Integration of Machine Learning and Synthetic Data for Long-term Prediction of Concrete Compressive Strength: Application in Structural Design

This research explores machine learning and synthetic data creation to improve the long-term forecasting of concrete compressive strength (CS), an essential structural design and longevity element. Conventional empirical datasets frequently do not provide enough long-term insights, which restricts predictive accuracy. We created synthetic datasets of ۱,۰۰۰, ۵,۰۰۰, and ۱۰,۰۰۰ data points to tackle this issue through K-means clustering and TGAN. We concluded that ۵,۰۰۰ data points best balanced computation efficiency and predictive accuracy. In contrast to earlier studies that mainly concentrate on short-term forecasts, this research seeks to enhance three-year CS predictions by combining ۷,۸۴۵ actual data points with ۳۲,۱۵۵ synthetic samples. We utilized sophisticated machine learning models, such as Random Forest (RF) and Nonlinear Auto-Regressive Neural Networks (NARX), with NARX attaining the greatest predictive accuracy. A significant contribution of this study is the thorough assessment of the lasting impacts of chemical admixture combinations, a factor frequently neglected in current research. The results emphasise K-means clustering as an effective computational method for producing synthetic data and improving the accuracy of CS forecasting. This research greatly enhances predictive modelling by combining empirical and synthetic data, strengthening AI-based decision-making in structural engineering. These findings provide actionable implications for enhancing material choice, boosting structural dependability, and encouraging eco-friendly building methods. The suggested framework offers a strong approach to evaluating durability, aiding in developing more resilient and effective civil engineering solutions.

کلیدواژه ها:

Compressive strength prediction ، Synthetic Data ، Machine learning ، TGAN ، K-means clustering ، Structural Design