Search

KR-102961271-B1 - METHOD AND SYSTEM FOR PREDICTING STRUCTURE STRENGTH USING ARTIFICIAL NEURAL NETWORK

KR102961271B1KR 102961271 B1KR102961271 B1KR 102961271B1KR-102961271-B1

Abstract

A method for predicting structural strength using an artificial neural network according to the present invention comprises: a step of generating training data; a step of training an artificial neural network; and a step of predicting structural strength. The above-mentioned learning data generation step comprises: a step of generating first learning data by striking a structure of first intensity with a first intensity and measuring a first response with a vibration sensor; a step of generating second learning data by striking a structure of second intensity different from the first intensity with the first intensity and measuring a second response with a vibration sensor; and a step of generating Nth learning data by striking a structure of Nth intensity different from the first and second intensity with the first intensity and measuring an Nth response with a vibration sensor. The above first, second, and Nth responses include vibration-related information resulting from a first-stroke intensity strike of the structure of the above first, second, and Nth strengths, and The above first, second, and Nth learning data includes time history data containing information related to changes in vibration over time during a first intensity strike of the structure of the first, second, and Nth intensity.

Inventors

  • 진원종
  • 김영진
  • 이규완

Assignees

  • 한국건설기술연구원
  • (주)카이센테크

Dates

Publication Date
20260511
Application Date
20221216

Claims (17)

  1. As a method for predicting the strength of a target structure, It includes a training data generation step; an artificial neural network training step; and a structural strength prediction step, The above training data generation step is, The method comprises: a step of generating first learning data by striking a structure of first intensity with a first intensity and measuring a first response with a vibration sensor; a step of generating second learning data by striking a structure of second intensity different from the first intensity with the first intensity and measuring a second response with a vibration sensor; and a step of generating Nth learning data by striking a structure of Nth intensity different from the first and second intensity with the first intensity and measuring an Nth response with a vibration sensor. The above first, second, and Nth responses are, Includes vibration-related information resulting from a first-stroke intensity strike of the above-mentioned first, second, and N-stroke strength structures, and The above first, second, and Nth training data are, It includes time history data containing information related to changes in vibration over time during a first intensity strike of the structure of the first, second, and N strengths, and The above artificial neural network learning step is, The method includes the step of training an artificial neural network using the above-mentioned first, second, and Nth training data; The above-mentioned structure of first and second N strengths is, It is a concrete specimen, installed on a support via a tightening screw, and fixed in a state of being suspended in the air by the tightening force of the tightening screw and the reaction force of a vibration sensor according to the tightening force, and The above training data generation step is, While the above structure is levitating in the air, a strike of the first intensity is performed using a striking device, and The above striking device is, The main body of the above striking device is installed so as to be fixed to the support, and It is configured to perform a strike by causing movement of the striking rod through solenoid action, and The above vibration sensor is, It is installed on the opposite side of the main body of the above-mentioned striking device, and The opposing surface of the above main body is, A surface facing the structure when the above-mentioned striking device is installed on the above-mentioned support, A method for predicting structural strength using an artificial neural network characterized by
  2. In Article 1, The above change in vibration is, A method for predicting structural strength using an artificial neural network, characterized by including at least one of a change in amplitude and a change in frequency over time.
  3. In Article 1, The above first learning data further includes first intensity information, and the first intensity information includes an intensity reference value that matches the first response, and The second learning data further includes second intensity information, and the second intensity information includes an intensity reference value that matches the second response, and A method for predicting structural strength using an artificial neural network, characterized in that the above Nth training data further includes Nth strength information, and the above Nth strength information includes a strength reference value that matches the above Nth response.
  4. In Paragraph 3, Each strength reference value of the above first, second, and Nth strength information is, A method for predicting structural strength using an artificial neural network, characterized in that the strength value is measured immediately after striking the structure with the first force, or the strength value is measured immediately before striking the structure with the first force.
  5. In Paragraph 3, Each strength reference value of the above first, second, and Nth strength information is, A method for predicting structural strength using an artificial neural network, characterized by being set to a single strength value or to a strength value within a predetermined strength range.
  6. delete
  7. delete
  8. In Article 1, The above structural strength prediction step is, A first step of striking the above target structure with the above first force; A second step of obtaining vibration information by measuring the response to the impact with a vibration sensor when striking the target structure with the first force; and A method for predicting structural strength using an artificial neural network, characterized by including a third step of inputting the above vibration information into the artificial neural network to output a predicted strength.
  9. In Article 8, The above vibration information includes vibration-related information resulting from a first-stroke of the target structure, and A method for predicting structural strength using an artificial neural network, characterized in that the above vibration-related information includes information related to changes in vibration over time during a first-stroke of the target structure.
  10. In Article 9, The above change in vibration is, A method for predicting structural strength using an artificial neural network, characterized by including at least one of a change in amplitude and a change in frequency over time.
  11. In Article 8, The above first step is, The above target structure is struck n times with the above first force, and The above second step is, Each response to the above n strikes is measured by the vibration sensor to obtain n pieces of vibration information, and The above third step is, A step of inputting each of the n vibration information into the artificial neural network to output n predicted intensities; and A method for predicting structural strength using an artificial neural network, further comprising the step of determining the final predicted strength of the target structure by calculating the average value of the n predicted strengths.
  12. In Article 1, The step of generating the first training data above is, Step 1-1 of generating 1-1 learning data by measuring a 1-1 response after striking the structure of the first strength with a first intensity for the first time; Step 1-2 of generating 1-2 learning data by measuring a 1-2 response after striking the structure of the 1-th strength a second time with the 1-th strength; and A method for predicting structural strength using an artificial neural network, characterized by including a first-n step of generating first-n learning data by measuring a first-n response after striking a structure of the first strength for the nth time with the first strength.
  13. In Article 1, The above artificial neural network learning step is, A method for predicting structural strength using an artificial neural network, characterized by learning with optimal weights and biases by undergoing a forward propagation process according to the following mathematical formula 2; and a backpropagation process. Mathematical formula 2 (In Equation 2, a: activation function, x: 1st, 2nd, and Nth training data as input values, W: weights, b: bias)
  14. delete
  15. delete
  16. delete
  17. delete

Description

Method and System for Predicting Structural Strength Using Artificial Neural Network The present invention relates to a method for measuring the strength of a structure, and more specifically, to a method and system for predicting structural strength that can accurately predict changes in strength, particularly during the hardening process of a concrete structure, by utilizing the response to impact on the structure and an artificial neural network. Concrete is the most widely used construction material, and its mix elements mainly consist of cement, water, fine aggregate, and coarse aggregate. Important characteristic factors of concrete include mix design, quality control, and concrete strength, and concrete strength includes compressive strength, tensile strength, flexural strength, shear strength, and adhesive strength. Safety evaluation indicators for concrete structures are assessed by investigating compressive strength, reinforcement arrangement, reinforcement stress, and the overall deformation of the structure. Among these, concrete compressive strength is an important indicator for evaluating the performance of concrete facilities. Although factors such as salt damage, frost thawing, reinforcement corrosion, and carbonation also affect concrete performance, these factors ultimately affect the compressive strength of the concrete that supports the facility; therefore, compressive strength can be considered a very important factor. Determining concrete compressive strength is recognized as critical not only during the construction phase but also during maintenance in service. Assessing compressive strength during construction is a key factor in determining the timing of formwork removal; premature removal can lead to serious consequences such as structural collapse, while excessively late removal results in delays in the overall construction schedule, increasing both costs and labor requirements. Consequently, determining the optimal removal time is considered a critical element for both cost reduction and construction safety. However, the only realistic alternative for determining concrete compressive strength during construction is currently considered to be underwater curing of specimens made from the same concrete as the structure under construction and compressive failure testing using a UTM in a laboratory. This method, however, has the disadvantage of requiring the production of multiple specimens, and the reliability of the results is compromised because the tests are conducted on specimens under conditions different from the actual structure due to underwater curing. Methods for measuring the compressive strength of concrete structures during maintenance are classified into destructive and non-destructive tests. Conventional destructive testing of concrete compressive strength involved directly extracting concrete specimens to evaluate their condition; however, this method requires significant equipment and manpower, incurs high costs, and raises safety concerns as it not only destroys the structure but can also cause internal reinforcing bars to sever or corrode. To address these problems, non-destructive testing methods are required to minimize damage to the structure during compressive strength testing. For strength measurement, such methods include the rebound hardness method and the ultrasonic probe method, with the rebound hardness method being widely used. Among rebound hardness methods, the Schmidt hammer method is particularly commonly used. Since this method measures rebound force by striking the surface of the concrete, it is influenced by the quality of the surface and striking conditions, making it difficult to accurately measure the internal strength of the concrete structure. Furthermore, in heterogeneous materials like concrete, results vary depending on the presence or absence of aggregates on the striking surface, the state of moisture, and the age of the concrete. Consequently, the results may differ depending on the measurement environment, such as the striking method. In addition, there was a problem where the results varied depending on the prediction formula used to estimate concrete compressive strength using the rebound hardness values obtained from the experiment. It is important to measure concrete compressive strength uniformly and accurately, even if surface conditions or experimental environments differ. As such, among conventional concrete performance evaluation methods, the destructive test involves taking a core from the base concrete and destroying it, which affects the durability of the base concrete and can cause problems with the safety of the structure, and it also has the disadvantage of being costly. In addition, non-destructive testing has the disadvantage of large deviations in measured values and low precision depending on the measurement conditions, and there is a problem of increased costs due to the input of expensive diagnostic equipment and sp