1. A comparative analysis of algorithms on concrete at elevated temperatures
- Author
-
Duan, Min
- Abstract
Elevated temperatures have a substantial influence on the properties of the materials used in concrete manufacturing, leading to a noticeable decrease in the strength properties of the concrete. Attaining the intended compressive strength of concrete is a demanding and time-consuming pursuit. However, the use of supervised machine learning methods allows for an accurate initial prediction of the intended result. This work demonstrates the utilization of multi-layered perceptron (MLP), support vector regression (SVR), and least squared SVR (LSSVR) for predicting the compressive strength of concrete at elevated temperatures. The prediction models were enhanced by using the reptile search algorithm (RSA) and were trained on a dataset consisting of 207 data points. The investigation included a collection of nine input variables, namely water, cement, coarse aggregate, fine aggregate, fly ash, superplasticizers, silica fume, nano silica, and temperature. The results indicate that all the LSSR, SVRR and MLPR show great potential in accurately predicting the fcof concrete at elevated temperatures with acceptable values of the R2. Regarding the values related to the error-based metrics, the lowest values belonged to the LSSR model with a decline of about 50% with respect to the SVRR and MLPR networks. Turning to uncertainty analysis, it was obvious that in all three training, validating, and testing phases, MLPR resulted in the highest values of uncertainty, compared to SVRR. On the other hand, the most reliable model called LSSR introduced the smallest values of uncertainty at 6.028, 5.98, and 6.52, respectively, which depicted remarkable accuracy and dependability.
- Published
- 2024
- Full Text
- View/download PDF