KR20220093942A - Apparatus and method for detecting fault of structure considering noise environment - Google Patents

Abstract

The present invention relates to an apparatus and a method for detecting the defect of a structure in consideration of a noise environment, wherein the severity of the defect of a structure made of various materials and having much environmental noise such as boilers, aircrafts, wind power generators, ships, concrete, bridges and rotating bodies is detected. A plurality of sensors are attached to detect the defect of a structure composed of various materials generating environmental noise, environmental noise is collected in advance using the same to generate a threshold value, and a plurality of sensors for sensing the occurrence of defects in accordance with the material characteristics of the structure are calibrated for optimization, the occurrence of defects is determined based on the threshold value, and the signals of the sensors, for which the occurrence of defects has been determined, are imaged to be analyzed using a neural network, thereby enabling the severity to be predicted accurately and enabling accurate defect determination and severity analysis while reducing false alarms by varying the threshold value depending on the severity information.

Description

Apparatus and method for detecting fault of structure considering noise environment}

The present invention relates to an apparatus and method for detecting a structure defect in consideration of a noisy environment, and more particularly, to a structure made of various materials such as a boiler, an aircraft, a wind generator, a ship, concrete, a bridge, a rotating body, etc. It relates to an apparatus and method for detecting a structure defect in consideration of a noise environment for detecting severity.

Acoustic emission technology is one of the non-destructive inspection methods, and real-time remote monitoring during operation is possible, and it is possible to estimate the location of defects, so its superiority is proven in the defect diagnosis technology of structures such as pressure vessels.

Most of these technologies are applied to industrial facilities such as high-pressure pipes and pressure vessels, and are established as professional safety diagnosis technologies, but are not widely applied to general safety diagnosis of large structures.

The conventional safety diagnosis system generally diagnoses the system by giving an alarm when it deviates from a reference signal based on changes in physical quantities such as sound, vibration, camera, temperature, and pressure.

However, in recent years, large structures have been developed in a way that concrete structures or metal structures are mixed with composite materials according to the diversification of material properties and strength in order to achieve light weight and safety. However, it is found that there is a weakness in safety diagnosis because it is out of the error range or a fatal error that fails to detect a defect occurs.

Furthermore, in the case of a system installed in a new site, false alarms frequently occur, making it difficult to apply, which means that it is not easy to apply the system to a structure generating a lot of environmental noise.

This is the result of not being able to diversify the correction techniques and analysis methods for physical quantities according to the characteristics of materials for existing structures. It is necessary to supplement in consideration of environmental noise rather than determining a fixed threshold value.

Korean Patent Registration No. 10-1949875 [Device and method for detecting defects in structures] Korean Patent Registration No. 10-1884096 [Device and Method for Detecting Internal Defects of Complex Structures Using Electromagnetic Acoustic Resonant Frequency]

In order to solve the above-mentioned problem, the present invention is to optimize the criteria for defect detection according to various materials constituting the structure according to values measured through a plurality of sensors attached to the structure for detecting structure defects in a noise generating environment. However, the reliability of the detection of structural defects in a noise-generating environment is increased by dynamically converting a threshold value for classifying the shock according to the noise environment and the shock according to the structural defect through the collection of prior measurement information on the noise environment. has its purpose.

Furthermore, the present invention converts the sensor measurement value for defect determination of a structure having a composite material into an image, visualizes it, and then learns it. Based on the pattern characteristics of the impact image visualized through the learned model, the severity determination is highly reliable. The purpose is to make it possible.

In addition, the present invention collects sensor measurement values according to environmental noise in a noise generating environment to variably set a threshold according to environmental noise, and after imaging the impact amount information according to environmental noise through a sensor that has been calibrated, it is The purpose of learning is to distinguish the visualization pattern according to environmental noise and the image pattern according to the impact amount conversion according to the defect, and to reduce the possibility of malfunction by setting the threshold value according to the environmental noise variably.

In the method for detecting a structure defect of an apparatus for detecting a structure defect in consideration of a noise environment according to an embodiment of the present invention, environmental noise measured in a normal state by a plurality of sensors attached to a structure is collected at a preset time or time point to determine a reference threshold value calculating; performing calibration for detecting a reference impact amount for each of a plurality of sensors attached to the structure; a determination step of receiving a defect response sound from the plurality of sensors, compensating for the reference threshold value, and determining whether the structure is defective; When a defect occurs as a result of the determination of the determination step, a learning step of mapping the received signals of the plurality of sensors according to the occurrence of a defect with elements constituting an image to form an image, and then matching the input severity information on the severity of the occurrence of the defect to learn the neural network Wow; If the learning step is completed, when the determination of the determination step results in a defect, the received signals for each sensor according to the occurrence of a defect are mapped with the elements constituting the image and imaged, and then the severity prediction information of the defect occurrence through the learned neural network. a step of predicting the severity of the calculation; and a threshold value correction step of varying the reference threshold value within a measurement range measured according to the environmental noise when the severity of the severity prediction step is less than or equal to a reference level.

As an example related to the present invention, the learning step further includes a learning step of mapping the received signal for each sensor according to the environmental noise collected after the performing step with the elements constituting the image to image it, and then learning the neural network as the environmental noise. .

As an example related to the present invention, the severity prediction step further calculates environmental noise similarity prediction information along with the severity prediction information of the occurrence of the defect through the learned neural network, and the threshold value correction step includes the severity prediction information being less than or equal to a standard The method further includes the step of increasing the threshold value when the environmental noise similarity prediction information is equal to or greater than the reference value, and maintaining or lowering the threshold value when the severity prediction information is greater than or equal to the reference value and the environmental noise similarity prediction information is less than or equal to the reference value.

As an example related to the present invention, the performing step includes a plurality of test positions that display the test positions of a uniform grid on an output device connected to the defect detection device and correspond one-to-one to the arrangement positions for each of the plurality of sensors, respectively. The measurement is performed for each sensor according to the generation of a reference impulse for the first test group that When the displacement between the first test position and the second test position on the shock delivery direction path to which the first test position of the first test group and the second test position belonging to the second test group belong is within a preset range The sensor of the first test position and the second test position receives the measurement value for each sensor corresponding to the reference shock amount for each of one or more other test positions located between the first test position and the second test position while belonging to the shock transmission direction path. The method may further include automatically interpolating the star measurement values according to a preset algorithm.

As an example related to the present invention, the learning step identifies the location of the defect on the grid based on the received signals received for each of the plurality of sensors, and includes a plurality of different characteristic variables set in advance according to the received signals for each sensor. The neural network is mapped with at least one of a color and a figure, which are elements constituting an image based on a parameter, to generate pattern information corresponding to the received signal for each sensor, and match the pattern information with the severity information according to a user input. It may be characterized in that it further comprises the step of learning.

An apparatus for detecting a structure defect in consideration of a noise environment according to another embodiment of the present invention collects environmental noise measured in a normal state by a plurality of sensors attached to a structure at a preset time or point in time to calculate a reference threshold value. collection department; a calibration unit for performing calibration for detecting a reference impact amount for each of the plurality of sensors attached to the structure; a determination unit for receiving a defect response sound from the plurality of sensors, compensating for the reference threshold, and determining whether the structure is defective; an imaging unit for mapping the received signals of the plurality of sensors according to the occurrence of a defect with elements constituting an image when a defect occurs as a result of the determination of the determination unit to form an image; By matching the image of the received signal provided from the imaging unit and the severity information on the severity of the occurrence of a defect corresponding to the received signal, the neural network is trained, and when the learning is completed, the imaged received signal for each sensor is analyzed with the learned neural network. It may include a learning unit for calculating the severity prediction information of the occurrence of the defect.

The present invention attaches a plurality of sensors to a structure made of various materials generating environmental noise to detect defects, collects environmental noise in advance through this, generates a threshold value, and senses the occurrence of defects according to the material characteristics of the structure After calibrating and optimizing a plurality of sensors, the occurrence of a defect is determined based on the threshold value, and the signal of the sensors determined to be defective is imaged and analyzed with a neural network to predict the correct severity, and also to predict the severity according to the severity information. By varying the value, it has the effect of reducing false alarms and enabling accurate defect determination and severity analysis.

Furthermore, the present invention divides the defect detection target surface of a large structure into grid units in a grid form and then uses the sensor-specific measurement values received from each of the optimized plurality of sensors to identify the location of the defect among the locations on the grid. By specifying the target surface for defect detection, it is possible to increase the reliability of defect detection as well as increase the reliability of the defect detection by supporting that the occurrence of a defect and the location of the defect are provided accurately and quickly only with the detection signal measured through the sensor based on acoustic emission sensing. This has the effect of easily preventing errors from occurring.

In addition, the present invention can generate the environmental noise identified based on the detection signal for each sensor in the form of an image, and then train the neural network to learn, and accordingly, it is possible to grasp how close the impact determined as the defect is to the environmental noise. By compensating the threshold value of the judgment unit, the threshold value is intelligently varied, so that it can be applied stably even to sites where it is difficult to measure non-destructive defects of the acoustic emission method due to severe environmental noise.

1 is a block diagram of a structure defect detection system including an apparatus for detecting a structure defect in consideration of a noise environment according to an embodiment of the present invention;
2 is a detailed configuration diagram of an apparatus for detecting a structure defect in consideration of a noise environment according to an embodiment of the present invention.
3 is an exemplary operation view of a structure defect detection apparatus based on sound emission according to an embodiment of the present invention.

Hereinafter, detailed embodiments of the present invention will be described with reference to the drawings.

1 is a block diagram of a structure defect detection system including a structure defect detection apparatus 100 (hereinafter, referred to as a defect detection apparatus 100) in consideration of a noise environment according to an embodiment of the present invention.

Prior to the description, the structure to which the present invention is applied is a composite structure such as an aircraft, a wind generator, a bridge, a rotating body, etc., and a structure in which a metal material and a composite material such as a boiler, a ship, and a concrete are mixed, and a structure generating significant environmental noise. can do. In the case of structures that normally generate such environmental noise, during acoustic-based non-destructive inspection, environmental noise is added to the detection sound for defect measurement, so false alarms are frequent. There is a problem of significant degradation.

For efficient description of the present invention, environmental noise applied to a large structure to which the present invention is applied is shown as a noise generating source (N) as shown, but this is only a conceptual expression, and environmental noise may occur in various locations, It may occur at a variable location, or it may occur at a variable size at a specific location.

Therefore, in order to solve this problem, the defect detection apparatus 100 according to the embodiment of the present invention provides environmental noise measured in a normal state for each of the plurality of sensors S1, S2, and S3 attached to the structure in which the noise source N exists. is collected at a preset time or time point to calculate a reference threshold value. Through this, it is possible to check the range of the sensor's measurement signal for environmental noise, and the reference threshold may be an average or maximum value of the environmental noise.

Thereafter, calibration is performed for detecting the reference impact amount for each sensor attached to the structure, and the defect response sound is received from the plurality of sensors S1, S2, and S3, and the reference threshold value is compensated for whether the structure is defective. When a defect occurs as a result of the determination, the received signals for each sensor according to the occurrence of a defect are mapped to elements constituting an image and imaged, and then the defect detection apparatus 100 corresponds to the input severity information on the severity of the occurrence of the defect. ) can be configured to train a neural network configured in Of course, when learning is completed, by providing the imaged received signal image for each sensor input to the neural network, it is possible to calculate the severity prediction information of the defect occurrence. If the severity is less than or equal to the reference, the reference threshold may be increased, and if the severity is greater than or equal to the reference, the reference threshold may be decreased. In another way, after calibration is completed, the received signal for each sensor according to the collected environmental noise is mapped with the elements constituting the image to be imaged, and then the neural network is trained as the environmental noise, and the received imaged sensor is determined to be defective. By providing the received signal image to the neural network in which the environmental noise has been learned, predicting how close the received signal for each sensor determined as the occurrence of a defect is to the environmental noise, and if it is close to the environmental noise above the standard, the reference threshold can be increased, When it is different from the abnormal environmental noise, the reference threshold value may be lowered.

At this time, in the present invention, acoustic emission applied during diagnosis for defect detection of the large structure includes high-frequency (eg, 20 kHz to 1 MHz) and low-frequency vibration (eg, 20 kHz to 1 MHz) including sound and vibration ( 2 kHz to 20 kHz), and the plurality of sensors may sense and detect the sound generated according to the sound emission diagnosis method.

2 is a detailed configuration diagram of the defect detection apparatus 100. As shown, environmental noise measured in a normal state by a plurality of sensors attached to a structure is collected at a preset time or point in time to calculate a reference threshold value. an environmental noise collecting unit 105 that performs a function, a calibration unit 110 that performs calibration for detecting a reference impact amount for each sensor attached to the structure, and a defect response sound from the plurality of sensors. A determination unit 120 for determining whether a large structure is defective, a calibration unit 110 for performing calibration for detecting a reference impact amount for each sensor attached to the structure, and a defect response sound to the plurality of sensors A determination unit 120 that determines whether the structure is defective after compensating for the reference threshold value, and when a defect occurs as a result of the determination of the determination unit 120, the received signals for each of the plurality of sensors according to the occurrence of a defect After mapping and imaging with the elements constituting the image, the neural network is trained by matching the input severity information (provided by the manager for learning) on the severity of the occurrence of a defect. and a learning unit 140 for calculating the severity prediction information of the occurrence of a defect by analyzing it with a neural network.

At this time, any one of the environmental noise collecting unit 105 , the calibrating unit 110 , the determining unit 120 , the imaging unit 130 , and the learning unit 140 controls the overall control function of the defect detection apparatus 100 . It may be configured as a control unit that performs the operation, and the control unit may include RAM, ROM, CPU, GPU, and a bus, and the RAM, ROM, CPU, GPU, etc. may be connected to each other through a bus.

In addition, at least one of the environmental noise collecting unit 105 , the calibrating unit 110 , the determining unit 120 , the imaging unit 130 , and the learning unit 140 may be configured to be included in the other. to be.

In addition, the defect detection apparatus 100 has various components such as a communication unit for communication with an external device, a user input unit for receiving user input, a storage unit for storing various information, and a display unit for displaying various information in addition to the above-described component unit An additional may be included, and the communication unit may communicate with an external device such as a user terminal through a communication network using various well-known wired and wireless communication methods.

Hereinafter, an example of a detailed operation configuration of the defect detection apparatus 100 based on the above-described configuration will be described.

First, the environmental noise collecting unit 105 is connected to communicate with each of the plurality of sensors attached to the structure, and collects the environmental noise signal in a normal state without applying a separate impact from each of the plurality of sensors. This calculates the range of noise collected as a signal for each sensor by environmental noise by collecting environmental noise for a preset time (eg, 30 minutes, 1 hour, 1 day, or several days). The environmental noise collecting unit 105 may select an average value, a specific value in consideration of variance, a maximum value, etc. as a reference threshold value for the environmental noise in the noise range. This may vary depending on the characteristics of the structure or the characteristics of the environment. The reference threshold value obtained in this way is transmitted to the determination unit 120 , and, if necessary, information on a measurement range of environmental noise may also be transmitted.

The calibration unit 110 is connected to communicate with the plurality of sensors attached to the structure to perform calibration for each of the plurality of sensors according to the characteristics of various materials constituting the large structure from each of the plurality of sensors. can

For example, the calibration unit 110 may include a collection unit 111 communicatively connected to the plurality of sensors, and the environmental noise collection unit 105 may include a collection unit 111 of the calibration unit 110 . ) can be shared.

In addition, in order to calibrate the plurality of sensors according to the characteristics of the composite material constituting the structure, the sound sensed corresponding to the reference impact amount from each of the plurality of sensors is corresponding to the reference impact amount forcibly generated through sound emission on the structure A test signal that is a related sensing signal may be received.

Also, the calibrator 110 may set and store the test signal initially received from each of the plurality of sensors as an initial value for each sensor.

Also, the calibrator 110 may periodically receive the test signal through the collection unit 111 in order to compensate for a decrease in sensitivity of each of the plurality of sensors as the usage period increases.

In addition, the calibrator 110 periodically compares the initial value for each of the plurality of sensors each time the test signal is received, and calculates and calculates an offset of the current test signal compared to the initial value for each of the plurality of sensors. can

Also, the calibrator 110 may correct a signal received from each of the plurality of sensors based on the offset (or offset information) calculated for each of the plurality of sensors.

In addition, the calibrator 110 sets reference values for the highest and lowest values reflecting the offset on the basis of the initial value whenever an offset is calculated for a specific sensor for each of the plurality of sensors, and the specific signal received from the specific sensor is When out of the reference range according to the reference value, the specific signal is ignored by the defect detection apparatus 100 so that the specific signal is not used in other components or by generating an alarm for abnormal occurrence of the specific sensor It is possible to notify the sensor in which an abnormality has occurred among the sensors.

In this case, the defect detection apparatus 100 may be connected to a user terminal by wire or wirelessly to communicate with the user terminal, and the defect detection apparatus 100 detects that the received signal of the specific sensor is outside the reference range set in response to the specific sensor. In this case, alarm information may be generated and transmitted to the user terminal, and through this, the user of the user terminal may be provided so that the user of the user terminal can check the sensor in which the abnormality has occurred through the alarm information received and output to the user terminal.

Here, the user terminal may include various terminals such as a PC, a smart phone, and a notebook computer.

In the case of a general safety diagnosis system, the system reliability is low due to frequent alarm occurrence, but the present invention can reduce the occurrence of false alarms due to environmental noise and temporary signal rise through the above-mentioned bar, so that the defects of the structure are reduced. The reliability of the diagnosis can be increased, and there is an advantage of providing the ability to immediately identify defects caused by damage to sensors installed nearby.

Meanwhile, when the offset setting by the calibration unit 110 is completed as described above, the defect detection apparatus 100 may perform defect diagnosis of a large structure made of various materials such as metal, composite, concrete, and the like.

In detail, when the calibration according to the offset setting for each sensor of the calibration is completed, the calibration unit 110 corrects the signals received from the plurality of sensors through the collection unit 111 based on the offset. It may be provided to the determination unit 120 .

In this case, the calibration unit 110 including the collection unit 111 may be included in the determination unit 120 .

In addition, the determination unit 120 transmits signals generated by the plurality of sensors to the defect occurrence response sound, which is a sound generated when a defect such as a crack occurs in the large structure, and is generated when the defect occurs through the calibration unit 110 . can receive

In addition, the determination unit 120 may determine whether there is a defect based on the received signals received from the plurality of sensors and the reference threshold calculated by the environmental noise collecting unit 105 , and the result information on the determination result is defective. It may be output through an external output device connected to the detection device 100 or transmitted to the user terminal. Determination of the presence of such a defect may be determined based on how much the received signal exceeds the reference threshold. For example, if it is 6 dB or more than the reference threshold, it may be determined as the occurrence of a defect. Of course, this threshold may be varied, and a defect determination criterion other than 6 dB may be set.

At this time, in the present invention, the defect occurrence position of the structure determined by the determination unit 120 to be determined to have occurred may be detected and the detection information may be provided thereto. The operation is performed, which will be described in detail with reference to FIG. 3 .

The defect detection apparatus 100 may automatically calculate the attenuation characteristics of the material of the structure and reduce the position error in parallel with the automatic correction function.

To this end, when the defect detection apparatus 100 knows the distance from adjacent sensors such as sensor 1, sensor 2, and sensor 3, the time arrival difference from each sensor is calculated using the basic formula that the product of time and speed is at a constant distance. Safety diagnosis can be performed by calculating and using this to automatically calculate the attenuation characteristics according to the characteristics of the material and correct the attenuation function.

In this case, when the different attenuation characteristics are not known for each physical property of the material, a large positional error appears, but the present invention can perform a function of automatically correcting such a function.

At this time, in FIG. 3, three sensors S1, S2, and S3 are configured on the assumption of a two-dimensional plane, but according to the arrangement shape of the structure in which the sensors are disposed, additional sensors may be added in addition to the three sensors. The error can be further reduced.

Meanwhile, in order to increase the accuracy of the detection information as described above, the defect detection apparatus 100 may include an imaging unit 130 and a learning unit 140 .

In a state in which a learning mode for increasing the accuracy of detection information is set, the imaging unit 130 and the learning unit 140 may perform learning.

That is, the imaging unit 130 may receive received signals for each of the plurality of sensors determined by the determination unit 120 to have a defect when setting the learning mode.

Also, the imaging unit 130 may perform a calibration on a defect occurrence location in conjunction with the calibrator 110 .

To this end, the imaging unit 130 may display preset test positions of a uniform grid on an output device connected to the sensing unit.

In this case, the imaging unit 130 displays the test positions of the uniform grid on the output device connected to the defect detection apparatus 100, and includes a plurality of test positions corresponding to the arrangement positions of a plurality of sensors one-to-one, respectively. After performing the measurement for each sensor according to the generation of the reference impulse for the primary test group, and performing the measurement for each sensor for the secondary test group, which are test locations spaced apart from the primary test group by a predetermined plurality of grid intervals, When the displacement between the first test position and the second test position on the shock delivery direction path to which the first test position of the first test group and the second test position of the second test group belong is within a preset range, the shock Measurement of each sensor at the first test position and the second test position corresponding to the reference impact amount for one or more other test positions located between the first test position and the second test position while belonging to the transfer direction path Values can be automatically interpolated according to a preset algorithm.

As an example, the imaging unit 130 interlocks with the calibration unit 110 to measure values according to the occurrence of a reference shock amount for a first test group including a plurality of placement positions for each sensor and a plurality of test positions corresponding one-to-one, respectively. can be collected.

In this case, a signal for the reference shock amount may be generated using the reference shock generating device, and a method of providing a signal by embedding the reference signal generating device in the sensor itself and detecting it by a nearby sensor may also be used. For example, by generating a signal from the first sensor S1 and receiving the signals from the adjacent second and third sensors S2 and S3, it is possible to automatically calculate the characteristic value according to the material, such as the attenuation ratio with respect to the distance. can

In addition, the imaging unit 130 may collect measurement values based on received signals received by a plurality of sensors provided from the calibration unit 110 , and these measurement values correspond to the reference shock amount and are measured (sensed) by the sensor. a) can mean a value.

In addition, the imaging unit 130 may interwork with the calibration unit 110 to measure the reference impact amount with respect to the first test group and the second test group, which are test locations spaced apart by a predetermined plurality of grid intervals.

In addition, the imaging unit 130 considers the shock transmission direction path, and the displacement between the first test position of the first test group located on the shock transmission direction path and the second test position belonging to the second test group is a preset range. In the case of within the range, the measured values for each of the one or more other test locations located between the first test location and the second test location while belonging to the shock delivery direction path are based on the measured values measured at the first test location and the second test location. can be automatically interpolated.

That is, the imaging unit 130 corresponds to the reference impact amount according to the reference sound generated for the large structure, and confirms the measured values of the first test group and the second test group, and the first of the impact delivery direction paths according to the reference impact amount It is possible to identify a specific path, which is a path in the shock transmission direction, in which the first test position belonging to the test group and the second test position belonging to the second test group belong to the path, and the attenuation characteristic calculated using the basic formula and the second test position Using the measurement values for each sensor of each of the first test position and the second test position, the measurement values for each sensor corresponding to the reference impact amount are predicted and calculated for each test position for one or more grids belonging to the specific path (or located on the specific path). can do.

Through this, the imaging unit 130 may omit direct testing for test locations that do not belong to the first test group and the second test group, and may interpolate the measured values.

Meanwhile, the imaging unit 130 may automatically set the measured value for each sensor calculated or confirmed for the reference impact amount for each test location as a reference contrast value to complete the calibration for determining the location of the defect.

As described above, the imaging unit 130 generates a target surface, which is a target for detecting defects in a structure, in a grid form, and corresponds to a reference impact amount for a plurality of test positions on a grid constituting the grid, so that the calibration unit ( 110) can be set for each test location, and through this, when a defect occurs in a large structure, the measured value of each sensor is compared with the reference contrast value so that the location of the defect corresponding to the measured value for each sensor is accurately identified. can do.

As an example, the imaging unit 130 detects (actually measured) by the plurality of sensors in response to the reference contrast value and the defect occurrence response sound of the large structure from the determination unit 120 when the calibration for the defect occurrence location is completed. ) can receive a detection signal.

In addition, the imaging unit 130 calculates the actual measured value for each sensor based on the detection signal, compares it with the reference contrast value for each test location, and compares it with a specific reference contrast value corresponding to the actual measured value for each sensor. One or more test locations where the difference from the test location or the measured value for each sensor is within a preset range is identified as a defect occurrence area, and detection information for the defect occurrence area is generated and provided to the user terminal or to the output device. can be printed out.

In this case, the imaging unit 130 may display an area or location corresponding to the detection information on a grid including a plurality of grids displayed through the output device or the user terminal.

Also, the imaging unit 130 may provide the detection information together with the result information to an output device or a user terminal, or may provide the detection information by including the detection information in the result information.

Also, the imaging unit 130 may provide the result information as alarm information in conjunction with the determination unit 120 .

On the other hand, the imaging unit 130 may provide an image of the severity of the occurrence of a defect when a defect occurs, and through linking with the learning unit 140, the predicted severity of the region where the defect occurred may be provided based on a neural network, This will be described in detail.

That is, the imaging unit 130 converts the sensing signal into an abstract image for severity measurement as follows, in addition to the function of displaying the calibration position or the defect occurrence position as an image of a coordinate shape as described above, and converts various components included in the signal. It further performs the function of converting an image pattern having a flatness and a sense of depth (brightness, color, etc.). After converting the signal into an abstract image in this way, it is possible to classify the severity based on the characteristics of the image pattern of the measurement signal through an image-based neural network.

As described above, the imaging unit 130 extracts features of data at the point in time when a defect is detected in the existing area A, and classifies the corresponding features, as a method of providing big data-based automatic defect detection and severity. For example, various characteristic variables and environmental variables such as frequency, amplitude, and energy at the time are stored and automatically stored in the fault diagnosis DB. It can detect a defect by using the method and operate in a way that notifies the severity of the defect.

To this end, the imaging unit 130 identifies a defect occurrence location, which is a test location on the grid of the grid, based on the received signals (sensing signals) received for each of the plurality of sensors, and sets a plurality of preset locations according to the received signals for each sensor. The pattern information corresponding to the received signal for each sensor is generated by mapping with at least one of color and figure, which are elements constituting an image, based on the parameters for each different characteristic variable, and the pattern information and the pattern information are interlocked with the learning unit 140 . The neural network may learn by matching the severity information according to a user input.

In this case, the imaging unit 130 may generate the pattern information in the form of the grid-based image. In addition, the imaging unit 130 may generate the pattern information based on the environment information at the time when the received signal for each sensor is received and the received signal for each sensor. In this case, the characteristic variable may include frequency, amplitude, energy, and the like.

As an example, if there are three sensors (two-dimensional plane impact detection), the imaging unit 130 may identify the location of an impact or defect, and image the measurement value of each sensor by matching the elements constituting the image. have.

In addition, the imaging unit 130 matches each sensor to the R, G, and B color elements, matches the energy to the size of each color, and expresses the frequency and amplitude as a pattern such as a line or circle based on the point of impact, Environmental information (material characteristics, attenuation characteristics, etc.) may be expressed in a predetermined patterned background.

In addition, the imaging unit 130 provides the pattern information to the learning unit 140 , and the learning unit 140 corresponds to the pattern information and provides the user with an actual measured value (sensed value) for each sensor corresponding to the pattern information. .

In this case, the pattern information may be image information in which the defect occurrence pattern is imaged in graphic form such as color and figure on a grid corresponding to the measured values for each sensor, and the learning unit 140 is configured to combine the image information with the image information. By matching the severity information with each other, the neural network may be trained.

In addition, such a neural network may be configured as a neural network model, and may include an input layer, one or more hidden layers, and an output layer. In addition, various types of neural networks such as Deep Neural Network (DNN), Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), and Support Vector Machine (SVM) may be applied to the neural network model.

In addition, in the above configuration, the pattern information may include various physical quantity signals (or physical quantity data) used to generate the pattern information.

In addition, the defect detection apparatus 100 may communicate with a central server through a communication network, and the central server detects one or more other defects that perform defect detection and diagnosis on large structures other than the defect detection apparatus 100 . Big data that is learning data including the pattern information and the severity information may be collected by communicating with the device and the defect detection apparatus 100 , and the big data may be transmitted to the defect detection apparatus.

Accordingly, the learning unit 140 of the defect detection device 100 can apply the big data received from the central server to the neural network, and through this, the learning unit 140 can apply the big data received from the central server to the neural network based on the big data collected for various structures. The accuracy of the calculated severity-related data can be increased.

On the other hand, when the learning of the neural network of the learning unit 140 is completed as described above, the imaging unit 130 measures the actual value for each sensor according to the detection signal for each sensor determined that the defect has occurred from the determination unit 120 . , and may generate detection information and pattern information as described above in response to the measured values for each sensor.

In addition, the imaging unit 130 provides the pattern information to the learning unit 140 , and the learning unit 140 applies the pattern information received from the imaging unit 130 to the completed neural network to learn the pattern. In correspondence with the information, it is possible to calculate severity prediction information about the expected severity of the location where the binding occurs.

In addition, the imaging unit 130 works in conjunction with the learning unit 140 to transmit at least one of the severity prediction information, detection information, and result information calculated in response to the defect determination of the determination unit 120 to the output device. may be output or transmitted to the user terminal.

By utilizing the characteristics of the imaging unit 130 and the learning unit 140 as they are to analyze environmental noise, the sensor is determined to be defective according to the determination unit 120 and provided and converted into an image by the imaging unit 130 . It is possible to predict how similar their received signal to the environmental noise is through the neural network trained by converting the environmental noise into an image.

That is, the environmental noise collecting unit 105 provides the collected environmental noise information (measured values measured by each sensor with respect to the environmental noise) to the imaging unit 130 , and the imaging unit 130 uses the environmental noise information according to the corresponding environmental noise information. After mapping the received signals of each sensor with elements constituting an image to form an image, the image is provided to the neural network of the learning unit 140 , and the corresponding image may be learned as environmental noise.

In this case, the learning unit 140 learns the sensor signal image for the occurrence of a defect, learns the sensor signal image for the environmental noise, but can learn by tagging identification information such as severity and environmental noise, respectively, and learning is performed accordingly The completed neural network can calculate severity prediction information and environmental noise prediction information for the received sensor signal image.

Accordingly, the learning unit 140 provides the imager 130 with a sensor signal determined to be a defect by exceeding the current threshold through the determination unit 120 , and the imager 130 converts the sensor signal into an image. converted into , and provided to the learning unit 140, the learning unit 140 further calculates environmental noise similarity prediction information along with the severity prediction information of defect occurrence through the learned neural network, and provides the severity prediction information to the user. have. Furthermore, when the severity prediction information is below the standard and the environmental noise similarity prediction information is equal to or greater than the standard, the threshold value of the determination unit 120 is increased, and when the severity prediction information is equal to or greater than the reference level and the environmental noise similarity prediction information is below the standard, the determination unit ( 120) can be maintained or adjusted downward. Of course, such a threshold correction method may vary.

Through this, the present invention attaches a plurality of sensors to a structure composed of various materials generating environmental noise to detect defects, collects environmental noise in advance through this, generates a threshold value, and creates a defect according to the material characteristics of the structure. After calibrating and optimizing a plurality of sensors that sense the occurrence, the occurrence of a defect is determined based on a threshold value, and the signal of the sensors determined as the occurrence of a defect is imaged and analyzed with a neural network to accurately predict the severity and severity information By changing the threshold value according to this, it is possible to accurately determine the defect and analyze the severity while reducing false alarms.

The various devices and components described herein may be implemented by hardware circuitry (eg, CMOS-based logic circuitry), firmware, software, or a combination thereof. For example, it may be implemented using transistors, logic gates, and electronic circuits in the form of various electrical structures.

Those of ordinary skill in the art to which the present invention pertains may modify and modify the above-described contents without departing from the essential characteristics of the present invention. Accordingly, the embodiments disclosed in the present invention are not intended to limit the technical spirit of the present invention, but to explain, and the scope of the technical spirit of the present invention is not limited by these embodiments. The protection scope of the present invention should be construed by the following claims, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of the present invention.

100: structure defect detection device 105: environmental noise collecting unit
110: calibration unit 111: collection unit
120: judgment unit 130: image conversion unit
140: study unit

Claims (6)

As a structure defect detection method of a structure defect detection device considering a noise environment,
calculating a reference threshold value by collecting environmental noise measured in a normal state by a plurality of sensors attached to the structure at a preset time or time point;
performing calibration for detecting a reference impact amount for each of a plurality of sensors attached to the structure;
a determination step of receiving a defect occurrence response sound from the plurality of sensors, compensating for the reference threshold value, and determining whether the structure is defective;
When a defect occurs as a result of the determination of the determination step, a learning step of mapping the received signals of the plurality of sensors according to the occurrence of a defect with elements constituting an image to form an image, and then matching the input severity information on the severity of the occurrence of the defect to learn the neural network Wow;
If the learning step is completed, when the determination of the determination step results in a defect, the received signals for each sensor according to the occurrence of a defect are mapped with the elements constituting the image and imaged, and then the severity prediction information of the defect occurrence through the learned neural network. a step of predicting the severity of the calculation;
and a threshold value correction step of varying the reference threshold value within a measurement range measured according to the environmental noise when the severity of the severity prediction step is less than or equal to the standard.
The method according to claim 1,
The learning step further comprises a learning step of mapping the received signal for each sensor according to the environmental noise collected after the performing step with the elements constituting the image to image it and then learning it as the environmental noise in the neural network considering the noise environment A method for detecting a structure defect in a structure defect detection device.
3. The method according to claim 2,
The severity prediction step further calculates environmental noise similarity prediction information together with the severity prediction information of defect occurrence through the learned neural network, and the threshold correction step includes the severity prediction information being less than or equal to the standard and the environmental noise similarity prediction information being equal to or greater than the standard. In case the threshold value is raised, and when the severity prediction information is greater than or equal to the reference level and the environmental noise similarity prediction information is less than or equal to the reference value, the method further comprises the step of maintaining or lowering the threshold value. of structural defects detection method.
The method according to claim 1,
In the performing step, the test positions of the uniform grid are displayed on the output device connected to the defect detection device, and the first test group including the plurality of test positions corresponding to the arrangement positions of the plurality of sensors one-to-one, respectively. After performing the measurement for each sensor according to the generation of the reference impulse, and performing the measurement for each sensor on the second test group, which is the test locations spaced apart from the first test group by a plurality of grid intervals, the first test group When the displacement between the first test position and the second test position on the shock delivery direction path to which the first test position and the second test position belonging to the second test group belong is within a preset range, while belonging to the shock delivery direction path The measured value for each sensor corresponding to the reference impact amount for one or more other test positions located between the first test position and the second test position is set to a preset algorithm for the measured value for each sensor of the first test position and the second test position. The method of detecting a structure defect in a structure defect detection apparatus considering a noise environment, characterized in that it further comprises the step of automatically interpolating according to the method.
The method according to claim 1,
The learning step identifies the location of the defect on the grid based on the received signals received by the plurality of sensors, and constructs an image based on a plurality of parameters for each different characteristic variable preset according to the received signals for each sensor Generating pattern information corresponding to the received signal for each sensor by mapping with at least one of elements color and figure, and matching the pattern information with the severity information according to a user input to train the neural network to learn A method of detecting a structure defect in a structure defect detecting apparatus in consideration of a noise environment, characterized in that.
an environmental noise collecting unit that collects environmental noise measured in a normal state by a plurality of sensors attached to a structure at a preset time or point in time and calculates a reference threshold value;
a calibration unit for performing calibration for detecting a reference impact amount for each of the plurality of sensors attached to the structure;
a determination unit for receiving a defect response sound from the plurality of sensors, compensating for the reference threshold value, and determining whether the structure is defective;
an imaging unit for mapping the received signals of the plurality of sensors according to the occurrence of a defect with elements constituting an image when a defect occurs as a result of the determination of the determination unit to form an image;
By matching the image of the received signal provided from the imaging unit and the severity information on the severity of the occurrence of a defect corresponding to the received signal, the neural network is trained, and when the learning is completed, the imaged received signal for each sensor is analyzed with the learned neural network. A structure defect detection apparatus in consideration of a noise environment including a learning unit for calculating the severity prediction information of the occurrence of a defect.

Images