Modeling of the system to predict the performance of steel fiber reinforced self-compacting concrete using machine learning methods

Self-compacting concrete, a special type of concrete, is successfully used in special structures. The use of self-compacting concrete has been increasing in recent years due to the advantages it provides. The most important of these advantages are the cost reduction and shortening of construction time compared to conventional concrete. At the design stage of self-compacting concrete, high dust content and low water-binding ratio can cause poor stability. Steel fiber is added to the concrete mix to improve this weakness. Steel fibers are excellent materials for controlling cracking development of concrete types and increasing tensile strength. For these reasons, self-compacting steel fiber concrete is a special type of concrete that has been widely researched in the literature due to its superior performance properties. At least two different fresh concrete and one hardened concrete experiments are required to determine the performance of this particular type of concrete in order to obtain proper mixtures. These experiments require a long time, cost and technical workforce. Steel fiber self-compacting concrete is considered to require an economical and fast design process, as it is difficult to obtain proper mixing ratios and is dependent on too many parameters. In order to eliminate this problem, in this thesis study, models that will predict the performance of fresh and hardened concrete of steel fiber reinforced self-compacting concrete by using machine learning methods are proposed for the first time. To predict fresh concrete performance, v-funnel, T50 and slump-span tests were selected, and 28-day standard cube sample tests were selected for the hardened concrete experiment. Data sets for these experiments were collected experimentally and from the literature for the first time. K- nearest neighbor, linear regression, regression trees, support vector machines and extreme learning machines were used as basic machine learning methods. Fresh and hardened concrete performances of steel fiber reinforced self-compacting concrete were successfully predicted using machine learning methods. These methods have been tried and applied successfully for the first time in the datasets and experimental data sets in the literature in different areas. The performance of the extreme learning machines method has been improved by using five different chaotic maps (Chebyshev, iterative, logistics, piecewise and tent). It is observed that the proposed new methods give better results in three of the four different experiments. It is thought that the proposed new methods will be applied successfully in different areas.