CN114594158A - Tunnel lining cavity acoustic identification method based on long-time memory neural network - Google Patents
Tunnel lining cavity acoustic identification method based on long-time memory neural network Download PDFInfo
- Publication number
- CN114594158A CN114594158A CN202111607455.6A CN202111607455A CN114594158A CN 114594158 A CN114594158 A CN 114594158A CN 202111607455 A CN202111607455 A CN 202111607455A CN 114594158 A CN114594158 A CN 114594158A
- Authority
- CN
- China
- Prior art keywords
- sound pressure
- time
- frequency
- tunnel lining
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 30
- 230000015654 memory Effects 0.000 title claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000004458 analytical method Methods 0.000 claims abstract description 26
- 238000012545 processing Methods 0.000 claims abstract description 15
- 239000013598 vector Substances 0.000 claims abstract description 14
- 238000012360 testing method Methods 0.000 claims abstract description 12
- 230000005284 excitation Effects 0.000 claims abstract description 10
- 230000009471 action Effects 0.000 claims abstract description 9
- 238000005070 sampling Methods 0.000 claims description 29
- 239000011159 matrix material Substances 0.000 claims description 24
- 238000007781 pre-processing Methods 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 18
- 238000010276 construction Methods 0.000 claims description 10
- 230000008859 change Effects 0.000 claims description 3
- 238000013527 convolutional neural network Methods 0.000 claims description 3
- 230000003595 spectral effect Effects 0.000 claims 3
- 239000011800 void material Substances 0.000 claims 2
- 238000001228 spectrum Methods 0.000 abstract description 12
- 238000005516 engineering process Methods 0.000 abstract description 2
- 238000003062 neural network model Methods 0.000 abstract 2
- 238000001514 detection method Methods 0.000 description 16
- 230000000694 effects Effects 0.000 description 6
- 239000011435 rock Substances 0.000 description 6
- 238000004519 manufacturing process Methods 0.000 description 5
- 238000009826 distribution Methods 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 4
- 238000012423 maintenance Methods 0.000 description 3
- 238000005086 pumping Methods 0.000 description 3
- 101000709114 Homo sapiens SAFB-like transcription modulator Proteins 0.000 description 2
- 102100032664 SAFB-like transcription modulator Human genes 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 239000004568 cement Substances 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 238000009415 formwork Methods 0.000 description 2
- 230000005484 gravity Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000005422 blasting Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000000875 corresponding effect Effects 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 238000002592 echocardiography Methods 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000008014 freezing Effects 0.000 description 1
- 238000007710 freezing Methods 0.000 description 1
- 239000003673 groundwater Substances 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 238000010079 rubber tapping Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000004575 stone Substances 0.000 description 1
- 238000001308 synthesis method Methods 0.000 description 1
- 238000009423 ventilation Methods 0.000 description 1
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
- G01N29/04—Analysing solids
- G01N29/045—Analysing solids by imparting shocks to the workpiece and detecting the vibrations or the acoustic waves caused by the shocks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/04—Analysing solids
- G01N29/12—Analysing solids by measuring frequency or resonance of acoustic waves
-
- 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
- G01N29/14—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 using acoustic emission techniques
-
- 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
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2291/00—Indexing codes associated with group G01N29/00
- G01N2291/02—Indexing codes associated with the analysed material
- G01N2291/023—Solids
- G01N2291/0232—Glass, ceramics, concrete or stone
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2291/00—Indexing codes associated with group G01N29/00
- G01N2291/02—Indexing codes associated with the analysed material
- G01N2291/028—Material parameters
- G01N2291/0289—Internal structure, e.g. defects, grain size, texture
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2291/00—Indexing codes associated with group G01N29/00
- G01N2291/26—Scanned objects
- G01N2291/269—Various geometry objects
- G01N2291/2698—Other discrete objects, e.g. bricks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Health & Medical Sciences (AREA)
- Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Acoustics & Sound (AREA)
- Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
Abstract
The invention discloses a tunnel lining cavity acoustic recognition method based on a long-time memory neural network, which comprises the steps of firstly, acquiring echo sound pressure time domain information generated by a lining cavity region and a non-cavity region under the action of external excitation; and then, carrying out standardization processing on the time-domain sound pressure data samples of the two working conditions by using a data standardization program. And then, carrying out time-frequency analysis on the two working condition standardized time domain data by using a short-time Fourier transform technology, acquiring sound pressure amplitude information of sound pressure information of the two working conditions in two dimensions of a time domain and a frequency domain, and acquiring a frequency spectrum characteristic vector of the sound pressure information which changes along with time and reflects the physical characteristics of the sound pressure. And finally, constructing a neural network model based on long-time memory, using the frequency spectrum characteristic vectors of sound pressure at different time points for training and testing the neural network model, completing model parameter adjustment work, and obtaining a mature lining cavity identification model. Finally, whether cavities appear in the lining is accurately judged through analysis of new sample data.
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 long-time memory neural network.
Background
In the construction and service stage, a 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 damage and the influence of the cavities at the back of the lining are particularly serious.
The concrete reasons for the occurrence of the hollow holes in the lining structure mainly include 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, due to the reasons of insufficient power of pumping concrete, poor concrete fluidity, too early pumping pipe drawing and pulling and the like, the concrete is in an unsaturated state, and further the lining of the tunnel 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 arch crown, relative displacement is possibly generated on the vertical wall, the arch crown is sunk, the arch crown lining concrete is sunk and hollowed, and then a cavity is formed.
The raw materials used in the tunnel construction process are not strictly controlled, the concrete shrinkage is too large due to improper sand-stone gradation, too large cement dosage, too large concrete water-cement ratio, improper temperature difference and ventilation control and the like, so that the lining concrete sinks and voids, and cavities are formed.
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 surrounding rocks all will change, and the fracture takes place easily in the lining upper 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 along cavity and fracture, and then leads to aggravating of seepage phenomenon, and then lead to freezing damage 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 cavity of the lining structure of the tunnel 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 structure comprise the following steps: the method comprises a knocking echo detection method based on acoustics, a geological radar method and an ultrasonic echo synthesis method.
The geological radar method and the ultrasonic echo comprehensive method have the advantages of automation, rapidness, no loss, 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 construction quality can not be ensured. Due to the existence of the contradiction, the geological radar method and the ultrasonic echo comprehensive method have high omission factor in the tunnel cavity detection process, and the condition of the tunnel cavity 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 processes of initial detection, rechecking and the like, 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 sound and make corresponding marks. In the rechecking process, a maintainer adopts a core drilling method to visually detect whether the tunnel is empty or not. The 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 for experience of engineering personnel in the initial detection process.
Disclosure of Invention
The invention provides a tunnel lining cavity acoustic recognition method based on a long-time memory neural network.
The technical scheme provided by the invention is as follows:
on one hand, the tunnel lining cavity acoustic identification method based on the long-time memory neural network comprises the following steps:
s1: acquiring sound pressure time domain data of the tunnel lining structure under the action of external excitation;
s2: sound pressure time domain data standardization preprocessing;
s3: carrying out short-time Fourier transform analysis on the standardized sound pressure time domain data to obtain sound pressure time-frequency characteristics;
s4: acquiring a sound pressure time-frequency characteristic matrix;
s5: setting an over-parameter and a loss function of a tunnel lining cavity acoustic recognition model based on a long-time memory neural network;
s6: processing the historical data according to S1-S4 to obtain a sound pressure time-frequency characteristic matrix sample, and performing model training and parameter adjustment;
s7: and (3) lining the tunnel to be identified, processing according to S1-S4 to obtain a sound pressure time-frequency characteristic matrix, and inputting the trained tunnel lining cavity acoustic identification model based on the long-time and short-time memory neural network to identify the cavity.
Further, external excitation is applied to the tunnel, the lining structure is made to vibrate and sound, and sound pressure time domain information of the cavity region and the non-cavity region is collected through a microphone sound pressure sensor.
The sampling frequency was set to 48kHz and the sampling time was set to 2.5s per tap test.
The process can be completed in the grouting maintenance stage of the tunnel lining cavity, and in the period, engineering personnel can already determine which areas have the tunnel cavity and which areas do not have the tunnel cavity, so that the accuracy of data is better.
Further, the sound pressure time domain data standardization preprocessing means that a peak point of sound pressure information collected within a sampling interval time is determined and extracted, then 4800 sampling points before the peak point and 57600 sampling points after the peak point are intercepted, and 62400 sampling points are counted and used as sound pressure time domain data after standardization preprocessing.
The data preprocessing work ensures that the data section contains the whole sound production process on one hand, and on the other hand, the sound pressure information is standardized by lower errors through alignment of peak values of the data.
Further, the short-time fourier transform analysis of the normalized sound pressure time domain data is performed according to the following formula:
A Hamming window is adopted in short-time Fourier analysis, the window length selected in the process is 1024 sampling points, and the window shift is set as 512 sampling points;
further, the frequency spectrum characteristics of the sound pressure within each sampling interval of 0.02s are obtained through short-time Fourier transform, each group of frequency spectrum characteristic data is the frequency spectrum characteristic vector of the sound pressure information at each time point, and the characteristic vectors are combined to obtain the characteristic matrix of the sound pressure information at all time periods.
Further, the hyper-parameter setting of the tunnel lining cavity acoustic recognition model based on the long-term and short-term memory neural network is as follows: the training round epoch is 80 rounds, the initial learning rate is set to be 0.01, the learning rate is updated by adopting an exponential decay method, and a cross entropy loss function is adopted as a loss function;
in the training process, the feature vectors contained in the feature matrix are input into the LSTM in the tunnel lining cavity acoustic recognition model based on the long-short time memory neural network one by one according to a time sequence, and all parameters in the tunnel lining cavity acoustic recognition model based on the long-short time memory neural network are updated by adopting a root-mean-square back propagation algorithm.
I.e., each time a training round (epoch) is completed, the learning rate decays to 0.9 times the last learning rate.
Furthermore, a frequency band division mode is adopted to take an average value of 10 frequency parameters at intervals, so that the number of the frequency parameters is reduced, and the bandwidth is 10 Hz.
In the technical scheme of the invention, the frequency parameter is analyzed to 8000Hz, an average value is taken at intervals of 10, namely, 8000 parameters are reduced by 10 times by adopting a frequency band division mode, and the bandwidth is 10 Hz. Through the arrangement, on one hand, the analyzed frequency components are ensured to contain necessary high frequency, on the other hand, the number of parameters is reasonably reduced, and further, the calculation speed of the model is improved.
In another aspect, a tunnel lining cavity acoustic identification system based on a convolutional neural network is characterized by comprising:
sound pressure time domain signal acquisition unit: the system is used for acquiring sound pressure time domain 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 short-time Fourier transform analysis on the sound pressure time-domain signal data subjected to the standardized preprocessing to obtain sound pressure time-frequency characteristics and construct a sound pressure time-frequency characteristic matrix;
an identification model construction unit: setting an over-parameter and a loss function of a tunnel lining cavity acoustic recognition model based on a long-time memory neural network;
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 using a sound pressure time-frequency characteristic matrix and an identification label of the training sample as input information and output information of a tunnel lining cavity acoustic identification model based on a long-time memory neural network;
an identification unit: and (3) carrying out hole recognition on a sound pressure time-frequency characteristic matrix obtained by processing a tunnel lining calling sound pressure time-domain signal acquisition unit, a signal preprocessing unit and a signal time-frequency analysis unit by using the trained tunnel lining hole acoustic recognition model based on the long-time and short-time memory neural network.
Advantageous effects
Compared with the prior art, the invention has the advantages that:
(1) dependence on expert diagnosis experience and complex signal processing is eliminated in the tunnel lining cavity identification process;
(2) combining two steps of feature extraction, cavity identification and the like in the identification process of the tunnel lining cavity;
(3) the method has higher accuracy and reliability for identifying the cavity of the tunnel lining structure and has better applicability.
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 time domain data of original sound pressure, wherein (a) is time domain data of vibration sound pressure of a cavity region, and (b) is time domain data of vibration sound pressure of a non-cavity region;
FIG. 4 is normalized data, wherein (a) is a normalized time domain graph of sound pressure in the cavity region, and (b) is a normalized time domain graph of sound pressure in the normal region;
FIG. 5 is a plot of discrete time point acoustic pressure spectra, wherein (a) the acoustic pressure spectra for a cavity regime and (b) the acoustic pressure spectra for a normal regime; (ii) a
Fig. 6 is a schematic diagram of a tunnel lining cavity acoustic recognition model structure and training based on a long-time and short-time memory 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 long-time and short-time memory neural network includes the following steps:
s1: acquiring sound pressure time domain data of the tunnel lining structure under the external excitation action;
and applying external excitation to the tunnel to enable the lining structure to vibrate and sound, and acquiring sound pressure time domain information of the cavity region and the non-cavity region by using a microphone sound pressure sensor.
The sampling frequency is set to be 48kHz, the sampling time of each knocking test is set to be 2.5s, and 100500 groups of time-domain sound pressure data are obtained under two working conditions of a hollow area and a non-hollow area.
The process can be completed in the grouting maintenance stage of the tunnel lining cavity, and in the period, engineering personnel can already determine which areas have the tunnel cavity and which areas do not have the tunnel cavity, so that the accuracy of data is better.
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 characteristics of the knocking echoes of all areas of the tunnel, the positions of the knocking points are selected according to the distribution density of the holes, one knocking point is selected at intervals of 10 degrees in the area with 0-30 degrees on the single side of the vault, and one knocking point is selected at 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 the working condition of a hollow, a knocking test is completed on one section under the normal working condition, and each measuring point is subjected to a knocking test for 100 times. Fig. 3 shows the data of the vibration sound pressure generated by the external excitation in the hollow region and the non-hollow region of the #1 measuring point.
S2: sound pressure time domain data standardization preprocessing;
the sound pressure time domain data standardization preprocessing means that a peak point of sound pressure information collected in sampling interval time is determined and extracted firstly, then 4800 sampling points before the peak point and 57600 sampling points after the peak point are intercepted, and 62400 sampling points are counted and used as sound pressure time domain data after standardization preprocessing.
The data preprocessing work ensures that the data section contains the whole sound production process on one hand, and on the other hand, the sound pressure information is standardized by lower errors through alignment of peak values of the data.
The time-domain sound pressure data standardization processing program is compiled by utilizing matlab, firstly, a section of peak value point of sound pressure information is determined and extracted, then, 4800 sampling points in front of the peak value point are intercepted, 57600 sampling points behind the peak value point are intercepted, 62400 sampling points are counted in total, and the duration is 1.3s in total. The data preprocessing work ensures that the data section contains the whole sound production process on one hand, and on the other hand, the data section is aligned with the peak value of the data, so that the sound pressure information is standardized with a low error. The normalized sound pressure time domain data is shown in fig. 4.
S3: carrying out short-time Fourier transform analysis on the standardized sound pressure time domain data to obtain sound pressure time-frequency characteristics;
and performing time-frequency analysis on the sound pressure time-domain data by adopting short-time Fourier transform to obtain the frequency amplitude (energy characteristic) of sound pressure information at each discrete time point so as to obtain the frequency spectrum characteristic of the sound pressure at each time point in the whole sound production process. In the short-time fourier analysis, a hamming window is used, in which the window length is 1024 samples, and the window shift is set to 512 samples. The spectrogram of discrete time points obtained in this process is shown in fig. 5.
S4: acquiring a sound pressure time-frequency characteristic matrix;
and (3) acquiring the frequency spectrum characteristics of sound pressure in each small time period (0.02s) through short-time Fourier transform, forming the characteristic vector of each time point by each group of frequency spectrum characteristic data, and combining the characteristic vectors to obtain a characteristic matrix of sound pressure information in the whole time period.
S5: setting a hyper-parameter and a loss function of a tunnel lining cavity acoustic recognition model based on a long-time and short-time memory neural network;
on the aspect of LSTM model setting, on one hand, the excellent capability of the LSTM model on time domain data analysis, extraction and identification is fully exerted, and on the other hand, the LSTM optimization setting is completed by combining the self-generated characteristics and requirements of the tunnel knocking echo.
The vibration and sound production process of the tunnel structure under the knocking action is about 1.3s approximately, the technical scheme of the invention adopts a 48000Hz sampling frequency, and the obtained echo sample is preprocessed to have an actual length of 1.3s and comprise 62400 sampling points. When the time-frequency analysis is carried out on the sample data by using the short-time Fourier technology, the effect is best when the length of the comparison window is set to be 1024 (namely the analysis time length of 0.021 s). The final piece of sample data may be divided into 61 hour periods.
Therefore, in the setting of the LSTM model, the number of the SLTM neurons is set to be 61, and the number of the SLTM neurons is matched with the frequency characteristic matrix of the data sample, so that better coordination and analysis effects are achieved. Since the distribution characteristics of the percussive echo in the frequency domain are strongly correlated with time, a strict one-way transmission arrangement is adopted for the LSTM in the patent, and the order and connection among the neurons are strictly limited.
S6: processing the historical data according to S1-S4 to obtain a sound pressure time-frequency characteristic matrix sample, and performing model training and parameter adjustment;
firstly, feature vectors which are obtained by short-time Fourier processing and reflect sound pressure frequency spectrum characteristics in different periods are input into an LSTM network model one by one according to a time sequence order and used for training the network model. And performing short-time Fourier analysis on the time domain signal of each 0.02s hour period to obtain the distribution characteristics of the sound pressure energy under the frequency domain scale. In the technical scheme, 8000Hz is analyzed, the frequency is averaged every 10 intervals, namely, 8000 parameters are reduced by 10 times by adopting a frequency band division mode, and the bandwidth is 10 Hz. Through the arrangement, on one hand, the analyzed frequency components are guaranteed to contain necessary high frequency, on the other hand, the number of parameters is reasonably reduced, and further the calculation speed of the model is improved. Finally each feature vector contains 800 elements as input x for each input LSTM neuront. At the input x of the first LSTM neuron1Is the first 0.021s frequency domain feature vector, c0And h0Can be set to an arbitrary value and then the feature vectors in each sample are input one by one in 61 neurons;
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 completing the adjustment of the hyper-parameters of the model in the process so as to obtain a mature prediction model.
A Graphic Processing Unit (GPU) is used as a computing core, training and parameter adjusting work of a model is completed by relying on a deep learning framework Tensorflow developed by Google, and the initial setting of the hyper-parameters is as follows: the training round (epoch) is 80 rounds, the initial learning rate is set to be 0.01 (the amplitude of each update of the model weight parameters), and the learning rate is updated by adopting an exponential decay method. I.e., each time a training round (epoch) is completed, the learning rate decays to 0.9 times the last learning rate. The model parameters are updated by adopting a root mean square back propagation algorithm (Rmrsprep) in the whole training process.
This example uses 80 sample data as the training set and 20 sample data as the test set. The training and testing effects of the model are judged through the loss function, the loss functions of the training set and the testing set gradually reach stability and convergence, when the error between the training set and the testing set is minimum, the fitting effect of the model is best at the moment, and then the final mature model is obtained. The loss function used in this patent is a cross-entropy loss function.
Wherein the cross entropy loss function for a single sample is:
the overall cross entropy loss function is:
according to the technical scheme, whether a hole exists in a knocking area or not is judged by using the knocking echo of the tunnel, and the problem belongs to a binary classification problem, and finally, the probability of the existence of the hole, namely the probability given by Softmax
And y is the probability of whether the sample is a hole, a hole is 1, and a non-hole is 0. The data value of L is always greater than or equal to 0, and the prediction effect is better when L is closer to 0, and the prediction value is close to the true value. As the number of training increases, the value of L gradually decreases and tends to stabilize. And when the overall cross entropy loss L tends to be stable, the model training is finished.
S7: and (3) the tunnel lining to be identified is processed according to S1-S4, and the obtained sound pressure time-frequency characteristic matrix is input into a trained tunnel lining cavity acoustic identification model based on a long-time memory neural network to identify the cavity.
A tunnel lining cavity acoustic identification system based on a convolutional neural network comprises:
a sound pressure time domain signal acquisition unit: the system is used for acquiring sound pressure time domain 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 short-time Fourier transform analysis on the sound pressure time-domain signal data subjected to the standardized preprocessing to obtain sound pressure time-frequency characteristics and construct a sound pressure time-frequency characteristic matrix;
an identification model construction unit: setting an over-parameter and a loss function of a tunnel lining cavity acoustic recognition model based on a long-time memory neural network;
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 using a sound pressure time-frequency characteristic matrix and an identification label of the training sample as input information and output information of a tunnel lining cavity acoustic identification model based on a long-time memory neural network;
an identification unit: and (3) carrying out hole recognition on a sound pressure time-frequency characteristic matrix obtained by processing a tunnel lining calling sound pressure time-domain signal acquisition unit, a signal preprocessing unit and a signal time-frequency analysis unit by using the trained tunnel lining hole acoustic recognition model based on the long-time and short-time memory 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 (8)
1. A tunnel lining cavity acoustic recognition method based on a long-and-short-term memory neural network is characterized by comprising the following steps:
s1: acquiring sound pressure time domain data of the tunnel lining structure under the external excitation action;
s2: sound pressure time domain data standardization preprocessing;
s3: carrying out short-time Fourier transform analysis on the standardized sound pressure time domain data to obtain sound pressure time-frequency characteristics;
s4: acquiring a sound pressure time-frequency characteristic matrix;
s5: setting an over-parameter and a loss function of a tunnel lining cavity acoustic recognition model based on a long-time memory neural network;
s6: processing the historical data according to S1-S4 to obtain a sound pressure time-frequency characteristic matrix sample, and performing model training and parameter adjustment;
s7: and (3) the tunnel lining to be identified is processed according to S1-S4, and the obtained sound pressure time-frequency characteristic matrix is input into a trained tunnel lining cavity acoustic identification model based on a long-time memory neural network to identify the cavity.
2. The method of claim 1, wherein applying an external stimulus to the tunnel vibrates the lining structure to sound and collects sound pressure time domain information for void and non-void regions using a microphone sound pressure sensor;
the sampling frequency was set to 48kHz and the sampling time was set to 2.5s per tap test.
3. The method of claim 1, wherein the sound pressure time domain data is normalized and preprocessed by firstly determining and extracting a peak point of sound pressure information collected within a sampling interval, and then intercepting 4800 sampling points before the peak point and 57600 sampling points after the peak point for a total of 62400 sampling points as the normalized and preprocessed sound pressure time domain data.
4. The method of claim 1, wherein the short-time fourier transform analysis of the normalized sound pressure time domain data is performed according to the following equation:
5. The method of claim 4, wherein the spectral features of the sound pressure within each sampling interval of 0.02s are obtained by short-time Fourier transform, each set of spectral feature data is the spectral feature vector of the sound pressure information at each time point, and the feature vectors are combined to obtain the feature matrix of the sound pressure information at all time intervals.
6. The method according to claim 1, wherein the hyper-parameters of the acoustic recognition model of the tunnel lining cavity based on the long-time memory neural network are set as follows: the training round epoch is 80 rounds, the initial learning rate is set to be 0.01, the learning rate is updated by adopting an exponential decay method, and a cross entropy loss function is adopted as a loss function;
in the training process, the feature vectors contained in the feature matrix are input into the LSTM in the tunnel lining cavity acoustic recognition model based on the long-short time memory neural network one by one according to a time sequence, and all parameters in the tunnel lining cavity acoustic recognition model based on the long-short time memory neural network are updated by adopting a root-mean-square back propagation algorithm.
7. The method of claim 1, wherein the frequency parameters are averaged once every 10 intervals by adopting a frequency band division mode, so that the number of the frequency parameters is reduced, and the bandwidth is 10 Hz.
8. A tunnel lining cavity acoustic recognition system based on a convolutional neural network is characterized by comprising:
sound pressure time domain signal acquisition unit: the system is used for acquiring sound pressure time domain 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 short-time Fourier transform analysis on the sound pressure time domain signal data subjected to the standardized preprocessing to obtain sound pressure time-frequency characteristics and construct a sound pressure time-frequency characteristic matrix;
an identification model construction unit: setting an over-parameter and a loss function of a tunnel lining cavity acoustic recognition model based on a long-time memory neural network;
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 using a sound pressure time frequency characteristic matrix and an identification label of the training sample as input information and output information of a tunnel lining cavity acoustic identification model based on a long-time and short-time memory neural network respectively;
an identification unit: and (3) carrying out hole recognition on a sound pressure time-frequency characteristic matrix obtained by processing a tunnel lining calling sound pressure time-domain signal acquisition unit, a signal preprocessing unit and a signal time-frequency analysis unit by using the trained tunnel lining hole acoustic recognition model based on the long-time and short-time memory neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111607455.6A CN114594158A (en) | 2021-12-27 | 2021-12-27 | Tunnel lining cavity acoustic identification method based on long-time memory neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111607455.6A CN114594158A (en) | 2021-12-27 | 2021-12-27 | Tunnel lining cavity acoustic identification method based on long-time memory neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114594158A true CN114594158A (en) | 2022-06-07 |
Family
ID=81814268
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111607455.6A Pending CN114594158A (en) | 2021-12-27 | 2021-12-27 | Tunnel lining cavity acoustic identification method based on long-time memory neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114594158A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110045016A (en) * | 2019-04-24 | 2019-07-23 | 四川升拓检测技术股份有限公司 | A kind of tunnel-liner lossless detection method based on audio analysis |
| CN111261146A (en) * | 2020-01-16 | 2020-06-09 | 腾讯科技（深圳）有限公司 | Speech recognition and model training method, device and computer readable storage medium |
| CN113111786A (en) * | 2021-04-15 | 2021-07-13 | 西安电子科技大学 | Underwater target identification method based on small sample training image convolutional network |
| CN113988142A (en) * | 2021-12-27 | 2022-01-28 | 中南大学 | Tunnel lining cavity acoustic identification method based on convolutional neural network |
-
2021
- 2021-12-27 CN CN202111607455.6A patent/CN114594158A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110045016A (en) * | 2019-04-24 | 2019-07-23 | 四川升拓检测技术股份有限公司 | A kind of tunnel-liner lossless detection method based on audio analysis |
| CN111261146A (en) * | 2020-01-16 | 2020-06-09 | 腾讯科技（深圳）有限公司 | Speech recognition and model training method, device and computer readable storage medium |
| CN113111786A (en) * | 2021-04-15 | 2021-07-13 | 西安电子科技大学 | Underwater target identification method based on small sample training image convolutional network |
| CN113988142A (en) * | 2021-12-27 | 2022-01-28 | 中南大学 | Tunnel lining cavity acoustic identification method based on convolutional neural network |
Non-Patent Citations (1)
| Title |
|---|
| 窦顺;贺磊;郑静;王念秦;: "隧道二衬脱空声振检测试验研究", 铁道工程学报, no. 07, 15 July 2017 (2017-07-15), pages 68 - 73 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100416269C (en) | Non destructive detection mothod used for anchor rod anchored system | |
| CN102736124A (en) | Tunnel excavation surrounding rock dynamic refined classification method based on integrated parameters | |
| CN113988142B (en) | Tunnel lining cavity acoustic identification method based on convolutional neural network | |
| CN109725366B (en) | Method and system for positioning rainwater blocking point | |
| US11789173B1 (en) | Real-time microseismic magnitude calculation method and device based on deep learning | |
| CN109034641A (en) | Defect of pipeline prediction technique and device | |
| CN106383172A (en) | Surrounding rock damage prediction method based on energy release coefficient | |
| CN111042866B (en) | Multi-physical-field cooperative water inrush monitoring method | |
| CN106952003A (en) | High Ground Stress Areas beded rock mass underground rock cavern Failure type Forecasting Methodology | |
| CN110924457A (en) | Foundation pit deformation monitoring method and system based on measuring robot | |
| CN103278843A (en) | Rockburst real-time forecasting technique device used in process of rock tunnel construction | |
| CN104614144B (en) | Flood-discharge energy-dissipating induces the Forecasting Methodology of place vibration | |
| CN106501285A (en) | The equipment of the mud jacking degree of compaction of Non-Destructive Testing prestress pipe and detection method | |
| CN114594158A (en) | Tunnel lining cavity acoustic identification method based on long-time memory neural network | |
| CN116882023A (en) | Method for predicting transverse settlement of upper earth covering layer in underground excavation construction of subway tunnel | |
| CN103953024B (en) | Foundation ditch automatic monitoring disorder data recognition method | |
| CN108278109B (en) | Method, equipment and system for determining reinforcement time of weakened surrounding rock of underground engineering | |
| CN105807321A (en) | Rock mass structure analysis and electromagnetic radiation monitoring combined rock burst prediction method | |
| CN114740089A (en) | Foundation pile low-strain signal acquisition system | |
| CN112946778A (en) | Method for early warning karst collapse based on underground water turbidity monitoring | |
| CN110909402B (en) | Advanced small catheter design method based on neural network technology | |
| CN115389405B (en) | Method and device for monitoring health state of externally hung flower bed of viaduct | |
| CN116630676B (en) | Large-scale-range field classification processing method and device and electronic equipment | |
| Wan et al. | From data to decision: combining Bayesian updating with a data-driven prior to forecast the settlement of embankments on soft soils | |
| RU2809469C1 (en) | Method and system for seismoacoustic monitoring of rock mass |
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 |