracing and wavelet transformation | |
CN103592690A (en) | Method for automatically recognizing reservoir cracks based on electric imaging logging porosity spectrum information | |
CN107132334A (en) | Rock physical and mechanic parameter intelligent integral test system and its method of testing | |
CN113988142B (en) | Tunnel lining cavity acoustic identification method based on convolutional neural network | |
CN110954033A (en) | Concrete crack depth detection method and system | |
CN104181233A (en) | B ultrasound scanning detection method of strain clamp crimping defect based on feature enhancement | |
CN113685188B (en) | TBM tunneling optimization method based on physical characteristics of rock slag | |
US11789173B1 (en) | Real-time microseismic magnitude calculation method and device based on deep learning | |
CN111364981A (en) | Method for measuring near-bit lithology while drilling and system for monitoring lithology while drilling | |
CN112593922A (en) | Method and device for evaluating cementing quality of two well cementation interfaces through array acoustic logging | |
CN109403958B (en) | Method for identifying formation fluid properties by using gas logging power spectrum equivalent index | |
CN116378648A (en) | Near-bit stratum detection method and device based on while-drilling acoustic wave forward looking | |
CN106501285B (en) | The equipment and detection method of the mud jacking compactness of non-destructive testing prestress pipe | |
CN113468804B (en) | Underground pipeline identification method based on matrix bundles and deep neural network | |
CN107506556A (en) | A kind of short-cut method for determining fresh intact rock sound wave velocity of longitudinal wave value | |
CN116559289B (en) | Submarine substrate measuring device and method | |
CN109029711B (en) | Dynamic bridge structure multi-order frequency identification method | |
CN116184401A (en) | System and method for engineering quality inspection | |
CN112581456B (en) | Image crack real-time extraction method suitable for ultrasonic imaging logging | |
CN109100243A (en) | A kind of reinforced direct shear test detection method | |
CN112949977A (en) | Intelligent supervision management system for periodic acceptance check of engineering project quality based on big data and artificial intelligence | |
Khazanovich et al. | Nondestructive analysis of alkali-silica reaction damage in concrete slabs using shear waves |
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 |