Buy article PDF
The purchased file will be sent to you
via email after the payment is completed.
US$ 35
|
Smart Structures and Systems Volume 34, Number 6, December 2024 , pages 377-406 DOI: https://doi.org/10.12989/sss.2024.34.6.377 |
|
|
|
Predicting the mechanical properties of high-performance concrete implementing boosting models integrated with metaheuristic algorithms |
||
Fangxiu Wang and Jiemei Zhao
|
||
| Abstract | ||
| Compressive Strength (CS) and Tensile Strength (TS) are vital in designing a reinforced concrete structure. In fact, ensuring structural strength and safety requires both properties. Establishing predictive models for CS and TS yields high advantages, ensuring considerable cost savings by reducing labor-intensive and time-consuming lab experiments. This becomes more critical where high performance is necessary, such as in testing advanced HPC, renowned for its remarkable durability and strength in critical infrastructure and construction undertakings. The use of machine learning has become one of the innovative approaches to predicting these concrete characteristics. Through data-driven analyses on ingredient ratios, curing conditions, and environmental exposure, ML returns quite accurate CS and TS predictions. Applying ML methodologies results in gains in efficiency, cost economy, design refinement, higher-quality control, and increased safety. The contribution of this paper is the realization of an extended comparison between many algorithms. In particular, this work investigates two ML-based models: Histogram Gradient Boosting (HGB) and Light Gradient Boosting (LGB). These models are combined in a structured fashion with three newer optimization algorithms: Snake Optimization Algorithm (SO), Fox Optimization Algorithm (FO), and the Prairie Dog Optimization Algorithm (PDO), along with an ensemble of all three optimizers, namely SO-FO-PDO. From the results represented by the R2 values, it is obvious that the HGPD model demonstrated far better forecasting performance for the CS, with a resultant R2 value of 0.9961 during training. Similar to TS, the HGPD model developed as the best estimator in the case of TS with an R2 value of 0.9947 for the training phase. Besides, from the statistical measures of accuracy w.r.t MAE and RMSE, it has been quite evident that the proposed PDO-based hybrid and ensemble models outperformed the rest by a long margin in estimating concrete mechanical properties. | ||
| Key Words | ||
| compressive and tensile strength; high-performance concrete; light gradient and histogram gradient boosting; meta-heuristic algorithms; sensitivity analyses | ||
| Address | ||
| School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan 430023, Hubei, China. | ||