Determination of aeration performance in conduits with different cross section geometries by artificial intelligence methods

Rapid population growth, industrialization and unplanned urbanization causes excessive and inefficient use of water resources, water pollution, and reduces the dissolved oxygen concentration in the water. To achieve ecological balance, the decreased oxygen concentration should be increased by adding the oxygen in the atmosphere to the water. This process is called aeration. Many experimental studies have been carried out on how to aerate of water most efficient and economically. These studies are quite time consuming since they include many parameters that affect the air intake and oxygen transfer mechanism. In addition, the large number of parameters used increases the experimental error rates. It is very important to minimize these problems in the experiments. Artificial intelligence methods, which have been used frequently in engineering applications recently, have also started to be used in aeration studies. In this thesis, it is proposed that the air flow values used to find aeration efficiency can be estimated with artificial intelligence methods. In addition, it has been shown that when conduit gates are evaluated in two classes as radial and sluice, this gate type can be determined by artificial intelligence methods according to the relevant parameters. In this thesis, air flow rate has been estimated by artificial algorithms trained with a data set containing 1008 data obtained from experiments to measure air intake rates. The study uses seven different regression methods and neural networks with two layers. The regression analysis methods used are multi regression, polynomial regression, support vector machines, decision tree regression, random forest regression, gaussian process regression and nearest neighbor regression. The data set used has been divided into two sub-data sets. The first subset which is 80% of the main data set is used in the training stage and the second subset which is 20% of the main data set is used in the test stage. The trained models have been tested with the second subset of data for the success evaluation. In addition, the class of conduit gates, which must be used, are found with seven different classification algorithms, namely Naive Bayesian, logistics, SMO, IBk, JRip, J48 and RF. The metrics used for success evaluations are R2 and correlation values in regression analyzes and accuracy rates in classification algorithms. As a result, it has been seen that the most successful regression models are polynomial and SVR polynomial regressions, and the most successful classification algorithms are RF, J48 and IBk algorithms. The results show that air flow values can be calculated in a short time and at low cost with artificial intelligence models trained with real data, instead of difficult and time-consuming traditional experiments.