A Study on Long-term Strength Prediction Models of Concrete exposed to Local Environment in South Korea

Abstract

The korean social infrastructure was constructed about 20 years ago, and many parts of it are now deteriorating. As a result, the maintenance costs for structures are increasing rapidly. Among the various performance indicators of concrete, the compressive strength is the most basic factor used to evaluate the quality and performance of concrete. thus, it is very important to accurately evaluate and predict this characteristic. However, because concrete is a heterogeneous mixture composed of various materials, it is difficult to predict its strength. Additionally, environmental factors also affect the strength of concrete. Long-term experiments are needed to identify the relationship between these environmental factors and the compressive strength, but there are many limitations. This lack of long-term data has led to unreliable results when predicting the performance of concrete exposed to domestic environments. Therefore, the purpose of this study is as follows: 1) Analyze the long-term strength characteristics of concrete exposed to a domestic environment over a long period of time. 2) Improve the long-term strength prediction models of concrete that reflects the characteristics of the domestic environment. For this purpose, a deterioration environment evaluation was conducted for the domestic environment, and the effects of the environment on the strength characteristics of the concrete were investigated based on this evaluation. In addition, the long-term strength characteristics of the concrete were analyzed according to the mixing conditions, including the water/cement ratio, binder type, and aggregate type. A strength-prediction model was derived based on the obtained long-term data. The strength prediction model was divided into an empirical-formula-based strength-prediction model and an integrated strength-prediction model that considers all variables using machine learning. To examine the performance of the prediction model improved in this study, additional test specimens exposed to the domestic environment for a long period of time were obtained. As a result of examining the performance of the empirical formula-based strength prediction model, ordinary concrete using type 1 ordinary Portland cement and concrete mixed with blast furnace slag showed an accuracy of approximately 92%. In addition, concrete mixed with fly ash showed an accuracy of approximately 80%. For the integrated strength prediction model developed using machine learning, an accuracy of approximately 88% was obtained under the conditions in which learning was performed. Therefore, it was demonstrated that the long-term strength prediction models proposed in this study have high reliability.