KR20210085168A - System and method for safety inspection by trainiing nature freqeuncy of structure based on machine learning - Google Patents

Abstract

The present invention relates to a safety inspection system by learning the natural frequency of a building structure based on machine learning. Provided is a safety inspection system by learning the natural frequency of a building structure based on machine learning, comprising: a cloud platform for collecting sensor data, which represents the safety of a building structure, from a sensor device, and managing the collected sensor data; a building structure safety analysis part for analyzing the safety of a building structure based on the sensor data, which represents the safety of a building structure, collected from the sensor device; and a monitoring and notification part for receiving and displaying an alarm signal based on the analysis result of the sensor data carried out by the building structure safety analysis part. The building structure safety analysis part includes: an absolute management part for analyzing the safety of a building structure in accordance with alarm setting standards based on preset specific sensor values; a continuous management part for analyzing the safety of a building structure in accordance with alarm setting standards based on the time series data of preset specific sensor values; and a composite management part for analyzing the safety of a building structure by a sorter, which learns using deep learning, based on sensor data from the sensor device.

Description

Safety diagnosis system and method through machine learning-based natural vibration value learning of building structures {SYSTEM AND METHOD FOR SAFETY INSPECTION BY TRAINIING NATURE FREQEUNCY OF STRUCTURE BASED ON MACHINE LEARNING}

The present invention relates to a method and apparatus for diagnosing the safety of a building structure, and more particularly, to a method and apparatus capable of efficiently and accurately performing a safety diagnosis of a structure by learning a natural vibration value of a building structure based on machine learning will be.

Although building safety accidents occur frequently in recent years, preventive measures against them are still insufficient. Although the proportion of social infrastructure among the first and second types of facilities is continuously decreasing, the proportion of buildings is continuously increasing, highlighting the importance of safety diagnosis and maintenance of buildings. As for the condition grades for each type of building (grades 5 - A, B, C, D, E), among the first and second types of facilities, the status grades of buildings are 18.0% of grade A and 79.8% of grade B, which is overall B grade or higher. In addition to the 1st and 2nd types, discussions are being conducted on the designation of 3 types of facilities, and in order to prevent building safety accidents in advance, it is necessary to improve the problem of limited participation of experts and develop and apply a maintenance system incorporating real-time monitoring. can be said to be the key.

However, the existing building safety diagnosis system has limitations in that it is based on manpower. Facility inspection and diagnosis work currently consumes a lot of time and effort in the work process, such as on-site investigation, inputting the investigation contents into a PC at the office, organizing data and writing reports, and although the technical level required for building safety and maintenance work is high, technicians The level of professionalism is insufficient, and the technical skills of professional technicians are declining due to frequent turnover of technicians due to poor work environment. In addition, a number of diagnostic equipment to support the work of technicians have been developed and applied, but there is a shortage of experts who can utilize them, and they are not effectively utilized. Therefore, it can be said that it is necessary to strengthen the competence of experts and develop core technologies for real-time data-based building safety diagnosis and evaluation in order to overcome the limitations of the manpower base of the existing building safety diagnosis system.

In addition, in Korea, many buildings were built in a short period of time due to the rapid economic growth and implementation of the housing supply expansion policy since the 1970s, but many of the buildings deteriorated severely because maintenance was not carried out systematically. If repairs to existing D and E grade facilities are not urgently performed, the number of disaster risk facilities is expected to increase rapidly every year. According to the statistical data of the Ministry of Land, Infrastructure and Transport (2014), 35.8% of buildings over 30 years after completion were surveyed nationwide, 24.5% in the metropolitan area, and 40.2% in regional areas. The status of buildings over 30 years by use was 38.8% for residential and 38.8% for commercial use. 38.5% for industrial use, 15.6% for industrial use, and 10.1% for educational and social use, indicating that the deterioration of buildings is severe and the deterioration is mainly concentrated in residential and commercial buildings. If the paradigm of facility maintenance is changed to preventive maintenance, the effect of huge economic cost reduction can be expected. According to the study, the US will cost 3.6 trillion dollars by 2020 ASCE (Mar. 2013). As a result of analysis of the impact of US infrastructure aging (3,840 trillion won), Japan will need 93.7 trillion won in annual maintenance costs from the Ministry of Land, Infrastructure and Transport (July 2012) 20 years later is expected to be

In general, most of the safety diagnosis and maintenance tasks performed by companies are performed visually or using non-destructive testing equipment by professional personnel, but there may be differences in the reliability of the diagnosis results depending on the skills and experience of the professional personnel. , there are problems in that it is difficult to diagnose a location where human access is difficult, and it takes a lot of time and money. In addition, since the manpower in charge of local governments is limited, it is virtually impossible to manage all hazardous facilities in real time. In order to improve the maintenance problems of manpower-oriented facilities, IT technology is used to increase maintenance efficiency and to reduce maintenance costs, a preventive safety diagnosis and maintenance system using ICT and IoT-based advanced convergence technology is implemented. There is a need to introduce

On the other hand, since the inner and outer walls of the building are finished, it is difficult to judge structural deterioration due to aging without safety diagnosis through professional personnel. There is a risk of collapse due to structural deterioration. Therefore, through structural deterioration monitoring, it is necessary to derive the time to repair and reinforce the building at the lowest cost and with the highest efficiency to promote optimal maintenance and management, and timely repair/reinforcement and preventive maintenance are essential for the longevity of the building. Because it can ultimately achieve enormous economic cost savings by promoting a masterpiece, safety diagnosis and automation of maintenance of buildings incorporating IT technology are desired.

An object of the present invention is to provide a system and method capable of accurately and efficiently analyzing the safety level of a building structure by sensor data of a building structure based on machine learning.

In particular, the present invention installs a sensor device capable of low-power long-distance communication in a building structure, and performs a safety analysis based on deep learning based on the sensor data collected therefrom in real time, thereby accurately detecting the possibility of risk occurrence in the building structure. Another object is to provide a system and method that can predict and prevent safety accidents caused by building structures in advance.

In order to solve the above problems, the present invention is a safety diagnosis system through machine learning-based building structure intrinsic vibration value learning, and collects sensor data indicating the safety level of a building structure from a sensor device, and manages the collected sensor data cloud platform; a building structure safety analysis unit that analyzes the safety level of the building structure based on the sensor data indicating the safety level of the building structure collected from the sensor device; and a monitoring/notification unit configured to receive and display an alarm signal according to a result of analyzing the sensor data by the building structure safety level analysis unit, wherein the building structure safety level analysis unit includes an alarm setting standard according to a preset specific sensor value The absolute management department that analyzes the safety level of the building structure by; a continuous management unit that analyzes the safety level of a building structure according to an alarm setting criterion according to time series data of a preset specific sensor value; and a complex management unit that analyzes the safety level of a building structure by a classifier learned based on deep learning based on sensor data from a sensor device. A safety diagnosis system through machine learning-based building structure intrinsic vibration value learning, characterized in that it comprises: provides

Here, the complex management unit, a first model for predicting the probability of reaching the dangerous state in the next cycle, a second model for predicting the time until reaching the dangerous state, and a third model for classifying the dangerous state The safety level may be analyzed by at least one.

In addition, the first model refines the data by collecting or processing sensor data for each measurement period, and calculating the probability of becoming a dangerous state such as caution, danger, and alarm in the next period for a certain period, and thus the state of the highest probability The second model predicts the mutation to analyze the degree of safety, and the second model trains the classifier based on time series data for the measurement period and cluster classification test data, and inputs a sensor value at a specific point in time to obtain a probabilistic transition to the cluster state. The third model analyzes the safety level by predicting the period of high probability of transition to the dangerous state by calculating the inference, and the third model learns the classification model by clustering the characteristics of the sensor value and classifies it into a preset number to analyze the safety level. can

According to the present invention, it is possible to provide a system and method capable of accurately and efficiently analyzing the safety level of a building structure based on machine learning based on sensor data of the building structure.

In particular, the present invention installs a sensor device capable of low-power long-distance communication in a building structure, and performs a safety analysis based on deep learning based on the sensor data collected therefrom in real time, thereby accurately detecting the possibility of risk occurrence in the building structure. It is possible to provide a system and method that can predict and prevent safety accidents caused by building structures in advance.

1 is a view showing the overall configuration and connection relationship of a safety diagnosis system 100 through machine learning-based building structure intrinsic vibration value learning according to the present invention.
2 shows sensor devices 300 used in the present invention.
3 is a diagram illustrating a process of selecting an installation location and measuring/reviewing the sensor devices 300 .
4 is a diagram for explaining an example of an operation of the analysis unit 30 .
5 illustrates a scenario analysis process for the main structural members of the analysis unit 30 .
6 is a diagram illustrating the configuration of the analysis unit 30 .
7 illustrates an alarm setting standard according to a specific sensor value used by the absolute management unit 31 .
8 illustrates an alarm setting standard according to time series data of a specific sensor value used by the continuity management unit 32 .
9 is a diagram for explaining the operation of the complex management unit 33 .
10 is a diagram for explaining the overall configuration of the complex management unit 33 .
11 and 12 show an example in which an alarm signal by the monitoring/notifying unit 40 is displayed.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a view showing the overall configuration and connection relationship of a safety diagnosis system 100 through machine learning-based building structure intrinsic vibration value learning according to the present invention.

Referring to FIG. 1 , a safety diagnosis system 100 (hereinafter simply referred to as “system 100”) through machine learning-based building structure natural vibration value learning according to the present invention includes a cloud platform 10, a sensor linkage unit ( 20), the building structure safety analysis unit 30 and the monitoring/notifying unit 40, and receive the safety data of the building structure from the sensor device 300 through the network server 200 to determine the safety level of the building structure perform the analysis function.

The system 100 is connected to the network server 200 through a network, and the network server 200 is connected to the sensor devices 300 . Although not explicitly shown, the system 100 and the network server 200 are coupled through a conventional network such as the Internet and a mobile communication network, and the sensor device 300 and the network server 200 are, for example, low-power, long-distance, such as LoRaWan. It can be connected through a communication network.

The sensor device 300 is provided in the building structure, collects safety data required for safety diagnosis of the building structure, and transmits it to the system 100 through the network server 200 and the network.

The sensor device 300 is various measurement sensors including an accelerometer, an inclinometer, a thermometer, a strain gauge, and the like, and collects safety data of a building structure. It is desirable that these are sensor devices based on LoRaWan so that they can be operated even in the city center where buildings are dense and can be operated even in old facilities where continuous power supply is difficult.

As the sensor device 300 , the accelerometer is used to estimate the rigidity of the macroscopic structural system level, and the microscopic structural member level damage/failure is tracked using an inclinometer, a thermometer, and a strain gauge.

In addition, it goes without saying that other sensor devices that can identify the static/dynamic sensor performance and check the normal and extreme building behavior measurement range can be used.

2 shows sensor devices 300 used in the present invention.

Referring to Figure 2, a GPS receiver and wind direction anemometer are installed on the roof, and it can be seen that a strain gauge can be installed on the 2nd Belt Truss, 1st Belt Truss, 2nd Outrigger, 1st Outrigger, and Truss Supporting Sloping Hanger. In addition, it can be seen that a uniaxial accelerometer was used for the cantilever girder, and a column axial force and foundation response accelerometer can be used for the foundation.

The sensor device 300 should be installed by appropriately selecting the installation location for each building type.

3 is a diagram illustrating a process of selecting an installation location and measuring/reviewing the sensor devices 300 .

Referring to FIG. 3 , for example, the sensor device 300 may be installed on a column, shear wall, etc. of the floor surface of a main floor of a high-rise building, and a sensor device 300 on a wall, column, beam, etc. by selecting an important structural member for a low-rise building can be installed This is done by selecting major structural members to be installed by building type, determining important structural members capable of collapsing by building type, and installing them on the relevant important structural members. It is desirable to review the allowable deflection amount of the main structural members, and to track the deformation, deflection, and inclination of the main structural members such as beams, columns, and load-bearing walls and review the allowable amount.

Meanwhile, the network server 200 performs a function of transmitting and receiving data necessary for the present invention between the sensor device 300 and the system 100, and as described above, it is preferably a LoRaWan network server including a LoRaWan gateway. The network server 200 may monitor the status of the gateway and the sensor device 300 , and it is preferable to configure the system in a structure that allows remote system control and update.

Referring again to FIG. 1 , the system 100 will be described.

As described above, the system 100 includes the cloud platform 10 , the sensor linkage unit 20 , the building structure safety level analysis unit 30 , and the monitoring/notification unit 40 , and the network server 200 . It receives safety data of the building structure from the sensor device 300 and performs a function of analyzing the safety level of the building structure.

The cloud platform 10 works with the sensor device 300 through the network server 200 to collect sensor data indicating the safety level of the building structure, and manage functions such as distributed storage, inquiry, and analysis of the collected sensor data. server that provides

The cloud platform 10 provides an interface capable of integrating and inquiring sensor data collected from the sensor device 300 , and provides functions such as central control monitoring for management of the sensor device 300 .

The sensor interworking unit 20 interworks with the network server 200 to receive and manage sensor data between the sensor device 300 and the system 100 .

The sensor linkage unit 20 models sensor data according to a preset sensor data transmission standard, and transmits the collected sensor data to the building structure safety level analysis unit 30 to analyze the safety level of the building structure.

The building structure safety level analysis unit 30 (hereinafter simply referred to as the analysis unit 30 ) is a means for analyzing the safety level of the building structure based on sensor data indicating the safety level of the building structure collected by the sensor device 300 .

The analysis unit 30 analyzes the safety level of the building structure according to an analysis algorithm for determining and predicting abnormal signs of the safety level of the building structure by using the simulation mathematical model for safety diagnosis of the building structure.

The analysis unit 30 is a system identification and stiffness estimation function using the constant vibration data collected from the sensor device 300, a macroscopic structural system level stiffness estimation function, a function of identifying spatial changes in the stiffness distribution of structures such as columns and bearing walls carry out

4 is a diagram for explaining an example of an operation of the analysis unit 30 .

4 shows a case in which the analysis unit 30 identifies a system using constant vibration data and performs a stiffness estimation function, and collects and distributes sensor data of a target building structure using an accelerometer, and performs data-based This shows the process of estimating the stiffness by identifying the data-based subspace system after performing the transformation analysis of the building structure by the model.

The analysis unit 30 also provides a risk estimation function based on deformation data of the major structural members of the building structure, selecting a scenario with a high probability of collapse for the target building structure, selecting the critical structural member of the target structure, Estimate the structural collapse scenario due to damage to the critical structural member structure.

In addition, the analysis unit 30 performs a function of identifying signs of damage and destruction of important structural members capable of collapsing of a building structure by using a structural collapse risk estimation technique based on member deformation data.

5 illustrates a scenario analysis process for the main structural members of the analysis unit 30 .

Referring to FIG. 5 , it can be seen that, after the analysis unit 30 collects sensor data for the major structural members of the building structure, the safety level is analyzed by calculating the structural collapse scenario for each major structural member.

6 is a diagram illustrating the configuration of the analysis unit 30 .

Referring to FIG. 6 , the analysis unit 30 includes an absolute management unit 31 , a continuous management unit 32 , and a complex management unit 33 .

The absolute management unit 31 is a means for performing an analysis of sensor data based on a mathematical model, and analyzes the safety level of the building structure according to an alarm setting standard according to a preset specific sensor value, and monitors/notifies an alarm signal according to the result It performs a function of transmitting to the unit 40 .

7 illustrates an alarm setting standard according to a specific sensor value used by the absolute management unit 31 .

Referring to FIG. 7 , it can be seen that safety/caution/warning three alarm signals are set according to each sensor device 300 such as an accelerometer, GPS, inclinometer, and strain gauge.

The continuous management unit 32 is a means for performing an analysis of sensor data based on a mathematical model, and analyzes the safety level of a building structure according to an alarm setting standard according to time series data of a specific sensor value set in advance, and an alarm signal according to the result It performs a function of transmitting to the monitoring/notifying unit 40 .

8 illustrates an alarm setting standard according to time series data of a specific sensor value used by the continuity management unit 32 .

Referring to FIG. 8 , when a preset reference value is continuously exceeded according to a time-series change of a sensor value of each sensor device 300 such as an accelerometer, a GPS, an inclinometer, or a strain gauge, safety / caution / warning three types of alarms You can see that the signal is set.

The complex management unit 33 is a means for analyzing the sensor data of the building structure based on deep learning, and based on the sensor data from the sensor device 300, the classifier ( classifier) and transmits an alarm signal according to the result to the monitoring/notification unit 40 .

9 is a diagram for explaining the operation of the complex management unit 33 .

In FIG. 9 , the complex manager 33 may use three predictive models.

The first predictive model is a model that predicts the probability of reaching a dangerous state in the next cycle. It refines the data by collecting or processing sensor data for each measurement cycle (minutes, hours, days, etc.), and calculates the probability of becoming a dangerous state, such as caution, danger, or alarm, in the next cycle for a certain cycle, resulting in the highest probability It is a model that predicts the state transition of

A classifier may be enabled to perform training and prediction using, for example, an LSTM deep learning model.

The second predictive model is a model that predicts the time until reaching the dangerous state. It trains the classifier by time series data and cluster classification test data for the measurement period, and calculates the probabilistic inference that can be transferred to the cluster state by inputting a sensor value at a specific point in time to transition to a dangerous state (danger, alarm, etc.) It is possible to use a method of predicting the period that is likely to be As a regression model, training and prediction can be performed using an LSTM deep learning model.

The third predictive model is a model for classifying risk states. This is a method of classifying a certain number (3 to 5) by learning a classification model by clustering the characteristics of the sensor values. It can be used when you want to define a more detailed alarm level based on the three-level classification of caution, danger, and alarm. In this model, classification learning is possible through unsupervised learning, and fine tuning may be required by human intervention on the classification result. In this model, the accuracy of the cluster classification model can be improved as data collection is accumulated.

10 is a diagram for explaining the overall configuration of the complex management unit 33 .

Referring to FIG. 10 , the complex management unit 33 includes a collection agent 331 , a distributed storage module 332 , a distributed processing module 333 , an analysis processing unit 334 , a deep learning operation unit 335 , and an external interface 336 . ) is included.

The collection agent 331 performs a function of collecting a large amount of sensor data transmitted from the sensor device 300 , and transmits the collected sensor data to the distributed storage module 332 .

The distributed storage module 332 distributes and stores the collected sensor data and transmits it to the distributed processing module 333 .

The distributed processing module 333 performs a function of distributing and processing the necessary pre-processing before transmitting the sensor data to the analysis processing unit 334 , and the processed sensor data is transmitted to the analysis processing unit 334 .

The analysis processing unit 334 and the deep learning operation unit 335 analyze the sensor data based on deep learning as described in FIG. 9 by a classifier learned based on deep learning, and an alarm signal according to the result It performs a function of transmitting to the monitoring/notifying unit 40 through the external interface 336 .

The external interface 336 performs a function of transmitting the alarm signal transmitted from the analysis processing unit 334 and the deep learning operation unit 335 to the monitoring/notifying unit 40 in conjunction with the monitoring/notifying unit 40 .

The monitoring/notifying unit 40 receives and displays an alarm signal according to a result of analyzing the sensor data by the analysis unit 30 .

The monitoring/notification unit 40 transmits an alarm signal delivered in real time to the administrator's computer, and enables the alarm signal to be displayed through a dashboard displayed on a display unit connected to the administrator's computer. In addition, an alarm signal is transmitted in real time to a device such as a smartphone or computer of an administrator or related person. In this case, it is preferable to display the alarm signal through an application that provides a monitoring/notification function.

11 and 12 show an example in which an alarm signal by the monitoring/notifying unit 40 is displayed.

11 shows that the alarm signal has been generated by displaying the location of the building structure in which the alarm signal has occurred on the map through the dashboard connected to the manager's computer.

FIG. 12 shows detailed information of a building structure in which an alarm signal is generated, and it can be seen that the alarm history, values of specific sensor data, trends indicating changes, etc. are visually displayed and shown.

Although the present invention has been described with reference to the preferred embodiment according to the present invention above, the present invention is not limited to the above embodiment, and of course, other various modifications and variations are possible.

100...Safety diagnosis system through machine learning-based natural vibration value learning
10...Cloud Platform
20...sensor linkage
30...Building Structure Safety Analysis Department
40...Monitoring/Notification Department

Claims (3)

As a safety diagnosis system through machine learning-based building structure natural vibration value learning,
a cloud platform that collects sensor data indicating the safety level of the building structure from the sensor device and manages the collected sensor data;
a building structure safety level analysis unit that analyzes the safety level of the building structure based on the sensor data indicating the level of safety of the building structure collected from the sensor device; and
Monitoring/notification unit for receiving and displaying an alarm signal according to the result of analyzing sensor data in the building structure safety analysis unit
including,
The building structure safety analysis unit,
an absolute management unit that analyzes the safety level of a building structure according to an alarm setting standard according to a preset specific sensor value;
a continuous management unit that analyzes the safety level of a building structure according to an alarm setting standard according to time series data of a preset specific sensor value; and
A complex management unit that analyzes the safety level of a building structure by a classifier learned based on deep learning based on sensor data from a sensor device
Safety diagnosis system through machine learning-based building structure intrinsic vibration value learning, characterized in that it comprises a.
The method according to claim 1,
The complex management unit,
The safety level is analyzed by at least one of a first model that predicts the probability of reaching a dangerous state in the next cycle, a second model that predicts the time to reach a dangerous state, and a third model that classifies the dangerous state A safety diagnosis system through machine learning-based building structure natural vibration value learning, characterized in that.
3. The method according to claim 2,
The first model refines the data by collecting or processing sensor data for each measurement cycle, and calculates the probability of becoming a dangerous state such as caution, danger, and alarm in the next cycle for a certain period to determine the highest probability of state change. The second model predicts and analyzes the safety level, and the second model trains the classifier based on time series data and cluster classification test data for the measurement period, and inputs a sensor value at a specific point in time to perform probabilistic inference that can be transferred to the cluster state. The third model analyzes the safety level by calculating and predicting a period with a high probability of transitioning to a dangerous state, and the third model learns the classification model by clustering the characteristics of the sensor value, classifying it into a preset number, and analyzing the safety level A safety diagnosis system through machine learning-based natural vibration value learning of building structures.

Images