CN114324580A - Intelligent knocking detection method and system for structural defects - Google Patents
Intelligent knocking detection method and system for structural defects Download PDFInfo
- Publication number
- CN114324580A CN114324580A CN202111460599.3A CN202111460599A CN114324580A CN 114324580 A CN114324580 A CN 114324580A CN 202111460599 A CN202111460599 A CN 202111460599A CN 114324580 A CN114324580 A CN 114324580A
- Authority
- CN
- China
- Prior art keywords
- detected
- signal
- detection model
- knocking
- intelligent detection
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 114
- 230000007847 structural defect Effects 0.000 title claims abstract description 14
- 230000005236 sound signal Effects 0.000 claims abstract description 60
- 230000007547 defect Effects 0.000 claims abstract description 26
- 238000007781 pre-processing Methods 0.000 claims abstract description 24
- 238000012549 training Methods 0.000 claims description 33
- 238000000034 method Methods 0.000 claims description 31
- 238000012360 testing method Methods 0.000 claims description 17
- 238000010079 rubber tapping Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 8
- 230000007246 mechanism Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 150000001875 compounds Chemical class 0.000 claims description 3
- 230000006870 function Effects 0.000 description 14
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000000605 extraction Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 230000008520 organization Effects 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000002950 deficient Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 239000011800 void material Substances 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003116 impacting effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000009659 non-destructive testing Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
Images
Classifications
-
- 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
- Y02B—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO BUILDINGS, e.g. HOUSING, HOUSE APPLIANCES OR RELATED END-USER APPLICATIONS
- Y02B10/00—Integration of renewable energy sources in buildings
- Y02B10/30—Wind power
Landscapes
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The utility model discloses an intelligent knocking detection method for structural defects, which comprises the following steps: collecting a knocking sound signal of a structure to be detected, and preprocessing the knocking sound signal of the structure to be detected; inputting the preprocessed knocking sound signal of the structure to be detected into the trained intelligent detection model for detection so as to obtain the defect category of the structure to be detected. The present disclosure also provides a structural defect's intelligence hits detecting system, includes: the signal collecting unit is used for collecting the knocking sound signal of the structure to be detected; the signal preprocessing unit is used for preprocessing the knocking sound signal of the structure to be detected; and the signal detection unit is used for inputting the preprocessed knocking sound signal of the structure to be detected into the trained intelligent detection model for detection so as to obtain the defect type of the structure to be detected.
Description
Technical Field
The disclosure belongs to the field of structural defect detection, and particularly relates to an intelligent knocking detection method and system for structural defects.
Background
With the development of socioeconomic and engineering technologies, large buildings and precision equipment including high-speed railways, super high-rise buildings, large bridges and tunnels, wind turbines, etc. are increasingly being constructed and put into practical use. During the production, construction and service of these buildings and equipment, the detection of the structural integrity of its internal main load-bearing members is key to ensure the safe service thereof.
Nondestructive testing is a common technical means for inspecting internal damage or defects of a structure to be tested on the premise of not damaging the structure; the knock detection is one of the most basic, easiest to implement and widest application range nondestructive detection methods for the near-surface defects in the structure. In practice, the work is usually completed by experienced technical workers, however, the method has the problems of low work efficiency, unstable detection quality, high training cost of the technical workers, small number of high-grade workers and the like. As more and more large buildings and precision equipment are built and put into practical use, the knocking detection requirement in production is gradually expanded, so that the method for implementing knocking detection by taking technical workers as a main body cannot meet the production requirement, and an automatic and intelligent technical means capable of breaking away from manual knocking detection is urgently needed.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide an intelligent knocking detection method for structural defects, which is based on a deep learning technology, does not need to artificially design signal characteristic parameters, reduces the dependence on expert knowledge, and can automatically learn an extraction and organization method of knocking detection acoustic signals from original acoustic signal samples.
In order to achieve the above purpose, the present disclosure provides the following technical solutions:
an intelligent knocking detection method for structural defects comprises the following steps:
s100: collecting and preprocessing knocking sound signals of a structure to be detected;
s200: inputting the preprocessed knocking sound signal of the structure to be detected into the trained intelligent detection model for detection so as to identify the defect type of the structure to be detected.
Preferably, the intelligent detection model adopts a one-dimensional convolutional neural network, which is composed of n combined blocks, specifically expressed as:
in the formula, M represents a one-dimensional convolutional neural network; f. ofiRepresents the ith combined block, i takes the value of [1, n]The value of n is selected according to the application scene of the intelligent detection model;
each combination block is represented as:
in the formula, the compound is shown in the specification,
representing a complex function operator; flRepresenting the input tensor of the l-th layer, l taking the value of [1, n]An integer of the interval; conv denotes a one-dimensional convolution transform layer; ReLU denotes the activation function layer; BN represents a batch regularization layer; pool denotes pooling layer.
Preferably, in step S200, the training process of the intelligent detection model includes:
s201: obtaining a plurality of structural knocking sound signal samples with known defect types, preprocessing the samples, and dividing the preprocessed sound signal samples into a training set, a verification set and a test set;
s202: inputting the training set into an intelligent detection model to train the model, inputting the verification set into the trained intelligent detection model to verify the model, taking the accuracy of the intelligent detection model in identifying and verifying the sample types in the set as a judgment standard, and finishing the training of the intelligent detection model when the accuracy is not improved any more;
s203: inputting the test set into the trained intelligent detection model for testing, taking the accuracy of the intelligent detection model for identifying the sample types in the test set as a judgment standard, if the accuracy meets a preset value, obtaining a final intelligent detection model, otherwise, optimizing the structural parameters and the training configuration parameters of the intelligent detection model until the accuracy requirements are met.
Preferably, in step S202, the step of inputting the training set into the intelligent detection model to train the model specifically includes the following steps:
s2021: calculating a loss function of the training set;
s2022: calculating the parameter gradient of the intelligent detection model through back propagation;
s2023: and updating the parameters of the intelligent detection model according to the parameter gradient.
Preferably, in step S100 and step S201, the preprocessing the tapping sound signal includes the following steps:
s1000: normalizing the amplitude of the knocking sound signal;
s2000: and slicing the normalized tapping sound signal according to the standard length.
The present disclosure also provides a structural defect's intelligence hits detecting system, includes:
the signal collecting unit is used for collecting the knocking sound signal of the structure to be detected;
the signal preprocessing unit is used for preprocessing the knocking sound signal of the structure to be detected;
and the signal detection unit is used for inputting the preprocessed knocking sound signal of the structure to be detected into the trained intelligent detection model for detection so as to obtain the defect type of the structure to be detected.
Preferably, the signal detection unit includes:
the signal input and preprocessing module is used for acquiring and preprocessing the knocking sound signal of the structure to be detected;
and the defect identification module is embedded with an intelligent detection model and used for analyzing the preprocessed knocking sound signal of the structure to be detected so as to identify the defect type of the structure to be detected.
Preferably, the signal collection unit includes:
the automatic knocking mechanism is used for automatically knocking the structure to be detected so that the structure to be detected generates knocking sound signals;
and the signal collecting device is used for collecting the knocking sound signals.
Preferably, the automatic knocking mechanism comprises a moving platform, a knocking arm is arranged on the moving platform, and a knocking head is arranged on the knocking arm.
Preferably, the signal collecting device comprises an acoustic sensor and an acoustic signal collector, and the acoustic sensor is arranged on the knocking arm.
Compared with the prior art, the beneficial effect that this disclosure brought does:
1. compared with knocking detection by manual operation, the method disclosed by the invention has the advantages of high efficiency, high automation degree, stable detection quality, low operation cost and small limitation of manpower conditions;
2. compared with an automatic knocking system built based on the prior art, the intelligent model built by the method can automatically learn the extraction and organization algorithm of the knocking detection acoustic signal from the original data sample, does not need to artificially design the characteristic parameters of the signal, reduces the dependence on expert knowledge, has lower practice difficulty in engineering, and is more accurate in the identification of the knocking detection acoustic signal type.
Drawings
FIG. 1 is a typical flow of a prior art implementation of automatic identification of a tap detection acoustic signal;
FIG. 2 is a flowchart of a method for implementing automatic recognition of a tap detection acoustic signal according to an embodiment of the present disclosure;
fig. 3(a) to 3(c) are schematic diagrams of an intelligent detection model according to another embodiment of the disclosure, where fig. 3(a) is a depth flat network, fig. 3(b) is a depth residual network, and fig. 3(c) is a depth tight network;
FIG. 4 is a flow chart of a method for training a smart detection model according to another embodiment of the present disclosure;
FIG. 5 is a block diagram of a smart tap detection system for structural defects according to another embodiment of the present disclosure.
Detailed Description
Specific embodiments of the present disclosure will be described in detail below with reference to fig. 1 to 5. While specific embodiments of the disclosure are shown in the drawings, it should be understood that the disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
It should be noted that certain terms are used throughout the description and claims to refer to particular components. As one skilled in the art will appreciate, various names may be used to refer to a component. This specification and claims do not intend to distinguish between components that differ in name but not function. In the following description and in the claims, the terms "include" and "comprise" are used in an open-ended fashion, and thus should be interpreted to mean "include, but not limited to. The description which follows is a preferred embodiment of the disclosure, but is made for the purpose of illustrating the general principles of the disclosure and not for the purpose of limiting the scope of the disclosure. The scope of the present disclosure is to be determined by the terms of the appended claims.
To facilitate an understanding of the embodiments of the present disclosure, the following detailed description is to be considered in conjunction with the accompanying drawings, and the drawings are not to be construed as limiting the embodiments of the present disclosure.
In one embodiment, as shown in fig. 2, the present disclosure provides a method for detecting a structural defect by smart tap, comprising the following steps:
s100: collecting and preprocessing knocking sound signals of a structure to be detected;
s200: inputting the preprocessed knocking sound signal of the structure to be detected into the trained intelligent detection model for detection so as to identify the defect type of the structure to be detected.
In this embodiment, fig. 1 is a common way of performing tap detection using the prior art: firstly, impacting an area to be tested by using a knocking head, exciting local vibration and generating an acoustic signal; then, the generated acoustic signals are analyzed to judge the local defect condition of the current test. Generally, for knock detection by manual operation, the acoustic signal is distinguished on site by a specially trained technician; for an automatic knocking system, sound signals are collected by a sensor, signal characteristics are extracted by using a specific algorithm in a computer and are compared with a reference value to realize automatic judgment.
The existing automatic knocking detection system generally comprises the following steps for the characteristic extraction and defect identification process of the acquired signal: firstly, according to the statistical theory and the spectrum analysis method of signal analysis, a plurality of signal characteristic parameters (usually including specific definitions of more than tens of characteristic parameters) are artificially defined; then, according to expert knowledge, on one hand, characteristic parameter screening is carried out, on the other hand, a threshold value of signal pattern recognition is determined, and in actual detection, defect signals are judged according to the threshold value corresponding to the specific parameters. In addition, based on artificially defining the characteristic parameters of the signals, the more advanced prior art can further introduce special algorithms, such as a Support Vector Machine (SVM), an Artificial Neural Network (ANN), and the like, in the processes of characteristic parameter screening and signal pattern recognition. These methods finally obtain a parameter classification boundary that can identify the defect signal by learning the feature parameters that have been extracted from the signal. In conclusion, the manual knocking detection has the limitations of low construction efficiency, unstable detection quality, high training cost of technical workers and small number of high-grade workers; for the existing automatic knocking system, the limitation is that the process of artificially defining, utilizing expert knowledge characteristic parameters and setting a threshold value according to the characteristic parameters cannot fully extract the mode characteristics of the acoustic signal, so that the knocking detection signal is difficult to accurately identify; for a more advanced knocking detection signal identification method introducing a characteristic parameter screening and pattern identification special algorithm, the method is limited in that the process of artificially defining the characteristic parameters cannot fully extract the acoustic signal pattern characteristics on one hand, and on the other hand, for different specific problems, the implementation difficulty of the process of artificially defining the characteristic parameters is high, and the requirement on field expert knowledge is high.
By constructing the intelligent detection model, the method can automatically learn the extraction and organization algorithm of the knocking detection acoustic signal from the original data sample without manually designing signal characteristic parameters, reduces the dependence on expert knowledge, has lower practical difficulty in engineering, and is more accurate in identifying the type of the knocking detection acoustic signal.
In another embodiment, the intelligent detection model employs a one-dimensional convolutional neural network, which is formed by n combined blocks, specifically expressed as:
in the formula, M represents a one-dimensional convolutional neural network; f. ofiRepresents the ith combined block, i takes the value of [1, n]The value of n is selected according to the application scene of the intelligent detection model;
each combination block is represented as:
in the formula, the compound is shown in the specification,
representing a complex function operator; flRepresenting the input tensor of the l-th layer, l taking the value of [1, n]An integer of the interval; c onv represents a one-dimensional convolution transform layer; ReLU denotes the activation function layer; BN represents a batch regularization layer; pool denotes pooling layer.
In this embodiment, the intelligent detection model may be any one of a depth flat-layer network, a depth residual error network, and a depth tight-lock network, where fig. 3(a) is a schematic structural diagram of the depth flat-layer network, the network is directly built by a plurality of combination blocks, and signal classification and identification are completed by using a global average pooling, a convolution layer with a convolution kernel width of 1, and a Softmax multi-classification activation function; fig. 3(b) is a deep residual network, which adds residual connection spanning two combination blocks based on the structure of a deep flat network, and completes signal classification and identification by using global average pooling, a convolution layer with convolution kernel width of 1, and a Softmax multi-classification activation function; fig. 3(c) is a deep tight network, which uses specially designed tight blocks as basic constituent units of the network, and the convolution operation in the tight blocks connects the outputs of all pre-convolutions as inputs, and the network uses global average pooling, convolution layer with convolution kernel width of 1, and Softmax multi-class activation function to complete signal classification and identification.
The three network structures have various characteristics in the aspects of model complexity, running speed, training difficulty and the like, the signal recognition capabilities are slightly different but are generally similar, and in practice, the model basic structure is selected according to specific problems and model training effects and the structure is optimally designed. For example, for the problems with distinct features and less damage types, training and tuning on the basis of the depth flat network shown in fig. 3(a) can be selected, so that the model training is easier and the recognition speed is faster while higher recognition accuracy is achieved; for the more difficult problems of complex features and rich damage types, the deep tight network shown in fig. 3(c) with a more complex structure may be considered and optimized to obtain a better recognition accuracy.
It should be noted that the above 3 network structures are only exemplary, and the common point is that the network structures are based on a one-dimensional convolutional network, and the difference is that the number of the combined blocks and the number of layers of each combined block are different.
In another embodiment, in step S200, as shown in fig. 4, the training process of the intelligent detection model includes the following steps:
s201: obtaining a plurality of structural knocking sound signal samples with known defect types, preprocessing the samples, and dividing the preprocessed sound signal samples into a training set, a verification set and a test set;
in this step, the method for obtaining a tap sound signal sample includes: knocking a target structure, recording knocking sound signals, dissecting the target structure to determine the real damage condition of a knocking point, and establishing a corresponding relation between the sound signals and the structural damage through repeated operation; alternatively, the damage may be preformed internally when the sample structure is manufactured, so that the damage types at different positions are known in advance, and then the acoustic signals are recorded by tapping at different positions, so that the correspondence between the damage types and the acoustic signals can also be established.
S202: inputting the training set into an intelligent detection model to train the model, inputting the verification set into the trained intelligent detection model to verify the model, taking the accuracy of the intelligent detection model in identifying and verifying the sample types in the set as a judgment standard, and finishing the training of the intelligent detection model when the accuracy is not improved any more;
s203: inputting the test set into the trained intelligent detection model for testing, taking the accuracy of the intelligent detection model for identifying samples in the test set as a judgment standard, if the accuracy meets a preset value, obtaining a final intelligent detection model, otherwise, optimizing the structural parameters and the training configuration parameters of the intelligent detection model until the accuracy requirements are met.
In the step, the accuracy is generally set to be 80% -95%, if the identification of the intelligent detection model for the samples in the test set meets the accuracy in the range, the intelligent detection model can be considered to pass the test, otherwise, the structural parameters and the training configuration parameters of the intelligent detection model need to be optimized until the accuracy reaches 80% -95%.
In another embodiment, in step S202, the step of inputting the training set into the intelligent detection model to train the model specifically includes the following steps:
s2021: calculating a loss function of the training set;
in this step, for a training set { x, y }, a loss function of the training set needs to be calculated first, where x is an input signal vector, y is a label vector, and a calculation expression of the loss function is as follows:
in the formula, c represents the number of classes to be identified, Mi(x) Presentation intelligenceThe detection model predicts the probability, y, according to the class given by the input xiThe component of the label vector y has two values of 0 and 1, and if and only if the real category of the input x is the ith category, y isiTake 1, otherwise 0.
S2022: calculating the parameter gradient of the intelligent detection model through back propagation;
in the step, after obtaining the damage function of the training set, calculating the parameter gradient of the intelligent detection model layer by layer in a reverse direction by a back propagation algorithm, wherein the calculation expression is as follows:
in the formula, thetaiRepresenting the parameter to be learned of the ith layer; flRepresenting the output of the l-th layer.
S2023: and updating the parameters of the intelligent detection model according to the parameter gradient.
In this step, after obtaining the parameter gradient of the intelligent detection model through back propagation, the parameters of the intelligent detection model need to be updated according to the following formula by the mathematical optimizer O:
in the formula, xi represents a learning rate and is selected according to different optimizers; optim (-) represents the optimization function determined by the optimizer with the parameter gradient as input and the optimization direction as output.
In another embodiment, in step S100 and step S201, the preprocessing the tapping sound signal includes the following steps:
s1000: normalizing the amplitude of the knocking sound signal;
in the step, the obtained knocking sound signal is divided by the maximum value of the absolute values of the signal amplitudes in all the collected signal samples, so that the amplitude of the knocking sound signal is controlled in an interval of [ -1, 1 ].
S2000: and slicing the normalized tapping sound signal according to the standard length.
In the step, the initial end of the knocking sound signal is moved backwards by a fixed step length, and each step intercepts the signal as a sample according to a standard length, wherein the standard length is selected according to the sound signal identification requirement and takes the characteristic capable of expressing most sound signals as a standard.
The present disclosure will now describe the above method in detail by taking the bonded laminate tap test as an example.
Three different materials of glued laminated board samples are prefabricated in a laboratory, and 6 types of defects are preset inside the laminated board samples, including: large area void, large area upper debond, large area lower debond, bubble-type void, bubble-type upper debond, and bubble-type lower debond. A total of about 10 ten thousand samples of the flawed and flawless tapping sound signals belonging to the above category 6 were collected by tapping different locations on the laminate test specimens. And respectively constructing a deep flat network model, a deep residual error network model and a deep tight network model according to the steps of the method, and then respectively training the three deep learning networks by using the collected knocking sound signal samples so as to classify the collected knocking sound signals. Meanwhile, the existing artificial feature extraction and machine learning technology (SVM) is used for automatically identifying the signals. Table 1 shows the comparison of the recognition accuracy of the above 6 types of defective signals and non-defective signals by three deep learning models and the existing automatic recognition technology.
TABLE 1
From the above table, compared with the existing acoustic signal identification method, the method disclosed by the invention has the advantage that the identification accuracy of the knocking acoustic signal is more accurate.
In another embodiment, the present disclosure further provides an intelligent tap detection system for structural defects, including:
the signal collecting unit is used for collecting the knocking sound signal of the structure to be detected;
the signal preprocessing unit is used for preprocessing the knocking sound signal of the structure to be detected;
and the signal detection unit is used for inputting the preprocessed knocking sound signal of the structure to be detected into the trained intelligent detection model for detection so as to obtain the defect type of the structure to be detected.
In another embodiment, the signal detection unit includes:
the signal input and preprocessing module is used for acquiring and preprocessing the knocking sound signal of the structure to be detected;
and the defect identification module is embedded with an intelligent detection model and used for analyzing the preprocessed knocking sound signal of the structure to be detected so as to identify the defect type of the structure to be detected.
In another embodiment, as shown in fig. 4, the signal collection unit includes:
the automatic knocking mechanism is used for automatically knocking the structure to be detected so that the structure to be detected generates knocking sound signals;
and the signal collecting device is used for collecting the knocking sound signals.
In another embodiment, the automatic knocking mechanism comprises a moving platform, a knocking arm is arranged on the moving platform, and a knocking head is arranged on the knocking arm.
In the embodiment, the knocking arm is formed by modifying a multi-degree-of-freedom mechanical arm, and the knocking angle and distance are controlled by a motion program stored in the mobile platform. The knocking head is composed of a spring device or an electromagnet and a mass block, and the mass block is accelerated by the spring device or the electromagnet to impact the surface of the structure to be detected so as to generate an acoustic signal.
In another embodiment, the signal collection device comprises an acoustic sensor and an acoustic signal collector, the acoustic sensor being disposed on the strike arm.
In this embodiment, the acoustic sensor transmits an acoustic signal generated by the knocking of the automatic knocking mechanism to the acoustic signal collector for collection, and the acoustic signal collector transmits the collected acoustic signal to the computer system, where the computer system includes a memory and a processor, where the memory stores an executable program executable on the processor, and the processor executes the executable program to implement the method described in the foregoing embodiment to perform defect type identification on the received knocking acoustic signal.
The foregoing description of the present disclosure has been presented with specific examples to aid understanding thereof, and is not intended to limit the present disclosure. Any partial modification or replacement within the technical scope disclosed in the present disclosure by a person skilled in the art should be included in the scope of the present disclosure.
Claims (10)
1. An intelligent knocking detection method for structural defects comprises the following steps:
s100: collecting and preprocessing knocking sound signals of a structure to be detected;
s200: inputting the preprocessed knocking sound signal of the structure to be detected into the trained intelligent detection model for detection so as to identify the defect type of the structure to be detected.
2. The method according to claim 1, wherein preferably, the intelligent detection model employs a one-dimensional convolutional neural network, which is composed of n combined blocks, specifically expressed as:
in the formula, M represents a one-dimensional convolutional neural network; f. ofiRepresents the ith combined block, i takes the value of [1, n]The value of n is selected according to the application scene of the intelligent detection model;
each combination block is represented as:
in the formula, the compound is shown in the specification,
representing a compositionA function operator; flRepresenting the input tensor of the l-th layer, l taking the value of [1, n]An integer of the interval; conv denotes a one-dimensional convolution transform layer; ReLU denotes the activation function layer; BN represents a batch regularization layer; pool denotes pooling layer.
3. The method according to claim 1, wherein in step S200, the training process of the intelligent detection model includes:
s201: obtaining a plurality of structural knocking sound signal samples with known defect types, preprocessing the samples, and dividing the preprocessed sound signal samples into a training set, a verification set and a test set;
s202: inputting the training set into an intelligent detection model to train the model, inputting the verification set into the trained intelligent detection model to verify the model, taking the accuracy of the intelligent detection model in identifying and verifying the sample types in the set as a judgment standard, and finishing the training of the intelligent detection model when the accuracy is not improved any more;
s203: inputting the test set into the trained intelligent detection model for testing, taking the accuracy of the intelligent detection model for identifying the sample types in the test set as a judgment standard, if the accuracy meets a preset value, obtaining a final intelligent detection model, otherwise, optimizing the structural parameters and the training configuration parameters of the intelligent detection model until the accuracy requirements are met.
4. The method according to claim 3, wherein in step S202, the step of inputting the training set into the smart detection model to train the model specifically includes the following steps:
s2021: calculating a loss function of the training set;
s2022: calculating the parameter gradient of the intelligent detection model through back propagation;
s2023: and updating the parameters of the intelligent detection model according to the parameter gradient.
5. The method according to claim 3, wherein the preprocessing the tapping sound signal in steps S100 and S201 comprises the steps of:
s1000: normalizing the amplitude of the knocking sound signal;
s2000: and slicing the normalized tapping sound signal according to the standard length.
6. An intelligent tap detection system for structural defects, comprising:
the signal collecting unit is used for collecting the knocking sound signal of the structure to be detected;
the signal preprocessing unit is used for preprocessing the knocking sound signal of the structure to be detected;
and the signal detection unit is used for inputting the preprocessed knocking sound signal of the structure to be detected into the trained intelligent detection model for detection so as to obtain the defect type of the structure to be detected.
7. The system of claim 6, wherein the signal detection unit comprises:
the signal input and preprocessing module is used for acquiring and preprocessing the knocking sound signal of the structure to be detected;
and the defect identification module is embedded with an intelligent detection model and used for analyzing the preprocessed knocking sound signal of the structure to be detected so as to identify the defect type of the structure to be detected.
8. The system of claim 7, wherein the signal collection unit comprises:
the automatic knocking mechanism is used for automatically knocking the structure to be detected so that the structure to be detected generates knocking sound signals;
and the signal collecting device is used for collecting the knocking sound signals.
9. The system of claim 8, wherein the automatic tapping mechanism comprises a mobile platform having a tapping arm with a tapping head disposed thereon.
10. The system of claim 9, wherein the signal collection device comprises an acoustic sensor and an acoustic signal collector, the acoustic sensor being disposed on the strike arm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111460599.3A CN114324580A (en) | 2021-12-03 | 2021-12-03 | Intelligent knocking detection method and system for structural defects |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111460599.3A CN114324580A (en) | 2021-12-03 | 2021-12-03 | Intelligent knocking detection method and system for structural defects |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114324580A true CN114324580A (en) | 2022-04-12 |
Family
ID=81049422
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111460599.3A Pending CN114324580A (en) | 2021-12-03 | 2021-12-03 | Intelligent knocking detection method and system for structural defects |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114324580A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116203131A (en) * | 2023-04-28 | 2023-06-02 | 中国铁建高新装备股份有限公司 | Method and device for detecting tunnel void, electronic equipment and storage medium |
| CN116858943A (en) * | 2023-02-03 | 2023-10-10 | 台州五标机械股份有限公司 | Hollow shaft intelligent preparation method and system for new energy automobile |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106680371A (en) * | 2016-12-30 | 2017-05-17 | 陕西科技大学 | Electric porcelain body online flaw detection device and electric porcelain body online flaw detection method |
| CN110222591A (en) * | 2019-05-16 | 2019-09-10 | 天津大学 | A kind of method for detecting lane lines based on deep neural network |
| US20190317633A1 (en) * | 2018-04-13 | 2019-10-17 | Silicon Integrated Systems Corp | Method and system for identifying tap events on touch panel, and touch-controlled end project |
| CN110702792A (en) * | 2019-09-29 | 2020-01-17 | 中国航发北京航空材料研究院 | Alloy tissue ultrasonic detection classification method based on deep learning |
| CN111160438A (en) * | 2019-12-24 | 2020-05-15 | 浙江大学 | Acoustic garbage classification method adopting one-dimensional convolutional neural network |
| WO2020156348A1 (en) * | 2019-01-31 | 2020-08-06 | 青岛理工大学 | Structural damage identification method based on ensemble empirical mode decomposition and convolution neural network |
| CN111833856A (en) * | 2020-07-15 | 2020-10-27 | 厦门熙重电子科技有限公司 | Voice key information calibration method based on deep learning |
| CN111983020A (en) * | 2020-08-25 | 2020-11-24 | 绍兴市特种设备检测院 | Metal component internal defect knocking detection and identification system and identification method |
| CN112946072A (en) * | 2021-01-27 | 2021-06-11 | 重庆大学 | Abrasive belt wear state monitoring method based on machine learning |
-
2021
- 2021-12-03 CN CN202111460599.3A patent/CN114324580A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106680371A (en) * | 2016-12-30 | 2017-05-17 | 陕西科技大学 | Electric porcelain body online flaw detection device and electric porcelain body online flaw detection method |
| US20190317633A1 (en) * | 2018-04-13 | 2019-10-17 | Silicon Integrated Systems Corp | Method and system for identifying tap events on touch panel, and touch-controlled end project |
| WO2020156348A1 (en) * | 2019-01-31 | 2020-08-06 | 青岛理工大学 | Structural damage identification method based on ensemble empirical mode decomposition and convolution neural network |
| CN110222591A (en) * | 2019-05-16 | 2019-09-10 | 天津大学 | A kind of method for detecting lane lines based on deep neural network |
| CN110702792A (en) * | 2019-09-29 | 2020-01-17 | 中国航发北京航空材料研究院 | Alloy tissue ultrasonic detection classification method based on deep learning |
| CN111160438A (en) * | 2019-12-24 | 2020-05-15 | 浙江大学 | Acoustic garbage classification method adopting one-dimensional convolutional neural network |
| CN111833856A (en) * | 2020-07-15 | 2020-10-27 | 厦门熙重电子科技有限公司 | Voice key information calibration method based on deep learning |
| CN111983020A (en) * | 2020-08-25 | 2020-11-24 | 绍兴市特种设备检测院 | Metal component internal defect knocking detection and identification system and identification method |
| CN112946072A (en) * | 2021-01-27 | 2021-06-11 | 重庆大学 | Abrasive belt wear state monitoring method based on machine learning |
Non-Patent Citations (1)
| Title |
|---|
| 公安部第三研究所: "《多摄像机协同关注目标检测跟踪技术", 北京邮电大学出版社, pages: 103 - 104 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116858943A (en) * | 2023-02-03 | 2023-10-10 | 台州五标机械股份有限公司 | Hollow shaft intelligent preparation method and system for new energy automobile |
| CN116203131A (en) * | 2023-04-28 | 2023-06-02 | 中国铁建高新装备股份有限公司 | Method and device for detecting tunnel void, electronic equipment and storage medium |
| CN116203131B (en) * | 2023-04-28 | 2023-09-15 | 中国铁建高新装备股份有限公司 | Method and device for detecting tunnel void, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020125668A1 (en) | Method and system for automatically identifying surrounding rock level by applying while-drilling parameters | |
| CN111507990A (en) | Tunnel surface defect segmentation method based on deep learning | |
| CN112462346A (en) | Ground penetrating radar roadbed defect target detection method based on convolutional neural network | |
| CN110796186A (en) | Dry and wet garbage identification and classification method based on improved YOLOv3 network | |
| CN111444940A (en) | Fault diagnosis method for critical parts of fan | |
| CN112858473A (en) | Turnout switch blade damage state monitoring method based on feature fusion | |
| CN110647830B (en) | Bearing fault diagnosis method based on convolutional neural network and Gaussian mixture model | |
| CN114324580A (en) | Intelligent knocking detection method and system for structural defects | |
| CN111507998B (en) | Depth cascade-based multi-scale excitation mechanism tunnel surface defect segmentation method | |
| CN110110768A (en) | Fault Diagnosis of Roller Bearings based on Concurrent Feature study and multi-categorizer | |
| CN110716512A (en) | Environmental protection equipment performance prediction method based on coal-fired power plant operation data | |
| CN111855816A (en) | Fan blade fault identification method integrating probability model and cnn network | |
| CN109886433A (en) | The method of intelligent recognition city gas pipeline defect | |
| CN117435958A (en) | Wind power equipment fault prediction method and system based on multi-source data acquisition | |
| CN111783616B (en) | Nondestructive testing method based on data-driven self-learning | |
| CN117370766A (en) | Satellite mission planning scheme evaluation method based on deep learning | |
| CN116051496A (en) | Real-time sewer defect detection method | |
| CN109447026A (en) | A kind of pile body integrity classification automatic identifying method based on support vector machines | |
| KR102311558B1 (en) | System and method for detecting structure damage using artificial intelligence, and a recording medium having computer readable program for executing the method | |
| CN115015394A (en) | Composite material defect ultrasonic detection method based on convolution network and trajectory tracking | |
| CN111553115A (en) | Large-span bridge vibration response prediction method under typhoon action based on data driving | |
| CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network | |
| CN115455764A (en) | Multi-node damage diagnosis method for frame structure | |
| CN114660180A (en) | Sound emission and 1D CNNs-based light-weight health monitoring method and system for medium and small bridges | |
| Kannadaguli et al. | Microwave imaging based automatic crack detection system using machine learning for columns |
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 |