Experimental and artificial intelligence approaches to measuring the wear behavior of DIN St28 steel boronized by the box boronizing method using a mechanically alloyed powder source

Abstract

Wear in moving materials in contact with each other is an inevitable cause of damage. To prevent this damage, various processes are applied to the material surfaces. The most widely used method is the surface hardening method. This study aims to examine the wear properties of the samples by forming a hard boride layer on the surface of low-carbon steel such as St28 with experimental and artificial intelligence approaches. In this context, it is aimed to obtain the boride layer at relatively low temperatures by pre-processing the powder mixture to be used as a boron source, such as Mechanical Alloying (MA). The boronizing process was carried out using the box boronizing technique. The wear behavior of the obtained samples was investigated by the block-on-disk method. In artificial intelligence approaches; The dataset is divided into three categories as 10N, 20N, and 40N. There are 39 sample types and attributes in each category. In this study, feature selection algorithms such as linear regression (LR), ridge, recursive feature elimination (RFE), f-regression, and multiple inclusion criterion (MIC) were used to select the most efficient samples. Then the best samples were classified according to their force types. Ensemble learning methods, machine learning methods, and Bayesian neural networks were used in the classification processes. Thanks to the proposed approach and feature selection algorithm, the best performance has been shown up to 10 feature selection. By ignoring 29 inefficient features, classification was performed with 10 efficient features. In the classification process, 100% overall accuracy was achieved.