Doğaltaşların aşındırıcı sujeti ile kesilmesine yönelik bir araştırma: kesme performansının iyileştirilmesi, modelleme ve optimizasyon için makine öğrenimi algoritmalarından yararlanılması

In this thesis, the performance of abrasive waterjet (AWJ) multi – pass cutting and forward angling the jet was experimentally investigated using the workpieces of various rocks including marble, travertine, basalt, onyx and tuffs. Additionally, ;the cutting performance indicators were modelled and optimized using machine learning algorithms (MLAs) such as artificial neural network, gradient boosted decision trees, Gaussian process regression, support vector machine and the particle swarm optimization. Once the MLAs-based models were developed, the robustness of the models was comprehensively assessed through the statistical indices of determination coefficient, root mean square error and mean absolute percentage error. In addition, a random forest regressor (RFR) was leveraged to determine the feature importance of predictor variables influencing the proposed models. In comparison to the single cutting, it was determined that the multi-pass cutting significantly improves cutting performance, depending on the cutting performance indicator and rock type. In experiments where the forward angling the jet was used, improvements of up to 42.94% were achieved, depending on the performance indicator and rock type, although not as much as the improvements achieved with the multi – pass cutting. In addition to these improvements, it was found that the increase in abrasive flow rate and water pressure during AWJ multi – pass cutting of rock led to substantial improvements. Moreover, it was concluded that the MLAs can be successfully used to model and optimize the cutting performances obtained by AWJ multi – pass and forward angling the jet cutting techniques. Key Words: Abrasive Waterjet, Natural Stone, Machine Learning, Modelling, Optimization