KR101965879B1 - Ann-based sustainable strain sensing model system, structural health assessment system and method - Google Patents

Abstract

The present invention relates to an artificial neural network based stress prediction model system, a safety evaluation system, and a method of evaluating the same, and more particularly, to a wind sensor that is installed in a high-rise structure and measures wind information in real time. A plurality of strain gauges installed on the high-rise structure, and a strain sensor for measuring the strain of the structure according to the wind information measured by the wind sensor and an artificial neural network (ANN) formed by a plurality of wind sensors and strain sensors And a stress prediction model generation unit for generating stress prediction models of the high-rise structure according to the wind loads by training the strain data according to the information.
The present invention provides a system for evaluating safety by predicting a stress of a structure using only wind information and generating a safety evaluation model of a sustainable high-rise structure, an evaluation system using the same, and an evaluation method thereof.

Description

Technical Field [0001] The present invention relates to an artificial neural network-based high-story structure stress prediction model system, a safety evaluation system using the same, and a method of evaluating the same. [0002]

The present invention relates to a stress prediction model system for a high-rise structure, more particularly, to an artificial neural network-based high-rise structure stress prediction model system capable of predicting a stress of a structure by wind information only and evaluating safety, .

Structural Health Monitoring (SHM) studies are being conducted to evaluate the safety of high-rise or high-rise buildings. SHM has installed a strain sensor such as an electric resistance sensor or FBG (Fiber Bragg Grating sensor) on the structure, estimates the stress through the measured strain value, and evaluates the safety of the structure against the allowable stress presented in the design standard.

Recently, in the design of high-rise buildings, sensors are installed in major structural members for safety evaluation in the construction and use stages of buildings. However, in order to evaluate the safety of high-rise structures, it is necessary to install a large number of strain sensors per a large number of members of the building. However, the installed strain gauge has problems such as instability of power supply, difficulties in maintaining and managing, And the occurrence of defects in the sensor itself. A wireless sensor system has been introduced in large structures such as high-rise buildings, but this has limitations in that measurement data is lost due to defects in the sensor itself and communication failure during wireless transmission / reception of measurement data.

Therefore, SHM system is needed for long term monitoring of high - rise buildings considering problems such as sensor defects and data loss. Artificial Neural Network (ANN) can be introduced to predict the response to wind load, which is one of the most important loads affecting high-rise buildings when stress measurement is impossible due to sensor defects and loss. The artificial neural network trains the pre-established input and output data to form a network and predicts the result when new input data is received.

The conventional invention using an artificial neural network uses an average value of strain data measured by wind information. Since the safety evaluation through the strain measurement of the structure is evaluated by the maximum strain acting on the member, the prior art has a limit in that the maximum strain can not be taken into consideration. Therefore, even if stress measurement is impossible due to sensor defects and loss, sustainable safety evaluation technology is needed considering the range of strain variation.

Korean Registered Patent No. 10-1227776 (Registration date: Jan. 23, 2013) Korean Registered Patent No. 10-1431237 (Registration date: Aug. 11, 2014)

The neural network-based high-layer structure stress prediction model system according to the present invention, the safety evaluation system using the same, and the evaluation method thereof have the following problems.

First, the present invention is to provide a stress prediction model system, evaluation system, and evaluation method of a high-rise structure which can estimate the stress of a structure by only wind information and evaluate safety.

Second, the present invention provides a system for generating a safety evaluation model of a sustainable high-rise structure, an evaluation system using the system, and an evaluation method thereof.

The present invention has been made in view of the above problems, and it is an object of the present invention to provide an apparatus and method for controlling the same.

According to a first aspect of the present invention, there is provided a stress prediction model system for a high-rise structure based on an artificial neural network, comprising: a wind sensor installed in a high-rise structure to measure wind information in real time; A plurality of strain sensors installed in the high-rise structure for measuring a strain of the structure according to the wind information measured by the wind sensor; And a stress prediction model generator for generating a stress prediction model of a high-rise structure according to wind loads by training strain data according to wind information based on an artificial neural network (ANN) formed through a plurality of wind sensors and strain sensors do.

Preferably, the wind sensor is a wind direction anemometer that measures the wind direction and the wind speed, and the wind sensor is preferably installed on the uppermost layer of the high-rise structure. The strain sensor is a strain sensor that measures stress of the high- It is preferably a fiber Bragg grating (FBG) sensor.

The artificial neural network may further comprise: an input layer having the wind information as an input value; And an output layer having the strain as an output value, and the output value is preferably a maximum strain value and a minimum strain value of the high-rise structure. The maximum and minimum strains are components that reflect the variation range of the strain to measure the dynamic response of the strain.

In addition, the stress prediction model generating unit generates the stress data by updating the connection weight value in the artificial neural network including the input layer to which the wind information is inputted, the output layer to output the strain, and the two hidden layers, It is preferable to generate the stress prediction model of the high-rise structure according to the wind load based on the training data, and it is preferable that the connection weight value is updated by the Back Propagation Learning Algorithm.

A second aspect of the present invention is a safety evaluation system using an artificial neural network-based high-rise structure stress prediction model, comprising: a wind sensor installed in a high-rise structure to measure wind information in real time; A plurality of strain sensors installed in the high-rise structure for measuring a strain of the structure according to the wind information measured by the wind sensor; A stress prediction model generation unit for generating a stress prediction model of a high-rise structure according to wind loads by training strain data according to wind information based on an artificial neural network (ANN) formed by a plurality of wind sensors and strain sensors; And a safety evaluation unit for evaluating the safety of the high-rise structure by inputting wind information to the generated stress prediction model when the strain sensor response is not possible.

Here, the artificial neural network includes: an input layer having the wind information as an input value; And an output layer having the strain as an output value, and the output value is preferably a maximum strain value and a minimum strain value of the high-rise structure.
The maximum and minimum strains of the strain due to the fluctuating wind load are components that reflect the range of variation of the strain to measure the dynamic response of the strain and the direction of the strain sensor 110 surface and wind direction are not coincident If the dynamic response of the structure is positive, tensile stress is generated, and if negative, compressive stress is generated.

In addition, the stress prediction model generation unit generates the stress data by updating the connection weight value in the artificial neural network including the input layer to which the wind information is input, the output layer to output the strain, and the two hidden layers, It is preferable to generate the stress prediction model of the high-rise structure according to the wind load based on the training data, and it is preferable that the connection weight value is updated by the Back Propagation Learning Algorithm.

According to a third aspect of the present invention, there is provided a safety evaluation method using a stress prediction model based on artificial neural network based on artificial neural network, wherein the above-described evaluation system is used to (a) acquire wind information data measured in real time by a wind speed anemometer installed in a high- ; (b) a plurality of strain sensors installed on the high-rise structure obtain strain data corresponding to the wind information data through real-time measurement; (c) calculating a training data by tracing strain data of a structure according to wind information data using an artificial neural network (ANN) of a stress sensor model generating unit; (d) generating a stress prediction model of a high-rise structure according to a wind load based on the trained data calculated by the safety evaluation section; And (E) if the strain sensor response is not possible, inputting wind information to the generated stress prediction model to evaluate the safety of the high-rise structure.

In the step (c), the strain data of the structure according to the wind information data is trained using the artificial neural network including the input layer having the wind information as the input value and the output layer having the strain as the output value, It is preferable that the output value is a maximum value and a minimum value of the strain of the high-rise structure.

In the step (c), the stress prediction model generation unit generates training data through updating the connection weight value in the artificial neural network including the input layer to which the wind information is input, the output layer to output the strain, and the two hidden layers step; And generating a stress prediction model of a high-rise structure according to a wind load based on the training data. Preferably, the connection weight value is updated using a Back Propagation Learning Algorithm.

The artificial neural network based high-layer structure stress prediction model system according to the present invention, the safety evaluation system using the same, and the evaluation method thereof have the following effects.

First, the present invention provides a stress prediction model system that forms a network for a nonlinear relationship between wind loads (wind information) and member stresses that largely affect high-rise building behavior using an artificial neural network.

Second, the present invention provides a safety evaluation system which can predict the stress of the structure only by wind information and evaluate the safety even if the stress measurement is impossible due to defects and loss of the sensor.

Third, the present invention provides a wind response prediction model system, a safety evaluation system, and a method of a structural member using an artificial neural network (ANN) for sustainable SHM (Structural Health Monitoring) of a high-rise building.

Fourth, the present invention provides a safety evaluation system and a method of evaluating the safety of a column, which is a main member of a high-rise structure, by predicting a strain range.

The effects of the present invention are not limited to those mentioned above, and other effects not mentioned can be clearly understood by those skilled in the art from the following description.

1 is a block diagram of an artificial neural network-based high-story structure stress prediction model system and its safety evaluation system according to an embodiment of the present invention,
2 is a connection model diagram of a stress prediction model system and its safety evaluation system according to an embodiment of the present invention.
3 is a schematic diagram of a wind tunnel test sensor configuration for verifying an embodiment of the present invention.
4 is a schematic diagram showing a training data set of an ANN applied to an embodiment of the present invention.
5 is a graph showing strain data when the static response is dominant.
6 is a graph showing the strain fluctuation width when the static response is dominant.
7 is a graph showing strain data when the dynamic response is dominant.
Fig. 8 is a graph showing the variation range of strain when the dynamic response dominates.
9 is a graph showing the correlation between input and output data of a training data set.
10 is a graph showing the correlation between input and output data of a training data set in polar coordinates.
11 is a graph showing the correlation between input and output data of a training data set in Cartesian coordinates.
12 is a schematic diagram of an artificial neural network (ANN) structure applied to an embodiment of the present invention when there are two hidden layers.
FIG. 13 is a graph showing a result of stress calculation for an artificial neural network (ANN) applied to an embodiment of the present invention when there are two hidden layers.
14 is a graph showing an error calculation result for the ANN applied to the embodiment of the present invention when there are two hidden layers.

Further objects, features and advantages of the present invention will become more apparent from the following detailed description and the accompanying drawings.

Before describing the present invention in detail, it is to be understood that the present invention is capable of various modifications and various embodiments, and the examples described below and illustrated in the drawings are intended to limit the invention to specific embodiments It is to be understood that the invention includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises" or "having" and the like refer to the presence of stated features, integers, steps, operations, elements, components, or combinations thereof, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Further, terms such as " part, "" unit," " module, "and the like described in the specification may mean a unit for processing at least one function or operation.

In the following description of the present invention with reference to the accompanying drawings, the same components are denoted by the same reference numerals regardless of the reference numerals, and redundant explanations thereof will be omitted. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail.

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

FIG. 1 is a block diagram of a stress prediction model system 100 and its safety evaluation system 200 according to an embodiment of the present invention. FIG. 2 is a block diagram of a neural network-based high-layer structure stress prediction model system 100 according to an embodiment of the present invention. Based stress prediction model system 100 and a safety evaluation system 200 thereof.

As shown in FIG. 1, an artificial neural network-based high-story structure stress prediction model system 100 according to an embodiment of the present invention includes a wind sensor 130 installed in a high-rise structure for measuring wind information in real time, A strain sensor 110 installed to measure the strain of the structure according to the wind information measured by the wind sensor 130 and an artificial neural network 110 formed by the plurality of wind sensors 130 and the strain sensor 110. [ : ANN) based on the wind information to generate a stress prediction model of the high-rise structure according to the wind load.

Thus, embodiments of the present invention provide a network of non-linear relationships between wind loads (wind information) and member stresses that primarily affect the behavior of high-rise structures using artificial neural networks, resulting in sensor defects and loss We propose a stress prediction model system and its safety evaluation system (200) for safety assessment of sustainable high-rise structures that can predict the stress of a structure only with wind information even if stress measurement is impossible.

Structural elements of the artificial neural network-based high-story structure stress prediction model system 100 according to an embodiment of the present invention will be described in more detail as follows.

As shown in FIG. 1, the wind sensor 130 installed in the high-rise structure is preferably a wind direction anemometer, and is preferably installed on the uppermost layer of the high-rise structure. The wind sensor 130 includes wind speeds and wind directions, Wind information is measured. The use of the wind sensor 130 in the embodiment of the present invention is advantageous in that the strain sensor 110 installed in the high-rise structure is lost or communication is interrupted and the strain prediction becomes impossible, In this paper, a prediction model system and its safety evaluation system which can accurately estimate the strain of a high - rise structure and evaluate stability at all times even when sensor measurement can not be done by creating and updating a stress prediction model of a high - 200). In addition, it is needless to say that the wind sensor 130 applied to the embodiment of the present invention can be any sensor including wind direction and wind speed and capable of measuring various wind information.

It is preferable to use FBG sensors as stress sensors for the strain sensors 110 provided in a plurality of members such as pillars of the high-rise structure. Fiber Bragg grating (FBG) sensors, which are installed to measure the strain or strain of high-rise structures, are optical sensors that are more efficient, harsh environments, dispersive systems, and longer usable than conventional electrical or mechanical sensors There are advantages.

Strain The FBG sensor is a measuring instrument that uses a principle that a specific wavelength is radiated according to the nature of the grating. When the FBG expands or contracts, the interval of the grating changes, and the wavelength of the reflected light changes Is an efficient and high-performance stress measuring instrument that uses the principle that A plurality of FBG stress or strain sensors 110 are installed on a column portion of a high-rise structure to measure stress or strain in real time. A plurality of angular sensors function as nodes constituting one artificial neural network .

The stress prediction model generation unit 150 trains strain data according to the wind information based on the artificial neural network (ANN) formed through the plurality of wind sensors 130 and the strain sensor 110 as described above, And the safety evaluation unit 210 is a device for generating a stress prediction model of a high-rise structure according to the above-described predictive model, using the model generated by the prediction model system, It is a device that can evaluate the safety of high-rise structures by predicting the strain.

1, the stress prediction model generation unit 150 and the safety evaluation unit 210 may be constituted by one computing device, and the computing device may be a server having a wireless communication device, a computer and a notebook computer A microprocessor, or the like, and may be an arithmetic processing unit capable of executing a computer program.

As described above, the present invention proposes a stress prediction model and a safety evaluation model of a high-rise structure using an artificial neural network as an alternative to continuous or long-term safety evaluation of a high-rise structure by sensing.

In the embodiment of the present invention, a system and a method for introducing an artificial neural network to predict a response to a wind load, which is one of the most important loads affecting a high-rise structure or a high-rise building when measurement data is lost due to a sensor defect I suggest. The artificial neural network trains the pre-established input and output data to form a network, and it is possible to predict and output the results through the network formed when new input data is received.

In addition, the stress prediction model system 100, the safety evaluation system 200, and the safety evaluation method according to the embodiment of the present invention are designed to solve complex nonlinear problems such as dynamic responses of wind and wind caused by many uncertainties, We propose a system and method to introduce neural network. If the nonlinear relationship between the wind load and the safety of the structure, ie the stress of the member, which mainly affects the behavior of the high - rise buildings, is defined by the neural network using the applicability to the nonlinear problem of the artificial neural network, Stress prediction of structures can be made with wind information alone and safety evaluation can be done even when stress measurement is impossible.

That is, in the embodiment of the present invention, a predictive model for response of the wind to the main structural member using the artificial neural network (ANN) is proposed for sustainable SHM (Structural Health Monitoring) of the high-rise buildings. That is, the embodiment of the present invention provides a method and apparatus for estimating the strain (strain) used to calculate wind information and member strain through real-time measurement of an FBG sensor used for at least one of a wind direction anemometer and strain sensor 110 for generating training data of ANN The data is measured and the ANN is used to train these data sets.

When the strain sensor 110 is lost, only the wind information obtained through the wind speed anemometer, which is relatively easy to measure and less likely to lose measurement data, is input to the trained artificial neural network as compared with the strain sensor 110 A system and a method for predicting the strain (stress) of a member (with a sensor attached thereto) are proposed.

In the artificial neural network proposed in the embodiment of the present invention, the wind direction and wind speed are set as input layers, and the strain of each column is set as an output layer. The output layer is constructed so that the maximum and minimum values of the strain are output in order to take into account the variation range of the strain response to the dynamic characteristics of the wind.

That is, the artificial neural network-based high-story structure stress prediction model system 100 according to the embodiment of the present invention, the safety evaluation system 200 using the same, and the evaluation method thereof are provided with a wind direction anemometer and a strain sensor 110 for generating training data of an artificial neural network, Strain data used to calculate wind information and member stresses are measured through real-time measurement of the FBG sensor, and these data sets are trained using artificial neural networks.

Then, when defects of the strain sensor 110 and data loss occur, only the wind information obtained through the wind direction anemometer, which is comparatively easy to measure and less likely to lose measurement data than the strain sensor 110, is input to the trained artificial neural network (Strain) of a member to which the sensor is attached, and a method therefor. In the artificial neural network proposed in the present invention, the wind direction and the wind velocity are set as input layers, the strain of each column is set as an output layer, and the maximum and minimum values of the strain The output layer is formed.

At the end of the artificial neural network training, the verification data set which is not used for the training and the training neural network are verified through the output of the wind direction, the wind speed input and the strain as in the case of the training. It is preferable to use a method using a Back Propagation Neural Network (BPNN). In addition, it can be applied through various techniques of artificial neural networks.

Hereinafter, in order to verify an artificial neural network-based high-layer structure stress prediction model system 100, a safety evaluation system 200 using the artificial neural network-based stress prediction model system 100, and a safety evaluation method thereof according to an embodiment of the present invention, The results of this study are as follows.

Scaled model specimens simulated core, outsourced columns and outrigger systems, which are representative structural systems of high - rise buildings, and measured strain rates of pillars of structures, which are one of the major members of each time, controlling wind speed and direction. Wind direction, wind speed data, and measured strain are used as training data sets for artificial neural networks.

Wind tunnel experiment

Lateral loads, which have the greatest influence on the behavior of high-rise buildings, are mainly determined by wind loads. When designing a high-rise building, it is necessary to estimate the wind load and various estimation methods are available. Most high-rise buildings are often located in urban areas, so they are affected by the airflow generated by other buildings. Therefore, wind tunnel tests are carried out in the design of high - rise buildings, and the wind loads and wind responses are calculated based on them. Wind tunnel tests are performed by constructing a scale model of the target building, constructing conditions similar to the area where the structure will be constructed, and measuring the response of the model considering the wind speed in the area. It is used to measure acceleration, displacement, wind pressure, etc. by installing a sensor on the model to judge the suitability of the design and the influence of the wind on the building. Cluni et al (2011) used a high-frequency force balance (HFFB) method and a multi-pressure sensing system (HFFB) to analyze the effects of wind on orthopedic and atypical high-rise buildings. (SMPSS) technology for wind tunnel experiments. Kim et al. (2009) conducted a study on the prediction of acceleration responses during typhoons through wind tunnel tests and FE models for three reinforced concrete high-rise buildings.

Li et al (2006) conducted a wind tunnel test using the HFFB method for the Jin Mao Building, one of China's high-rise buildings. Through the response measurement of the actual building during typhoon, This study was carried out to analyze the impact of typhoons. Most wind tunnel tests measure wind pressure through a sensor attached to the model body, but experiments are also carried out through various sensors such as acceleration and displacement gauges.

Hwang et al (2011) conducted a wind tunnel experiment to install an LED target on the top layer of a high-rise building model and to measure displacement through a camera installed on the floor to verify the algorithm for predicting the wind response of high-rise buildings. In addition, a motion capture system (MCS) capable of measuring displacement by measuring three-dimensional coordinates through the markers attached to the object is used in a wide variety of applications, including wind field measurement (Park et al., 2013) Studies on wind tunnel experiments have been conducted.

In the embodiment of the present invention, to verify the wind response prediction technique of high-rise building using ANN, a miniature model having a typical structure type of a high-rise building was manufactured and a wind tunnel test was performed.

FIG. 3 is a schematic diagram of a wind tunnel test sensor configuration for verifying an embodiment of the present invention. As shown in FIG. 3, the model used in the wind tunnel test is a regular structure, and in order to simulate a typical structural system of a high- And 8 outboard columns were placed on the outside of the building. The outrigger was installed on the top floor of the building, and the mass was divided into five layers.

During the wind tunnel test, the outer shell was installed outside the test body to make the building similar to the wind received through the wind pressure area. In the wind tunnel test, the reduction ratio of the model, the reduction ratio of the wind speed, the data measurement interval, and the measurement time should be considered, but the test object used in the embodiment of the present invention is not designed for a specific building, The specific boss ratio was not taken into consideration when setting the experimental wind speed. The wind tunnel test was carried out at the Experimental Experiment Station of wind tunnel experiment of Korea D Company. The specimen is placed in the wind tunnel turntable. The turntable is rotatable so that the building can adjust the wind angle to the desired angle. Also, the experiment was conducted by changing the wind power source for wind load input from 100 rpm to 400 rpm in order to control the wind speed.

An FBG sensor, a strain measuring sensor, is installed in the test specimen, and it is installed at the lower part of the specimen column where the stress is most generated by the lateral force. The position of the sensor installed in the test object can be confirmed in Fig. The FBG sensor was installed at the lower end of six columns to measure strain. Wind speed experiments were performed on 35 cases at 5cases (0 °, 30 °, 45 °, 60 °, 90 °) for 7cases (2.5, 3.75, 5, 6.25, 7.5, 8.75 and 10m / s) Respectively. The angle of the wind direction was changed counterclockwise with reference to 0 °.

A total of 32 cases were used for ANN model training and verification except for 3 cases where measurement errors occurred in 35cases total. A total of 10 cases were excluded from the training for the verification of the ANN model, including case containing 6.25m / s and case containing 30 ° of 32 cases. Therefore, the ANN model was trained with the data set of the remaining 22 cases. A total of 22 training data sets were constructed with the maximum and minimum values of the strain of the 6 columns in wind speed and wind direction as one set. The data set used for training and verification is shown in Figure 4 by wind speed and wind direction.

Stress prediction model using artificial neural network (ANN)

Artificial Neural Network (ANN)

Artificial neural network model is a mathematical model that imitates human brain learning method. Like the cerebral neural network, the input signal to the neuron and the output signal from the neuron are represented by the respective input / output nodes, and the connection strength between the nodes is determined by the coupling strength of the synapse. In Artificial Neural Networks, MultiLayer Perceptron is often applied. In this case, it consists of an input layer, an output layer, and one or more hidden layers.

Each layer is made up of a number of neurons, and each layer of neurons is connected to neurons in the other layer by a connection weight. One neuron in the artificial neural network acts as a biological neuron in the biological neural network. 1) The value input to the input layer is added to the hidden layer neuron by adding the product of the connection weight between the input layer and the hidden layer. 2) The input value is different from the input value through the nonlinear activation function in the neuron. 3) The process of transmitting output results to other hidden neurons or other neurons in the output layer is called a feed forward operation, and the process of modifying the value of the connection weight in the neural network so that the calculated output value coincides with the desired target value is repeated , Which is defined as learning. Here, the update of the value of the connection weights is performed by the Back Propagation Learning Algorithm (Rumelhart et al, 1986). When the specified epoch is reached or the training error reaches the target value, the operation is stopped.

Training data selection method

In the embodiment of the present invention, the prediction of response of the high-rise buildings according to the wind is handled using ANN, and the wind information is considered as the input signal of the ANN training data. On the top floor of the high-rise building, wind direction anemometer is generally installed, and the wind direction and wind speed blowing to the building can be measured and transmitted to the user in real time. The strain measured on the column was used as the output signal of the ANN training data for the stress prediction of high - rise column members. The input signal of the wind direction and the wind speed of the same time and the output signal of the strain of the column to be monitored become training data in one set. At the end of the ANN training, the verification data set that was not used for the training, and the neural network trained through the input of wind direction, wind speed and strain, as in the case of training, are verified.

The wind response of the tall building to the wind, in other words the wind load, can be divided into static and dynamic components. In the case of displacement in the wind response, for example, a certain amount of lateral displacement occurs in a high-rise building due to the stiffness of the structure due to wind load. At the same time, a dynamic displacement occurs, which is difficult to predict its direction and value due to turbulence effect and vortex effect. The former case can be regarded as a static component of the high-rise wind response and the latter as a dynamic component. The average wind and the fluctuating wind load can be estimated by using the wind force measured through the wind tunnel test when designing the wind storm of a high-rise building.

Since the average wind load is the load caused by the average component of the wind, it is calculated by using the average value of the measured wind force and is related to the above-mentioned static response component. Variable wind loads are generated by winds that fluctuate irregularly without a constant pattern. This causes the building to vibrate and the inertial force is generated by this vibration, which acts as an additional load on the building and corresponds to the dynamic response component.

Therefore, in the embodiment of the present invention, the strain to be subjected to the high-rise building wind response should also be considered in terms of the static component and the dynamic component. As a result of the measurement of the wind tunnel test, the strain measured at the column according to the angle between the blowing wind and the column surface on which the strain sensor 110 is installed is mainly composed of a static component, It can be divided into the case where the dynamic component is mainly represented.

First, when the static response dominates, as described above, the general case in which the static response dominates among the measurement results of the wind tunnel test is described as an example. Strain data measured at FBG1 at 6.25 m / s wind speed and 0 degree wind direction were used as examples. This tendency is conspicuous when the direction of the surface of the strain sensor 110 installed on the column coincides with the direction of the wind.

It can be seen that the strain response of the column when a uniform wind is blown for a certain period of time remains at a relatively constant value as shown in FIG. However, even if the static response dominates, a fine dynamic response component is included. 6 shows the relationship between the wind speed and the strain at the time of measuring the strain. 13.27eu shows the fluctuation range.

Second, the response characteristics of dynamic responses dominated by the FBG1 sensor were examined in the case of the above-mentioned static response in the dominant case and in the case of 90 ° in the wind direction. This tendency is conspicuous when the direction of the strain sensor 110 installed on the column and the direction of the wind are perpendicular to each other. The pylon facing the wind is bent in the same direction as the direction of the wind with the stiffness of the column, causing a static displacement (of course, a fine dynamic response is added).

When the windward pillar surface is flexed in the wind direction, the plane perpendicular to the wind direction produces only a static displacement and responds only to the dynamic characteristics of the wind, such as vibration. If the strain sensor 110 is installed on a plane perpendicular to the wind direction at this time, the dynamic response will be measured. This characteristic can be confirmed by the strain measurement results in the corresponding situation. Although the uniform wind was blown for a certain period of time, the strain response of the column to the wind at right angles to the wind was large as shown in Fig. Here, the + sign means that a tensile stress is generated in the column, and the - sign means that a compressive stress occurs.

FIG. 8 shows the variation of 34.13ue due to the relationship between the wind speed and the strain at the time of the corresponding strain, which is larger than the first case where the static response dominates the same wind speed.

Therefore, in the embodiment of the present invention, in order to take into account the dynamic components appearing in the above-mentioned two cases when the measured wind response, that is, the strain of the column, is selected by the training data, The range of variation of strain was considered without selecting a value. The strain range can be expressed by the maximum and minimum values of the strain that occur for a specific wind direction and wind speed. Therefore, as the training data, the wind direction and the wind speed are input, and the maximum value and the minimum value of the generated strain values are selected as one set. Therefore, the strain training data was selected by the following procedure.

1) Designate a section where a constant strain response occurs (the green line in FIGS. 5 and 7)

2) Filter the data judged as noise in sensor measurement in the applicable range (10Hz lowpass filter, yellow line in Fig. 5 and Fig. 7)

3) Extracting the maximum and minimum values (black mark in FIGS. 5 and 7)

Correlation analysis of training data

The training data set was generated according to the data selection method described above. The relationship between the input data and the output data of the training data set was examined prior to ANN training. First, the relationship between the wind speed and strain rate, which is one of the input data, is shown in Fig. The data shown in Fig. 9 is the strain value (blue) measured at the FBG1 installed on the column when the wind speed condition is increased from 2.5 m / s to 10 m / s under the 0 degree wind direction condition.

The entire strain value is expressed by curve fitting with a second order polynomial function (red line), and the maximum strain value and the strain minimum value are also expressed by curve fitting by a second order polynomial function (pink line) . As can be seen from FIG. 9, the strain tends to increase nonlinearly as the wind speed increases. This is similar to the measured data for other wind directions.

As described above, the strain range of each wind speed varies depending on the dynamic component of the wind, and the magnitude of the variation range tends to increase as the wind speed increases. In addition, curve fitting can be performed by curve fitting with a quadratic polynomial function having a relatively low order. However, due to the influence of dynamic response to wind, fitting to the entire measured strain data, It can be seen that the fitting curve is somewhat different. Therefore, it can be seen that the application of ANN to the relationship between wind speed and strain having such a nonlinear characteristic is valid.

In order to examine the relationship between wind direction and strain, which is one of the input data, polar coordinate type and rectangular coordinate type are shown in FIGS. 10 and 11, respectively. Similarly, the relationship between wind direction and strain rate was examined only at wind speeds of 6.25 m / s in FBG1. As the wind direction changes from 0 ° (wind direction) to 90 ° (wind right angle direction), the strain value increases slightly at 30 ° and then decreases again. In the other cases except for 90 degrees, the range of variation is relatively constant, while the range of variation is drastically increasing at 90 degrees in the direction of the wind.

This tendency was confirmed in other wind data. FIG. 11 shows curve fitting for the entire measurement strain and the maximum and minimum values. Curve fitting was difficult with the second order polynomial function and fitting was performed with the third polynomial function. As can be seen from the fitted curve of FIG. 11, it can be seen that the nonlinearity increases slightly at around 30 degrees and decreases again.

In addition, fitting (red line) for all data and fitting (pink line) for maximum and minimum strains tend to have a similar tendency to the strain change as a whole, but at 90 degrees, the influence of the dynamic response component is expanded, fitting results. Therefore, as can be seen from the relationship between the wind speed and the strain rate, the application of ANN to the nonlinear correlation between the input data and the output data of the training data set is judged to be reasonable.

Neural network composition

The relationship between wind direction and wind speed and strain is trained by ANN, and the strain generated in the high-rise building columns is predicted by the new wind direction and wind speed through the trained neural network. In this study, ANN was used as a multi - layer perceptron ANN and it was composed of a feedforward back propagation network. This is implemented in MATLAB. The ANN is composed of neurons and transmits information of the input layer to the output layer. In the case of a multilayer perceptron, a hidden layer exists, and the input values arriving at the hidden layer are multiplied by weights and summed to apply a nonlinear transfer function. In the output layer, the values delivered from the hidden layer are multiplied by weights, summed, and the linear transfer function is applied to output the final output value.

In general, there are no specific rules for the number of hidden layers and the number of neurons in each layer, and the network structure is found by trial and error due to a given problem. In the embodiment of the present invention, the result of case (case 1) in which two hidden layers are formed is analyzed.

Also, if too many neurons are used in the hidden layer, overfitting can result in poor training results, and if too few neurons are used, underfitting can also result in poor training (Haykin, S. 1999). Therefore, it is important to find the number of neurons suited to the problem. However, in the embodiment of the present invention, experimental results are shown by applying two cases of hidden layers and three neurons (case 1).

On the other hand, the input node has two nodes, wind direction and wind speed. The output layer was set to be twice the number of columns to be monitored, that is, the number of strains monitored, so that the maximum and minimum values of column strain were output. FIG. 12 shows an ANN architecture to which an embodiment of the present invention is applied. The nonlinear transfer function of the hidden layer uses a sigmoid function and the output layer uses a linear transfer function. The termination condition was set to 1000 epochs.

Verification and Results by Wind Tunnel Experiment

ANN application result (6.25m / s / 30 degree verification)

As described above, the ANN architecture was used to train each set of training data obtained through wind tunnel experiments. For the trained ANN Architecture, verification was performed with data sets not used for training. All data sets including 6.25 m / s and 30 ° in training data selection were not used for training. Of the 10 data sets not used for training, only 6.25m / s and 30 ° were selected for the verification data set. A training set of wind speed and direction combinations other than 6.25 m / s wind speed conditions and 30 degree wind direction conditions, for example, for the remaining nine data sets, such as 6.25 m / s wind speed conditions and 0 degree wind direction conditions Will be described below.

As described above, Case 1 is composed of two hidden layers, and each hidden layer has three neurons. The strain prediction results and errors are shown in FIGS. 13 and 14, respectively. As can be seen from FIG. 13, it can be seen that the blue box as the verification strain and the red box as the predicted strain overlap considerably closely. As can be seen from FIG. 14, the maximum error was -11.5%, which was able to predict the strain of most columns relatively accurately.

In addition, the maximum strain is an important factor in the structural member stress evaluation. In the maximum strain prediction of each sensor, only one sensor (FBG2) can predict the maximum strain with less than 10% error. In particular, it can be seen that the red box, which is the strain range predicted in FIG. 13, covers the range of the verification strain in all sensors except the FBG sensor No. 4. It can be predicted that the maximum and minimum range of strain can be accurately estimated, and it can be practically used for column safety evaluation.

As described above, the embodiment of the present invention introduces ANN which is used for a problem having a high nonlinear characteristic in order to evaluate the safety of a sustainable high-rise building, and a nonlinear relationship between the dynamic characteristic of the wind and the response of the building through ANN We propose a prediction model system that generates a stress prediction model for high - rise building columns according to wind loads. The model generated by the proposed system was verified by using strain and wind data measured by wind tunnel test. In order to simulate actual high - rise buildings, a miniature model consisting of core, outer column and outrigger was fabricated and wind tunnel test was performed under wind direction and wind speed. In order to measure the stresses of columns, which are the main structural members of high - rise buildings, the strains according to wind velocity and wind direction were measured through strain gauge FBG sensor and used as training data of ANN.

In order to propose a new ANN model that predicts the strain generated on a building column by inputting the wind direction and wind speed, we have selected a suitable ANN model by changing the number of hidden layers and the number of neurons. The ANN model was trained with the same training set as the three case 1 cases, and the strain rate was predicted for wind direction and wind speed not used for training.

As a result, case 1 predicted column strain relatively accurately. It is confirmed that more accurate models are generated than the existing studies on the number of hidden layers. The number of neurons is so high that the training is not done well, but it needs to be adjusted according to the characteristics of the problem.

Also, in the embodiment of the present invention, a system and method for evaluating the safety of a structural member according to the wind of a high-rise building have been proposed. In the case of a tall building in which the response includes the variability due to dynamic factors due to wind, in addition to the static factor, it is more reasonable to predict the range of response values covering the dynamic response than to predict specific response values.

In the embodiment of the present invention, the maximum and minimum values of the response strain values are considered in both training data, verification data, and predicted values. The results of the ANN application show that the prediction of strain is relatively accurate in case 1 and at the same time the range of reference strain is included in the range of reference strain to accurately predict the maximum and minimum values of strain have.

It can be seen that the stress prediction model according to the embodiment of the present invention enables the prediction of the range of the strain, and thus it is well applicable in terms of ensuring the safety of the column, which is a main member of the high-rise structure.

The embodiments and the accompanying drawings described in the present specification are merely illustrative of some of the technical ideas included in the present invention. Accordingly, the embodiments disclosed herein are for the purpose of describing rather than limiting the technical spirit of the present invention, and it is apparent that the scope of the technical idea of the present invention is not limited by these embodiments. It will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

100: Stress prediction model system 110: strain sensor
130: Wind sensor 150: Stress prediction model generation unit
200: Safety evaluation system 230: Safety evaluation unit

Claims (12)

delete delete delete delete delete A wind sensor installed on the uppermost layer of the high-rise structure for real-time measurement of wind direction and wind speed information;
A plurality of strain sensors installed in the high-rise structure for measuring a strain of the structure according to the wind information measured by the wind sensor;
A stress prediction model generation unit for generating a stress prediction model of a high-rise structure according to wind loads by training strain data according to wind information based on an artificial neural network (ANN) formed by a plurality of wind sensors and strain sensors; And
And a safety evaluation unit for evaluating the safety of the high-rise structure by inputting wind information to the generated stress prediction model when the strain sensor response is impossible,
The stress prediction model generation unit includes an input layer to which wind information is input, an output layer that outputs a strain maximum value and a minimum strain value of the high-rise structure measured by the strain sensor, an artificial neural network including three hidden neurons in two hidden layers and hidden layers Based on the training data, a stress prediction model for a high-rise structure is generated based on the wind load, and the model is constructed by using the Back Propagation Learning Algorithm.
The maximum strain value and the minimum strain value due to the fluctuating wind load are constituent components reflecting the variation range of the strain to measure the dynamic response of the strain,
If the strain dynamic response of the structure generated when the direction of the strain sensor 110 surface and the direction of the wind do not coincide due to the fluctuating wind load, the tensile stress is generated, and if the (-) sign is compressive stress A safety evaluation system using artificial neural network based stress prediction model based on high - rise structure. delete Using the evaluation system of claim 6,
(a) obtaining wind information data measured in real time by a wind direction anemometer installed on the uppermost layer of the high-rise structure;
(b) a plurality of strain sensors installed on the high-rise structure obtain strain data corresponding to the wind information data through real-time measurement;
(c) calculating a training data by tracing strain data of a structure according to wind information data using a neural network (ANN) of a stress sensor model generating unit (ANN);
(d) generating a stress prediction model of a high-rise structure according to a wind load based on the trained data calculated by the stress prediction model generator; And
(E) a step of evaluating the safety of the high-rise structure by inputting wind information to the generated stress prediction model when the strain sensor response is not possible, and a safety evaluation method using a stress prediction model based on an artificial neural network . The method of claim 8,
The step (c)
Wherein the step of calculating the training data comprises the step of tracing the strain data of the structure according to the wind information data by using the artificial neural network including the input layer having the wind information as the input value and the output layer having the strain as the output value, Safety Assessment Method Using Neural Network - Based High Story Stress Prediction Model. delete The method of claim 8,
The step (c)
Generating training data through renewal of a connection weight value in an artificial neural network including an input layer to which wind information is input, an output layer in which strain is output, and two hidden layers; And
And generating a stress prediction model of the high-rise structure according to the wind load based on the training data, using the stress prediction model of the high-rise structure based on the artificial neural network.

delete

Images