KR102256239B1 - Deep learning-based system for predicting 3d thermal diffusivity into concrete structures after fire, and method for the same - Google Patents

Abstract

By predicting the fire damage temperature by damage depth of the fire damaged concrete structure from the sample of fire damaged concrete structure collected from the fire site and calculating the fire temperature curve, the heat diffusion path of the fire damaged concrete structure is accurately predicted and the flame moving path is easy. In addition, it is possible to trace and reproduce the fire site directly from the fire site by coring-collecting samples of fire-damaged concrete structures, and performing thermogravimetric analysis (TGA), Fourier transform infrared spectroscopy (FTIR) analysis, and other chemical analysis in parallel. Through an experiment, the correlation with high temperature exposure temperature can be accurately identified, and by applying a deep neural network (DNN) model using a deep learning analysis tool, the predicted rate of fire damage temperature for fire damaged concrete structures A system and method for predicting a heat diffusion path of a deep learning-based fire damaged concrete structure that can increase the value are provided.

Description

DEEP LEARNING-BASED SYSTEM FOR PREDICTING 3D THERMAL DIFFUSIVITY INTO CONCRETE STRUCTURES AFTER FIRE, AND METHOD FOR THE SAME}

The present invention relates to the prediction of the heat diffusion path of a fire damaged concrete structure, and more specifically, by predicting the Fire Damaged Temperatures by damage depth of the fire damaged concrete structure and calculating the fire heating curves. The present invention relates to a deep learning-based system for predicting thermal diffusion paths of fire damaged concrete structures and a method for predicting 3D Thermal Diffusivity of damaged concrete structures.

In general, when a concrete structure is exposed to high temperatures such as fire for a long time, the cement hardened body and aggregate undergo different expansion and contraction behaviors, causing cracks or weakening of the structure due to chemical dissociation of hydration products. Changes in the structure and physical properties are greatly degraded, and a huge crack occurs in the concrete due to thermal stress generated by the confinement of the end, and in serious cases, the structure may collapse.

Specifically, the cement hardened body has a large amount of chemically-bonded water in addition to the free water. When the hardened body is exposed to 100℃ or higher, the excess water present in the capillary pores is evaporated by 1300 times. Abnormal volume expansion occurs.

In addition, when the heating temperature is further increased to about 180℃, some of the chemically bonded water in the cement hardened body begins to evaporate. At about 250~350℃, calcium silicate hydration, a key hydrate for the strength development of the hardened cement body, begins to evaporate. About 20% of the moisture contained in the product is lost, and most of the moisture contained is lost when it reaches 400 to 700℃. In addition, in a similar temperature range, calcium hydroxide (Ca(OH) ₂ ), which is a free alkali component in concrete, is also thermally decomposed into quicklime and water, causing chemical damage, increasing somewhat large voids, and losing rigidity, which becomes structurally very dangerous. Thereafter, the concrete gradually melts on the surface when heated at about 1200°C or higher for a long time.

1 is a diagram illustrating that the compressive strength of a concrete structure decreases with an increase in temperature during a fire.

As shown in FIG. 1, in the concrete structure, cement hydrates in the concrete are chemically decomposed as the heating temperature increases, and the cement paste shrinks but the aggregate expands as it is exposed to high temperatures. This is mainly because the internal fracture caused by the difference in the coefficient of thermal expansion according to the quality characteristics of cement paste and aggregate affects the mechanical properties of concrete. Moreover, as a result of moisture expansion of excess water existing in the pores of the concrete capillary, internal stress gradually increases and the internal structure is destroyed, so mechanical properties such as strength and elasticity are deteriorated.

At this time, the degree of reduction varies depending on the type, formulation, age, etc. of the materials used, and exhibits a tendency as shown in FIG. 1. That is, there is little decrease in strength up to 300°C, but it becomes 50% or less when it exceeds 500°C. In addition, at about 700° C., it may be reduced to about 60 to 80% of the compressive strength at room temperature. Accordingly, it can be seen that when the concrete is subjected to high heat by fire, the compressive strength is greatly reduced. In addition, it can be seen that the modulus of elasticity decreases by heating and becomes almost half at 500°C. This is because when the concrete is subjected to high heat, it loses its elastic properties and gradually changes to plasticity.

The performance of these concrete structures is degraded due to the change in the microstructure in the event of a fire. Such microstructure changes, for example, use a gas adsorption method that measures the pore structure by adsorbing nitrogen at the boiling point of nitrogen (-195.8℃). Or, by performing a porosimeter analysis by mercury intrusion, it can be grasped by pore distribution or pore structure characteristics.

Based on the fire mechanism in such concrete structures, it is necessary to accurately diagnose the deterioration of the concrete structure in order to determine whether or not to reuse the fire damaged concrete structure and to determine the damage class.

On the other hand, as a prior art for solving the above-described problem, Korean Patent No. 10-1554165 discloses an invention entitled "Survival Life Prediction System and Method for Fire Damaged Concrete Structure", which is referred to in the present specification. To form part of the present invention.

2 is a block diagram of a system for predicting the remaining life of a fire damaged concrete structure according to the prior art, and FIG. 3 is an operation flow diagram of a method for predicting the remaining life of a fire damaged concrete structure.

Referring to FIG. 2, a system for predicting the remaining life of a fire damaged concrete structure according to the prior art is largely a sample of fire damaged concrete structures 10, a concrete structure remaining life predictor 20, a standardized DB 30, and a fire. It includes a damaged concrete structure diagnosis device 40, wherein the concrete structure residual life prediction unit 20 includes a chemical analysis unit 21, a data analysis unit 22, and a fire damage diagnosis and residual life evaluation unit 23. Includes.

The fire-damaged concrete structure sample 10 is collected by drilling to the target point of the concrete structure from the fire-damaged concrete structure, for example, using a moving core drill, and processed with precision. At this time, the size of the sample 10 can be various sample sizes, and it is preferable that the diameter is 10 mm x length 40 mm. In addition, the collected fire damaged concrete structure sample 10 is precisely cut using various cutting machines. For example, the collected fire damaged concrete structure sample 10 can be cut in various sizes, but a total of 4 samples can be prepared for each depth by cutting by 10 mm.

The concrete structure residual life prediction unit 20 obtains sample data by performing a chemical analysis on the fire damaged concrete structure sample 10 processed as a sample for chemical analysis, and stores the sample data in the standardized DB 30. By predicting the fire damage temperature, void structure, and neutralization depth, respectively, compared with, it is possible to diagnose fire damage to the fire damaged concrete structure sample 10 and evaluate the remaining life.

Specifically, the chemical analysis unit 21 of the concrete structure residual life prediction unit 20 performs a chemical analysis on the fire damaged concrete structure sample 10 processed into a sample for chemical analysis to obtain sample data.

The data analysis unit 22 of the concrete structure residual life prediction unit 20 compares the sample data obtained by the chemical analysis unit 21 with the data previously stored in the standardized DB 30 to make the fire damage temperature, void structure, and neutralization. Predict each depth. Here, the data analysis unit 22 may include a fire damage temperature prediction unit 22a, a void structure prediction unit 22b, and a neutralization depth prediction unit 22c.

Specifically, the fire damage temperature prediction unit 22a of the data analysis unit 22 compares the sample data obtained by the chemical analysis unit 21 with the data previously stored in the damage temperature DB 31 of the standardized DB 30 Thus, the fire damage temperature is predicted, and the pore structure predicting unit 22b of the data analysis unit 22 compares the data previously stored in the pore structure DB 32 for each temperature of the standardized DB 30 to compare the pore structure for each temperature. In addition, the neutralization depth prediction unit 22c of the data analysis unit 22 predicts the neutralization depth by comparing the data previously stored in the neutralization depth DB 33 for each pore structure of the standardized DB 30.

The fire damage diagnosis and residual life evaluation unit 23 of the concrete structure residual life prediction unit 20 is based on the fire damage temperature, void structure, and neutralization depth analyzed by the data analysis unit 22. ), and evaluate the remaining life.

Standardized DB (Standardized Database: 30) standardizes and stores the experimental data obtained through precision experiments using the fire damaged concrete structure diagnosis device 40, and compares the sample data obtained by the chemical analysis unit 21. It is used as data.

Specifically, the standardized DB 30 includes a damage temperature DB 31, a pore structure DB 32 for each temperature, and a neutralization depth DB 33 for each pore structure.

The damage temperature DB 31 of the standardized DB 30 stores real-time chemical property change data according to the fire damage temperature through various chemical analysis such as XRD analysis of fire damaged concrete structures. In addition, the temperature-specific pore structure DB 32 stores data on changes in pore structure characteristics according to the fire damage temperature by a pore amount analysis device such as BET analysis and porosimeter analysis of fire damaged concrete structures. In addition, the neutralization depth DB 33 for each pore structure stores data on the neutralization depth according to the pore structure change by the accelerated neutralization analysis of the fire damaged concrete structure.

The fire damaged concrete structure diagnosis device 40 is applicable whenever the heating temperature of the concrete structure test body increases for various chemical analysis such as XRD, various pore analysis such as BET (Brunauer-Emmett-Teller) and accelerated carbonation analysis test. By measuring the change in characteristics of the sample by exposure temperature and exposure time, and storing each data, it can be established as a standardized DB (30).

3, the method for predicting the remaining life of a fire damaged concrete structure according to the prior art is, first, a sample 10 is collected from the fire damaged concrete structure, and the sample 10 is precisely cut and processed (S10). , Then, a chemical analysis (XRD) is performed on the precisely cut-processed sample 10 (S20). After that, the chemically analyzed data is used as the first input data.

Next, the sample data subjected to the chemical analysis is compared with the standardized DB 30 (S30). At this time, the standardization DB 30 standardizes and stores the experimental data obtained through a precision experiment using the fire damaged concrete structure diagnosis device 40.

Next, the fire damage temperature, void structure, and neutralization depth corresponding to the sample data are predicted (S40), and then, the fire damage of the concrete structure is diagnosed and the remaining life is evaluated (S50).

According to the prior art, it is possible to obtain sample data by performing a chemical analysis of a fire damaged concrete structure sample, and to quickly predict the remaining life of a fire damaged concrete structure by comparing it with data previously stored in a standardized DB. In addition, it is possible to accurately and scientifically evaluate the level of fire damage to fire damaged concrete structures so that correct repair and reinforcement of fire damaged concrete structures can be made.

In the case of the system and method for predicting the remaining life of fire damaged concrete structures according to the prior art, the fire damage temperature prediction unit 22a of the data analysis unit 22 standardizes the sample data obtained by the chemical analysis unit 21 Damage temperature of DB 30 The fire damage temperature is predicted by comparing with the data previously stored in DB 31. Because only the chemical composition change by exposure temperature of concrete exposed to high temperature is considered through the experiment, the fire damage temperature is accurately determined. There is a problem that it is difficult to predict. That is, there is a problem that the predicted rate of the fire damage temperature is lowered.

Republic of Korea Patent No. 10-1554165 (filing date: December 19, 2014), title of invention: "Survival Life Prediction System and Method for Fire Damaged Concrete Structures" Republic of Korea Patent Registration No. 10-1991825 (filing date: October 24, 2017), title of invention: "Method of estimating the water heat temperature of concrete structures that have been damaged by fire" Republic of Korea Patent No. 10-1920886 (filing date: June 19, 2018), title of invention: "Chemical healing method for preventing secondary damage to fire damaged concrete structures" Republic of Korea Patent Registration No. 10-1792435 (filing date: September 13, 2016), title of invention: "Method for evaluating the degree of damage to structures that have been damaged by fire"

The technical problem to be achieved by the present invention for solving the above-described problem is by predicting the fire damage temperature by damage depth of the fire damaged concrete structure from the fire damaged concrete structure sample collected by coring at the fire site and calculating the fire temperature curve, It is to provide a deep learning-based thermal diffusion path prediction system and method for fire damaged concrete structures that can predict the heat diffusion path of fire damaged concrete structures and trace and reproduce the flame movement path.

Another technical problem to be achieved by the present invention is a deep learning-based thermal diffusion path of a fire damaged concrete structure, which can increase the predictive rate of the fire damage temperature for a fire damaged concrete structure by analyzing it using a deep learning analysis tool. It is to provide a prediction system and method thereof.

As a means for achieving the above-described technical problem, the deep learning-based heat diffusion path prediction system of a fire damaged concrete structure according to the present invention physical and chemical analysis of a sample of a fire damaged concrete structure obtained by coring at a fire site and cut and processed. A sample analysis unit that performs; A deep learning analysis tool that applies a deep neural network model to the analysis result of the sample analysis unit for deep learning analysis and then compares it with a standardized directory to derive an optimal analysis result; A fire damage temperature prediction unit for predicting a fire damage temperature at a specific location for each damage depth of the fire damaged concrete structure according to the deep neural network model analysis of the deep learning analysis tool; A fire temperature curve calculating unit for calculating a fire temperature curve according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure according to the deep neural network model analysis of the deep learning analysis tool; And a heat diffusion path prediction unit for predicting a heat diffusion path according to the passage of time outside the fire damaged concrete structure according to the predicted fire damage temperature and the calculated fire temperature curve, wherein the heat diffusion path prediction unit It is characterized in that the heat diffusion path over time is identified in response to the prediction of the fire damage temperature at the time of fire for the fire damaged concrete structure sample.

The deep learning-based heat diffusion path prediction system of a fire damaged concrete structure according to the present invention tracks the movement path of the flame from the fire place of the fire site as a function of time according to the predicted fire damage temperature and the calculated fire temperature curve, and It may further include a reproducing unit for reproducing the flame movement path tracking.

Here, the sample analysis unit may include a thermogravimetric analysis unit configured to perform thermogravimetric analysis (TGA) measuring a mass reduction rate of the fire damaged concrete structure sample; Fourier transform infrared spectroscopy (FTIR) analysis unit for analyzing the maximum value of the peak appearing at a specific wavelength number and the area of the corresponding part according to the fire damage temperature of the fire damaged concrete structure sample; And a chemical analysis unit that performs chemical analysis in parallel with the chemical analysis equipment according to the analysis purpose.

Here, the higher the fire damage temperature in the thermogravimetric analysis unit, the smaller the total mass reduction rate is, and the higher the overall fire damage temperature in the Fourier transform infrared spectroscopy unit, the maximum peak value and the area of the area decrease. It is done.

Here, the analysis result of the sample analysis unit may be partially borrowed or a value may be changed, and it is preferable to undergo a noise removal process so that the analysis result is not affected by environmental influence factors.

Here, the standardization directory is the result of thermogravimetric analysis (TGA) showing the mass reduction rate by exposure temperature of the sample of fire damaged concrete structure collected by coring at the fire site, by exposure temperature of the sample of fire damaged concrete structure collected by coring at the fire site. It is characterized by including the results of Fourier Transform Infrared Spectroscopy (FTIR) analysis showing chemical composition changes and other analysis results of samples of fire damaged concrete structures collected by coring at the fire site.

Here, the deep learning analysis tool uses the result of the single analysis of the thermogravimetric analysis unit as input data, and calculates the fire damage temperature by the damage depth inside the fire damaged concrete structure and the fire temperature curve of the part corresponding to the heat source outside the fire damaged concrete structure. A first deep neural network model for deriving a final output for calculation; When the correlation value of the first deep neural network model is less than an appropriate level, the analysis result of the thermogravimetric analysis unit, the analysis result of the Fourier transform infrared spectral analysis unit, and the analysis result of the chemical analysis unit are used as input data, and weights for features that are clearly shown are used. A second deep neural network model for deriving a final output for calculating the fire damage temperature by damage depth inside the fire damaged concrete structure and the fire temperature curve of the part corresponding to the heat source outside the fire damaged concrete structure by reanalyzing it; And a data comparison and analysis unit comparing the analysis result of the first deep neural network model or the analysis result of the second deep neural network model with the established standardization directory.

Here, in the second deep neural network model, the feature may be a result value clearly displayed at a specific temperature.

Here, the fire temperature curve calculation unit compares and analyzes the result value obtained from the outer surface of the fire damaged concrete structure and the result value obtained from the inner area, which is each data analyzed in units of one set, and expresses it as a function of time. By comparing and analyzing the results of the fire damaged concrete structure sample, the fire temperature curve is calculated according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure, but the fire temperature curve is the size of the heat receiving temperature and the exposure time according to time. Characterized in that it is given as a value of.

Here, the fire damaged concrete structure sample is collected by coring the building member of the wall, column or slab of the fire damaged concrete structure, but the coring diameter is 30 mm or less, and the coring depth is at least 40 mm. It is characterized in that it is carried out to the position of the main reinforcement.

Here, the coring sample of fire damaged concrete structure is precisely cut at intervals of 10 mm, and the sample of fire damaged concrete structure collected by coring at one location is defined as one set of at least 4 cut fire damaged concrete structure samples. And, all analyzes are performed in units of one set, but the output can be derived by finding the value that corresponds to all or the most similar values with the standardized DB that has been established.

Here, in the one-set analysis, by cutting at least four samples coring from one location at 10 mm intervals and analyzing at least four, the diffusion rate of heat can be expressed as a function of time when comparing the result values for each section. .

On the other hand, as another means for achieving the above-described technical problem, the method for predicting the heat diffusion path of the deep learning-based fire damaged concrete structure according to the present invention includes: a) coring sampling and cutting a fire damaged concrete structure sample at a fire site. The step of doing; b) performing physical and chemical analysis on the fire damaged concrete structure sample; c) selecting a deep neural network model of a deep learning analysis tool; d) inputting the sample analysis result to the deep neural network model, and comparing and analyzing the final output with the standardized directory; e) predicting the fire damage temperature at a specific location for each damage depth of the fire damaged concrete structure according to the deep neural network model analysis of the deep learning analysis tool; f) calculating a fire temperature curve according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure according to the deep neural network model analysis of the deep learning analysis tool; And g) predicting the heat diffusion path of the fire damaged concrete structure according to the predicted fire damage temperature and the calculated fire temperature curve, and tracking and reproducing the movement path of the flame from the fire site of the fire site as a function of time. However, in response to the prediction of the fire damage temperature at the time of the fire for the unknown fire damaged concrete structure sample collected at the fire site in step g), the heat diffusion path over time is identified.

According to the present invention, the heat diffusion path of the fire damaged concrete structure is accurately predicted by predicting the fire damage temperature by damage depth of the fire damaged concrete structure and calculating the fire temperature curve from the fire damaged concrete structure sample collected at the fire site. The flame movement path can be easily traced and reproduced.

According to the present invention, coring samples of fire damaged concrete structures are directly collected from the fire site, and thermogravimetric analysis (TGA), Fourier transform infrared spectroscopy (FTIR) analysis, and other chemical analysis are performed in parallel, so that high temperature through chemical and physical experiments. Correlation with exposure temperature can be accurately identified.

According to the present invention, by applying a deep neural network (DNN) model using a deep learning analysis tool, it is possible to increase a predictive rate of a fire damage temperature for a fire damaged concrete structure.

1 is a diagram illustrating that the compressive strength of a concrete structure decreases with an increase in temperature during a fire.
2 is a block diagram of a system for predicting the remaining life of a fire damaged concrete structure according to the prior art.
3 is an operation flow diagram of a method for predicting the remaining life of a fire damaged concrete structure according to the prior art.
4 is a block diagram of a system for predicting a heat diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention.
5 is a diagram showing the results of thermogravimetric analysis (TGA) for establishing a standardized directory in the deep learning-based heat diffusion path prediction system of a fire damaged concrete structure according to an embodiment of the present invention.
6 is a diagram showing that the overall mass reduction rate decreases as the fire damage temperature increases according to the results of thermogravimetric analysis (TGA).
7 is a diagram showing a correlation between standardized directory construction data and measured temperature.
8 is a Fourier Transform Infrared Spectroscopy (FTIR) analysis result for establishing a standardized directory in a deep learning-based thermal diffusion path prediction system of a fire damaged concrete structure according to an embodiment of the present invention. It is a diagram showing that the maximum value of (peak) is different or the area of the corresponding part is different.
9 is a diagram schematically illustrating a deep neural network (DNN) model applied to a system for predicting a thermal diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention.
10 is a diagram illustrating selection of a DNN model with high predictive power by searching for an applicable deep neural network (DNN) algorithm.
11 is a diagram illustrating a first deep neural network (DNN) model analysis in a system for predicting a thermal diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention.
12 is a diagram illustrating a second deep neural network (DNN) model analysis in a system for predicting a thermal diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention.
13 is an operation flow diagram of a method for predicting a heat diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention.
14 is a detailed operation flow diagram of a method for predicting a heat diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a part "includes" a certain component, it means that other components may be further included rather than excluding other components unless specifically stated to the contrary. In addition, terms such as "... unit" described in the specification mean a unit that processes at least one function or operation, which may be implemented by hardware or software, or a combination of hardware and software.

Hereinafter, a system for predicting a heat diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention will be described with reference to FIGS. 4 to 12, and an embodiment of the present invention with reference to FIGS. 13 and 14 A deep learning-based method for predicting the heat diffusion path of fire damaged concrete structures according to the following will be described.

[Deep learning-based heat diffusion path prediction system for fire damaged concrete structures (100)]

4 is a block diagram of a system for predicting a heat diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention.

4, a deep learning-based thermal diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention includes a fire damaged concrete structure sample 110, a sample analysis unit 120, and a standardized directory ( 130), a deep learning analysis tool 150, a fire damage temperature prediction unit 160, a fire temperature curve calculation unit 170, a heat diffusion path prediction unit 180, and a flame movement path tracking reproduction unit 190.

The fire damaged concrete structure sample 110 is collected by coring the building member of the wall, column or slab of the fire damaged concrete structure, but the coring diameter is 30 mm or less, and the coring depth is It is carried out to the position of the main reinforcing bar at least 40mm or more. At this time, it is preferable to collect the fire-damaged concrete structure sample 110 through coring with a minimum diameter so as not to damage the fire-damaged concrete structure.

In addition, the coring-collected fire-damaged concrete structure sample 110 is precisely cut at intervals of 10 mm, and the fire-damaged concrete structure sample 110 coring-collected at one location is at least four cut fire-damaged concrete structures. The sample 110 is defined as one set, and all analyzes are performed in units of one set, but an output may be derived by searching for a value that matches or is the most similar to all corresponding values. Accordingly, in the analysis of one set unit, the sample coring at one location is cut at 10 mm intervals and at least four are analyzed, so that the heat diffusion rate is a function of time when comparing the result values for each section. It can be expressed as

At this time, the coring-collected fire damaged concrete structure sample 110, as described later, was conducted through various chemical and physical experiments, such as chemical analysis and mass reduction rate, to determine the correlation with the high-temperature exposure temperature, and fire-damaged concrete at a specific temperature. After organizing how the chemical composition and physical characteristics inside the structure sample 110 change, it can be easily found while comparing specific values during the process of finding data. At this time, a collection of previously stored data is defined as the standardization directory 130.

The sample analysis unit 120 performs physical and chemical analysis on the fire-damaged concrete structure sample 110 that is coring-collected and cut at the fire site.

More specifically, the sample analysis unit 120 includes a thermogravimetric analysis (TGA) 121, a Fourier transform infrared spectroscopy (FTIR) analysis unit 122 and a chemical analysis unit 123 Includes.

The thermogravimetric analysis unit 121 of the sample analysis unit 120 measures the mass reduction rate of the fire damaged concrete structure sample 110. In addition, the Fourier transform infrared spectroscopy analysis unit 122 of the sample analysis unit 120 includes a maximum value of a peak appearing at a specific wavelength number and a corresponding portion according to the fire damage temperature of the fire damaged concrete structure sample 110. Analyze the area of In addition, the chemical analysis unit 123 of the sample analysis unit 120 performs chemical analysis in parallel with the chemical analysis equipment according to the analysis purpose.

Here, the higher the fire damage temperature in the thermogravimetric analysis unit 121, the smaller the total mass reduction rate appears, and the higher the overall fire damage temperature in the Fourier transform infrared spectroscopy unit 122, the maximum value of the peak and The area of the site becomes smaller.

In addition, the analysis result of the sample analysis unit 120 may be partially borrowed or a value may be changed, and it is preferable to undergo a noise removal process so that the analysis result is not affected by environmental factors.

The standardization directory 130 includes thermo-gravimetric analysis (TGA) results of the fire damaged concrete structure sample 110 exposed to a specific high temperature and the measured temperature results at the location of the fire damaged concrete structure sample. That is, in the deep learning-based heat diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention, the standardization directory 130 is the exposure temperature of the fire damaged concrete structure sample coring collected at the fire site. Thermogravimetric analysis (TGA) results showing the mass reduction rate of each, Fourier transform infrared spectroscopy (FTIR) analysis results showing the change in chemical composition by exposure temperature of fire damaged concrete structure samples coring from the fire site, and coring collection at the fire site It is constructed by including the results of other chemical and physical analysis of samples of damaged fire damaged concrete structures.

The deep learning analysis tool 150 applies a deep neural network (DNN) model to the analysis result of the sample analysis unit 120 for deep learning analysis, and then the standardized directory (Standard Directory). : 130) to derive the optimal analysis result. In other words, in the case of the deep learning-based thermal diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention, the correlation between the thermogravimetric analysis (TGA) result and the measured temperature through the deep neural network (DNN) analysis Analyze and identify the algorithm for it.

For example, a fire damaged concrete structure sample 110 for each damage depth coring-collected from a concrete structure at a fire site is included in the standardization directory 130 by thermogravimetric analysis (TGA), and the result is deep-learning (deep learning). -learning) By inputting into the analysis tool 150 and matching according to the DNN algorithm, the fire damage temperature value is predicted, and the fire temperature curve of the fire damaged concrete structure at the fire site is calculated according to the DNN algorithm. That is, the fire damage temperature of the fire damaged concrete structure can be accurately predicted, and the movement path of heat and flame from the fire site of the fire site can be expressed as a function of time. In addition, the results of chemical analysis using Fourier transform infrared spectroscopy (FTIR) analysis and other equipment are also included in the standardization directory 130 in the same way as thermogravimetric analysis (TGA), so that they are used as statistical data to increase the prediction rate during deep learning analysis. do.

Specifically, the deep learning analysis tool 150 includes a first deep neural network (DNN) model 151, a second deep neural network model 153, and a data comparison and analysis unit 152.

The first deep neural network (DNN) model 151 of the deep learning analysis tool 150 uses the result of the single analysis of the thermogravimetric analysis unit 121 as input data, and the fire damage temperature according to the damage depth inside the fire damaged concrete structure And derive the final output for calculating the fire temperature curve of the part corresponding to the heat source outside the fire damaged concrete structure.

When the second deep neural network model 153 of the deep learning analysis tool 150 is less than an appropriate level, the analysis result of the thermogravimetric analysis unit 121, the Fourier By using the analysis result of the converted infrared spectral analysis unit 122 and the analysis result of the chemical analysis unit 123 as input data, weighted and reanalyzed for clearly indicated features, and the depth of damage inside the fire-damaged concrete structure The final output for calculating the fire damage temperature and the fire temperature curve of the part corresponding to the heat source outside the fire damaged concrete structure is derived. Here, in the second deep neural network model 153, the feature may be a result value clearly displayed at a specific temperature.

The data comparison and analysis unit 152 of the deep learning analysis tool 150 stores the analysis result of the first deep neural network model 151 or the analysis result of the second deep neural network model 153 in the standardized directory 130 ) And compare.

The fire damage temperature prediction unit 160 predicts the fire damage temperature at a specific location for each damage depth of the fire damaged concrete structure according to the deep neural network model analysis of the deep learning analysis tool 150.

The fire temperature curve calculation unit 170 is absorbed by the outer surface of the concrete according to the size of the flame corresponding to the first heat source outside the fire damaged concrete structure according to the deep neural network model analysis of the deep learning analysis tool 150. Heat information is calculated as a fire temperature curve. Specifically, the fire temperature curve calculation unit 170 compares and analyzes the result value obtained from the outer surface of the fire damaged concrete structure and the result value obtained from the inner area, which is each data analyzed in units of one set, and is a function of time (T). The fire temperature curve is calculated according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure by comparing and analyzing the result of the fire damaged concrete structure sample 110 cored at the periphery, The fire temperature curve is given by the magnitude of the water heat temperature over time and the exposure time value at this time. For example, the X-axis represents the exposure time, and the Y-axis represents the hydrothermal temperature.

The heat diffusion path prediction unit 180 predicts the heat diffusion path according to the passage of time outside the fire damaged concrete structure according to the predicted fire damage temperature and the calculated fire temperature curve. Accordingly, the thermal diffusion path prediction unit 180 may determine the thermal diffusion path over time in response to the prediction of the fire damage temperature at the time of the fire for the unknown fire damaged concrete structure sample 110 collected at the fire site. .

The flame movement path tracking reproducing unit 190 may track and reproduce the movement path of the flame from the fire location of the fire site according to the predicted fire damage temperature, the calculated fire temperature curve, and the heat diffusion path prediction result.

In the end, the deep learning-based thermal diffusion path prediction system of a fire damaged concrete structure according to an embodiment of the present invention includes: 1) Thermogravimetric analysis showing the mass reduction rate by exposure temperature of the fire damaged concrete structure sample coring from the fire site ( TGA) results, 2) Fourier transform infrared spectroscopy (FTIR) analysis results showing the chemical composition change by exposure temperature of fire damaged concrete structure samples coring collected at the fire site, and 3) fire damaged concrete structures coring collected at the fire site Establish a standard directory by including the results of other analysis of the sample.

Accordingly, the deep learning-based heat diffusion path prediction system of a fire damaged concrete structure according to an embodiment of the present invention is based on deep learning after physical and chemical analysis of samples of fire damaged concrete structures collected at the fire site. By predicting the fire damage temperature by damage depth of the fire damaged concrete structure and calculating the fire temperature curve, the heat diffusion path of the fire damaged concrete structure can be accurately predicted and the flame movement path can be easily traced and reproduced.

On the other hand, FIG. 5 is a diagram showing the results of thermogravimetric analysis (TGA) for establishing a standardized directory in the deep learning-based thermal diffusion path prediction system of a fire damaged concrete structure according to an embodiment of the present invention, and FIG. 6 is a thermogravimetric analysis ( TGA) is a diagram showing the decrease in the total mass reduction rate as the fire damage temperature increases according to the result, and FIG. 7 is a diagram showing the correlation between the standardized directory construction data and the measured temperature.

5, in the case of the deep learning-based heat diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention, a standardization directory 130 is built through thermogravimetric analysis (TGA).

Specifically, in the conventional thermogravimetric analysis (TGA), as shown by the red line in FIG. 5, it can be confirmed that the mass decrease occurs rapidly at a specific location, that is, a specific temperature according to DTA (Differential Thermo Analysis: DTA). , It is possible to analyze which components are contained in the fire damaged concrete structure sample 110 and how much by checking that this corresponds to the result that appears as a specific chemical component is decomposed. In addition, as shown by the black line of FIG. 5, it can be seen how the mass is overall reduced according to the TG (Thermo Gravimetric: TG).

In the case of the deep learning-based heat diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention, the result value of the black line (TG) of FIG. 5 is used, and the concrete structure exposed to a specific fire damage temperature The sample 110 exhibits a characteristic that mass reduction does not occur in a certain section because mass reduction has already occurred, and accordingly, it can be seen that the overall mass reduction rate becomes smaller as the fire damage temperature increases.

In addition, as shown in Figure 6, according to the result of thermogravimetric analysis (TGA), as the fire damage temperature increases, the total mass reduction rate decreases, and as shown in Equation 1 below, the relative weight of the thermogravimetric analysis (TGA) Can be represented by ).

는 특정 도에서의 질량을 나타내고,

는 시험이 종료된 이후 최종 잔류된 샘플의 질량을 나타낸다.here,

Denotes the mass in a specific degree,

Denotes the mass of the final remaining sample after the end of the test.

In FIG. 6, the final mass reduction rate corresponding to a value corresponding to one line is expressed as a'dot', and then the correlation with temperature is shown. It shows the correlation results, and the collection of these results is defined as a standardized directory.

Meanwhile, FIG. 8 is a Fourier Transform Infrared Spectroscopy (FTIR) analysis result for establishing a standardized directory in the deep learning-based thermal diffusion path prediction system of a fire damaged concrete structure according to an embodiment of the present invention. It is a diagram showing that the maximum value of the peak that appears is different or that the area of the corresponding area is different.

In the case of the deep learning-based heat diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention, a standardized directory may be established through Fourier Transform Infrared (FTIR) analysis.

Specifically, Fourier Transform Infrared Spectroscopy (FTIR) analysis is a comparison of cement and blast furnace slag, respectively, and conventionally, after defining a material that exhibits a peak at a specific wavelength number, the value is large and small. It is possible to distinguish mixed substances.

In the case of the deep learning-based heat diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention, as shown in FIG. 9, the maximum of the peak appearing at a specific wavelength number according to the fire damage temperature The value changes or the area of the corresponding part appears differently. At this time, it can be seen that the higher the fire damage temperature as a whole, the smaller the maximum value of the peak and the area of the area.

For example, in the case of the deep learning-based heat diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention, as shown in FIG. 8, 6 peaks appearing at a specific wavelength number After determining the degree, the values at that time are sequentially listed. At this time, a graph is defined as a standardized directory that is collected by expressing the correlation with temperature after expressing a graph with about 6 numbers. For example, the main peak values of a 600 degree exposure sample are p1 = 1.54, p2 = 1.20, p3 = 1.02, p4 = 1.23, p5 = 1.64, and p6 = 1.52, respectively.

On the other hand, FIG. 9 is a diagram schematically showing a deep neural network (DNN) model applied to a system for predicting a thermal diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention, and FIG. 10 is a diagram schematically showing an applicable deep neural network (DNN). ) A diagram showing selecting a DNN model with high predictive power by searching an algorithm.

In the case of the deep learning-based heat diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention, a deep neural network (DNN) analysis tool is used, and the DNN model construction is specifically performed. Explained as follows.

First, a deep neural network (DNN) is an artificial neural network (ANN) including multiple hidden layers between an input layer and an output layer, as shown in FIG. 9. Artificial Neural Network), such a deep neural network (DNN) corresponds to one of machine learning, and it can learn various nonlinear relationships by including multiple hidden layers.

In general, this deep neural network (DNN) analysis derives a functional equation that predicts certain characteristics based on learning data. For example, we can derive the k value from Hooke's law F = kx.

The above-described standardization directory was constructed by selecting only specified features from TGA and FTIR analysis results for only deep neural network (DNN) analysis. In other words, the standardization directory is a directory organized by selecting a specific section or a specific value as a representative value and then numerically organizing it. For example, only specific values are extracted from the standardized directory and organized.

In the case of the deep learning-based heat diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention, after analyzing the results by searching for an applicable deep neural network (DNN) algorithm, a deep neural network (DNN) with high predictive power ) The optimized deep neural network (DNN) model is completed by selecting and optimizing the model.

The completed DNN model is implemented in the form of software, and as shown by the broken line in FIG. 10, the most similar value is found by matching the fire damaged concrete structure sample 110 collected at the site and the standardized directory sample.

On the other hand, FIG. 11 is a diagram showing an analysis of a first deep neural network (DNN) model in a system for predicting a thermal diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention, and FIG. 12 is a diagram showing an analysis of a first deep neural network (DNN) model according to an embodiment of the present invention. A diagram showing the analysis of a second deep neural network (DNN) model in a deep learning-based heat diffusion path prediction system for fire damaged concrete structures.

More specifically, in the deep learning-based heat diffusion path prediction system 100 of a fire-damaged concrete structure according to an embodiment of the present invention, the artificial neural network (DNN) model includes two types of analysis of the first DNN model and the second DNN model. The method is used in combination, but if the result of the first DNN model analysis is higher than the appropriate level, the analysis is terminated. However, when the correlation value is less than an appropriate level, the accuracy of the final result value can be improved by further analyzing the second DNN model.

First, in the case of the first DNN model analysis, as shown in FIG. 11, by using the above-described thermogravimetric analysis (TGA) alone analysis result as input data, the final result is the fire damage temperature and fire damage inside the concrete structure. Each fire temperature curve corresponding to the heat source outside the concrete structure is calculated.

In addition, when the correlation of the first DNN model analysis result is low, re-analysis is performed. In the case of the second DNN model analysis, as shown in Fig. 12a), the thermogravimetric analysis (TGA) result and Fourier transform infrared spectroscopy (FTIR) ) This is a method of re-analyzing by placing weights on features that clearly appear in the analysis result and other analysis result values. In other words, the results are analyzed by placing weights on the results that are clearly displayed at a specific temperature.

In addition, as shown in b) of FIG. 12, a weight is given to a result value having a high prediction rate for a specific temperature section for each analysis device. For example, since the prediction rate of the analysis results at temperatures below 100 degrees FTIR is high, weights are given to the results of the FTIR analysis at temperatures below 100 degrees, and weights are given to the results of the TGA at temperatures above 100 degrees. You can derive the output.

At this time, the output is calculated by calculating the fire damage temperature for each depth inside the concrete structure and the fire temperature curve of the part corresponding to the outer heat source of the concrete structure, as in the above-described first DNN model analysis, so that the prediction rates of the heat diffusion path and the flame movement path are calculated. It can be greatly increased.

In the end, the result of the first DNN model analysis and the second DNN model analysis is the fire damage temperature inside the fire damaged concrete structure, that is, after analyzing the fire damaged concrete structure sample 110 cored at the fire site, After entering the DNN model as the input data into the software to which the DNN model is applied, and passing through the matrix weighted by the matching algorithm, the fire damage temperature prediction and fire at the location of the finally cored fire damaged concrete structure sample 110 The fire temperature curve outside the damaged concrete structure can be calculated. This can be used to evaluate the damage scale and damage degree of the fire damaged concrete structure, and to present a recovery plan.

Meanwhile, in the deep learning-based heat diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention, a method of predicting a heat diffusion path and tracking and reproducing a flame movement path is as follows.

The existing analysis method judges the heat source as an approximate experience value, determines the value first, and then tracks the movement path of the flame. In this case, the prediction rate is less than 30% and a simulation analysis tool that can be applied in various ways to all fire sites. Does not exist. In addition, no related technology has been reported not only in Korea but also in foreign countries as a system for predicting heat damage to concrete. On the other hand, the deep learning-based heat diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention analyzes the concrete that only has temperature damage information at the fire site to achieve a high back calculation. With it, you can track the heat source.

The deep learning-based thermal diffusion path prediction system 100 of a fire damaged concrete structure according to an embodiment of the present invention does not use a specific value existing in the existing literature because it directly collects samples from fire damaged concrete structures at all fire sites. Does not. In addition, when implementing the simulation, it does not proceed using external values, but uses a back calculation method starting from the inner sample and moving to the outside. For example, the prediction rate may increase as various environmental factors disappear as the process proceeds based on the results of analysis in the molecular and atomic units of the cement paste.

In particular, when analyzing the DNN model, the data constructed in the standardization directory 130 are not matched one by one, but the samples collected by coring at the corresponding location are grouped into at least four and matched in units of one set. All of these match or look for the most similar values to derive the output.

Since this one-set analysis analyzes at least four or more by cutting the cored concrete structure samples 110 at 10mm intervals at one location, when comparing the result values for each section, the diffusion of heat. By expressing the speed as a function of time, the heat diffusion path can be accurately predicted. In addition, the maximum temperature and exposure time can be calculated according to the size of the heat source outside the concrete structure corresponding to the source of such heat damage, and a fire temperature curve can be derived through this.

Meanwhile, the above-described standardization directory 130 also stores result values in units of one set, and at this time, since the fire temperature curve corresponding to the heat source is also stored, By matching the data with the analysis result of the fire damaged concrete structure sample 110 collected by coring at the fire site, it is possible to easily calculate the fire damage temperature curve.

This fire temperature curve calculation can be used to accurately predict the heat diffusion path outside the fire damaged concrete structure, and can be used to trace and reproduce the flame movement path, thereby significantly increasing the prediction rate of the final result.

[Deep learning-based method for predicting the heat diffusion path of fire damaged concrete structures]

13 is an operation flow diagram of a method for predicting a heat diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention.

Referring to FIG. 13, a method for predicting a heat diffusion path of a fire damaged concrete structure based on deep learning according to an embodiment of the present invention includes, first, coring collecting and cutting a fire damaged concrete structure sample 110 at a fire site ( S110).

Next, physical and chemical analysis of the fire damaged concrete structure sample 110 is performed (S120). Specifically, a thermogravimetric analysis (TGA) measuring the mass reduction rate of the fire damaged concrete structure sample 110 is performed, and a specific wavelength according to the fire damage temperature of the fire damaged concrete structure sample 110 Fourier Transform Infrared (FTIR) analysis is performed to analyze the maximum value of the peak appearing in the number and the area of the corresponding region, and chemical analysis is performed in parallel with chemical analysis equipment according to the analysis purpose. .

Next, deep neural network (DNN) models 151 and 152 of the deep learning analysis tool 150 are selected (S130).

Next, the sample analysis result is input to the deep neural network models 151 and 152, and the final output is compared and analyzed with the standardized directory 130 (S140). At this time, the standardization directory 130 is a result of thermogravimetric analysis (TGA) showing the mass reduction rate by exposure temperature of the fire damaged concrete structure sample coring collected at the fire site, and the result of the fire damaged concrete structure sample coring collected at the fire site. It is constructed by including the results of Fourier Transform Infrared Spectroscopy (FTIR) analysis showing the change of chemical composition by exposure temperature and other analysis results of samples of fire damaged concrete structures collected by coring at the fire site.

Specifically, calculating the fire damage temperature by the damage depth inside the fire damaged concrete structure and the fire temperature curve of the part corresponding to the heat source outside the fire damaged concrete structure by using the result of the single analysis by the thermogravimetric analysis unit 121 as input data. A first deep neural network (DNN) model 151 is selected to derive a final output for the first deep neural network (DNN) model 151, and then, when the correlation value of the first deep neural network (DNN) model 151 is less than an appropriate level, the column Using the analysis result of the weight analysis unit 121, the analysis result of the Fourier transform infrared spectral analysis unit 122 and the analysis result of the chemical analysis unit 123 as input data, a weight is applied to a feature that is clearly indicated. A second deep neural network model (153) to derive the final output for calculating the fire damage temperature by damage depth inside the fire damaged concrete structure and the fire temperature curve of the part corresponding to the heat source outside the fire damaged concrete structure by reanalyzing it. ) Is selected, and then, the analysis result of the first deep neural network model 151 or the analysis result of the second deep neural network model 153 is compared with the established standardization directory 130.

Next, according to the deep neural network model analysis of the deep learning analysis tool 150, the fire damage temperature at a specific location for each damage depth of the fire damaged concrete structure is predicted (S150).

Next, according to the deep neural network model analysis of the deep learning analysis tool 150, a fire temperature curve is calculated according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure (S160). Specifically, the result obtained from the outer surface of the fire damaged concrete structure, which is each data analyzed in units of one set, and the result obtained from the inner area are compared and analyzed, and expressed as a function of time (T), and the fire coring from the periphery By comparing and analyzing the results of the damaged concrete structure sample 110, the fire temperature curve is calculated according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure, but the fire temperature curve is the number of heats over time. It is given as a value of the magnitude of the temperature and the exposure time.

Next, the heat diffusion path of the fire damaged concrete structure is predicted according to the predicted fire damage temperature and the calculated fire temperature curve, and the movement path of the flame from the fire site of the fire site is traced and reproduced as a function of time (S170). .

Accordingly, in response to the prediction of the fire damage temperature at the time of the fire for the unknown fire damaged concrete structure sample 110 collected at the fire site, the heat diffusion path over time can be identified.

Specifically, FIG. 14 is a detailed operation flow diagram of a method for predicting a heat diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention.

Referring to FIG. 14, the method for predicting the heat diffusion path of a deep learning-based fire damaged concrete structure according to an embodiment of the present invention includes, first, a fire site survey on various building members such as walls, columns, and slabs of a fire damaged concrete structure. Is carried out (S201).

Next, a fire damaged concrete structure sample 110 is collected by performing a coring in at least five positions in each of the building members (S202). At this time, it is preferable that the coring diameter is 30 mm or less at each position, and the coring depth is at least 40 mm or more to the main reinforcing bar position.

Next, the fire damaged concrete structure sample 110 collected through coring is subjected to precise analysis within 24 hours after being placed in the sample bottle (S203).

Next, as a pretreatment step for precise analysis, the coring collection fire damaged concrete structure sample 110 is precisely cut at 10 mm intervals (S204). At this time, the fire-damaged concrete structure sample 110 collected by coring at one location defines at least four or more cut fire-damaged concrete structure samples 110 as one set, and all analyzes are performed in units of one set. Conduct. In addition, all cut samples are prepared as samples for analysis in an optimal state through additional cutting or grinding or other methods according to the requirements of the sample analysis equipment, and then sealed with a vacuum pack to complete the preparation.

Next, physical and chemical analysis is performed on the sample for analysis within 30 minutes after taking the sealed sample for analysis out of the vacuum pack (S205). Specifically, the analysis of the fire damaged concrete structure sample 110 mainly uses Thermo-Gravimetric Analysis (TGA) and Fourier Transform Infrared (FTIR) analysis, and various chemistry according to the purpose of analysis. Chemical analysis can be performed in parallel with the analysis equipment, and at this time, it is preferable to borrow only a part of the experimental result or change the value, and to undergo a noise removal process so that the experimental result is not affected by various environmental influence factors.

Next, deep neural network (DNN) models 151 and 152 of the deep learning analysis tool 150 are selected (S206).

Next, data in a standardized directory (or standardized DB) in which each sample analysis result is input to the deep neural network models (151, 152), and the final output is established by applying the matching algorithm of the deep neural network (DNN) model. Compare and analyze with (S207).

Next, according to the deep neural network model analysis, the fire damage temperature at the corresponding location of the fire damaged concrete structure may be predicted (S208).

Next, each data analyzed in units of one set, i.e., the result obtained from the outer surface of the fire damaged concrete structure and the result obtained from the inner area, are compared and analyzed and expressed as a function of time (T). By comparing and analyzing the results of the fire damaged concrete structure sample 110, the fire temperature curve is calculated according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure (S209). In this case, the fire temperature curve is given by the magnitude of the water heat temperature and the value of the exposure time over time.

Next, according to the fire damage temperature and fire temperature curve, the heat diffusion path and the flame movement path outside the fire damaged concrete structure are tracked and reproduced as a function of time (S210).

After all, according to an embodiment of the present invention, the heat diffusion of the fire damaged concrete structure by predicting the fire damage temperature by damage depth of the fire damaged concrete structure and calculating the fire temperature curve from the fire damaged concrete structure sample collected at the fire site The path can be accurately predicted and the flame moving path can be easily traced and reproduced.In addition, coring samples of fire damaged concrete structures from the fire site are directly collected, thermogravimetric analysis (TGA), Fourier transform infrared spectroscopy (FTIR) analysis, and By performing other chemical analysis in parallel, it is possible to accurately identify the correlation with high temperature exposure temperature through chemical and physical experiments, and by applying a deep neural network (DNN) model using a deep learning analysis tool. It is possible to increase the predicted rate of fire damage temperature for damaged concrete structures.

The above description of the present invention is for illustrative purposes only, and those of ordinary skill in the art to which the present invention pertains will be able to understand that other specific forms can be easily modified without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative and non-limiting in all respects. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as being distributed may also be implemented in a combined form.

The scope of the present invention is indicated by the claims to be described later rather than the detailed description, and all changes or modified forms derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present invention. do.

110: Fire damaged concrete structure sample
120: sample analysis unit
130: Standard Directory
150: Deep Learning Analysis Tool
160: Fire damage temperature prediction unit
170: fire temperature curve calculation unit
180: heat diffusion path prediction unit
190: flame movement path tracking reproduction unit
121: thermogravimetric analysis unit (TGA)
122: Fourier transform infrared spectral analysis unit (FTIR)
123: Chemical Analysis Department
151: first deep neural network (DNN) model
152: data comparison and analysis unit
153: the second deep neural network model

Claims (20)

A sample analysis unit 120 for performing physical and chemical analysis of the fire damaged concrete structure sample 110 which is coring collected at the fire site and cut and processed;
For deep learning analysis, after applying a deep neural network (DNN) model to the analysis result of the sample analysis unit 120, the optimum A deep learning analysis tool 150 for deriving an analysis result;
A fire damage temperature prediction unit 160 for predicting a fire damage temperature at a specific location for each damage depth of the fire damaged concrete structure according to the deep neural network model analysis of the deep learning analysis tool 150;
A fire temperature curve calculation unit 170 for calculating a fire temperature curve according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure according to the deep neural network model analysis of the deep learning analysis tool 150; And
Including a heat diffusion path prediction unit 180 for predicting a heat diffusion path according to the passage of time outside the fire damaged concrete structure according to the predicted fire damage temperature and the calculated fire temperature curve,
The heat diffusion path prediction unit 180 is a dip, characterized in that to identify the heat diffusion path over time in response to the prediction of the fire damage temperature at the time of the fire for the unknown fire damaged concrete structure sample 110 collected at the fire site. A running-based fire-damaged concrete structure's thermal diffusion path prediction system. The method of claim 1,
Deep learning further includes a heat diffusion and flame movement path tracking reproduction unit 190 for tracking and reproducing the heat movement path as a function of time from the fire place of the fire site according to the predicted fire damage temperature and the calculated fire temperature curve -Based fire damage concrete structure heat diffusion path prediction system. The method of claim 1, wherein the sample analysis unit 120,
A thermogravimetric analysis unit 121 performing a thermogravimetric analysis (TGA) measuring the mass reduction rate of the fire damaged concrete structure sample 110;
Fourier Transform Infrared (FTIR) analysis unit 122 that analyzes the maximum value of the peak appearing at a specific wavelength number and the area of the corresponding part according to the fire damage temperature of the fire damaged concrete structure sample 110 ); And
A deep learning-based heat diffusion path prediction system for fire damaged concrete structures, including a chemical analysis unit 123 that performs chemical analysis in parallel with chemical analysis equipment according to the analysis purpose. The method of claim 3,
The higher the fire damage temperature in the thermogravimetric analysis unit 121, the smaller the total mass reduction rate appears, and the higher the overall fire damage temperature in the Fourier transform infrared spectroscopy unit 122, the maximum value of the peak and the portion thereof. Deep learning-based heat diffusion path prediction system of a fire damaged concrete structure, characterized in that the area of is reduced. The method of claim 3,
Deep learning-based fire damage concrete, characterized in that the analysis result of the sample analysis unit 120 may be partially borrowed or the value may be changed, and a noise removal process is performed so that the analysis result is not affected by environmental influence factors. Structure's thermal diffusion path prediction system. The method of claim 3,
The standardization directory 130 is the result of thermogravimetric analysis (TGA) showing the mass reduction rate by exposure temperature of the fire damaged concrete structure samples coring collected at the fire site, the exposure temperature of the fire damaged concrete structure samples coring collected at the fire site. A deep learning-based fire-damaged concrete structure, characterized in that it was constructed by including the results of Fourier Transform Infrared Spectroscopy (FTIR) analysis showing the change in chemical composition of each and other analysis results of fire-damaged concrete structure samples coring from the fire site. Thermal diffusion path prediction system. The method of claim 3, wherein the deep learning analysis tool 150,
Final output for calculating the fire damage temperature by the damage depth inside the fire damaged concrete structure and the fire temperature curve of the part corresponding to the heat source outside the fire damaged concrete structure using the result of the single analysis of the thermogravimetric analysis unit 121 as input data a first deep neural network (DNN) model 151 to derive (output);
When the correlation value of the first deep neural network (DNN) model 151 is below an appropriate level, the analysis result of the thermogravimetric analysis unit 121, the analysis result of the Fourier transform infrared spectral analysis unit 122, and the chemical analysis unit Using the analysis result in (123) as input data, weighted and reanalyzed for clearly indicated features, the fire damage temperature and the heat source outside the fire damaged concrete structure by damage depth were re-analyzed. A second deep neural network model 153 for deriving a final output for calculating a fire temperature curve; And
Deep learning including a data comparison and analysis unit 152 that compares the analysis result of the first deep neural network model 151 or the analysis result of the second deep neural network model 153 with the established standardization directory 130 A system for predicting the heat diffusion path of the base fire damaged concrete structure. The method of claim 7,
In the second deep neural network model (153), the characteristic is a result value clearly displayed at a specific temperature. A system for predicting a heat diffusion path of a fire damaged concrete structure based on deep learning. The method of claim 1,
The fire temperature curve calculation unit 170 compares and analyzes the result value obtained from the outer surface of the fire damaged concrete structure and the result value obtained from the inner area, which is each data analyzed in units of one set, and expresses it as a function of time (T), By comparing and analyzing the result of the fire damaged concrete structure sample 110 cored at the periphery, the fire temperature curve is calculated according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure, but the fire temperature A deep learning-based heat diffusion path prediction system for fire damaged concrete structures, characterized in that the curve is given by the size of the water heat temperature and the value of the exposure time according to time. The method of claim 1,
The fire damaged concrete structure sample 110 is collected by coring the building member of the wall, column or slab of the fire damaged concrete structure, but the coring diameter is 30 mm or less, and the coring depth The deep learning-based heat diffusion path prediction system of a fire damaged concrete structure, characterized in that to conduct at least 40㎜ or more to the position of the main reinforcement. The method of claim 10,
The coring-collected fire-damaged concrete structure sample 110 is precisely cut at intervals of 10 mm, and the fire-damaged concrete structure sample 110 coring-collected at one location includes at least four fire-damaged concrete structure samples ( 110) is defined as one set, and all analyzes are performed in units of one set, and the corresponding value matches all or the most similar values to the standardized DB and derives the output. Deep learning-based heat diffusion path prediction system for fire damaged concrete structures. The method of claim 11,
The one-set analysis is performed by cutting at least four samples by cutting at 10 mm intervals at one location, so that the heat diffusion rate can be expressed as a function of time when comparing the result values for each section. Deep learning-based heat diffusion path prediction system of fire damaged concrete structures, characterized in that there is. a) coring sampling and cutting the fire damaged concrete structure sample 110 at the fire site;
b) performing physical and chemical analysis on the fire damaged concrete structure sample (110);
c) selecting a deep neural network (DNN) model 151, 152 of the deep learning analysis tool 150;
d) inputting the sample analysis result to the deep neural network models (151, 152), and comparing and analyzing the final output with the standardized directory (130);
e) predicting the fire damage temperature at a specific location for each damage depth of the fire damaged concrete structure according to the deep neural network model analysis of the deep learning analysis tool 150;
f) calculating a fire temperature curve according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure according to the deep neural network model analysis of the deep learning analysis tool 150; And
g) predicting the heat diffusion path of the fire damaged concrete structure according to the predicted fire damage temperature and the calculated fire temperature curve, and tracking and reproducing the movement path of the flame from the fire site of the fire site as a function of time. ,
Deep learning-based fire, characterized in that the heat diffusion path over time is identified in response to the prediction of the fire damage temperature at the time of the fire for the unknown fire damaged concrete structure sample 110 collected at the fire site in step g). Method for predicting the thermal diffusion path of damaged concrete structures. The method of claim 13,
The fire damaged concrete structure sample 110 of step a) is collected by coring the building member of the wall, column or slab of the fire damaged concrete structure, and the coring diameter is 30 mm or less. , Deep learning-based method for predicting heat diffusion paths of fire damaged concrete structures, characterized in that the coring depth is carried out to the position of the main reinforcement at least 40 mm or more. The method of claim 14,
The coring-collected fire-damaged concrete structure sample 110 is precisely cut at intervals of 10 mm, and the fire-damaged concrete structure sample 110 coring-collected at one location includes at least four fire-damaged concrete structure samples ( 110) is defined as one set, and all analyzes are performed in units of one set, and the corresponding value matches all or the most similar values to the standardized DB and derives the output. A deep learning-based method for predicting the heat diffusion path of fire damaged concrete structures. The method of claim 15,
The one set unit analysis is that the sample coring at one location is cut at 10 mm intervals and at least four are analyzed to represent the heat diffusion rate as a function of time when comparing the result values for each section. A deep learning-based method for predicting heat diffusion paths of fire damaged concrete structures. The method of claim 13, wherein step b),
b-1) performing Thermo-Gravimetric Analysis (TGA) for measuring the mass reduction rate of the fire damaged concrete structure sample 110;
b-2) Fourier Transform Infrared (FTIR) analyzing the maximum value of the peak appearing at a specific wavelength number and the area of the corresponding part according to the fire damage temperature of the fire damaged concrete structure sample 110 Analysis step; And
b-3) A deep learning-based method for predicting heat diffusion paths of fire damaged concrete structures, including the step of conducting chemical analysis in parallel with chemical analysis equipment according to the purpose of analysis. The method of claim 13,
The standardization directory 130 of step d) is a thermogravimetric analysis (TGA) showing the mass reduction rate by exposure temperature of the fire damaged concrete structure sample coring collected at the fire site, and the fire damaged concrete structure coring collected at the fire site. A deep learning-based fire characterized by including the results of Fourier Transform Infrared Spectroscopy (FTIR) analysis showing the change of chemical composition by exposure temperature of the sample and other analysis results of fire damaged concrete structure samples coring from the fire site. Method for predicting the thermal diffusion path of damaged concrete structures. delete The method of claim 13,
The results obtained from the outer surface of the fire damaged concrete structure, which are each data analyzed in units of one set in the step f) above, are compared and analyzed with the results obtained from the inner area, and are expressed as a function of time (T), and cored from the periphery. By comparing and analyzing the results of the fire damaged concrete structure sample 110, the fire temperature curve is calculated according to the size of the flame corresponding to the initial heat source outside the fire damaged concrete structure, but the fire temperature curve is A deep learning-based method for predicting the heat diffusion path of a fire damaged concrete structure, characterized in that given by the size of the water heat temperature and the value of the exposure time.

Images