Akademik Veri Yönetim Sistemi
JOURNAL OF POLYTECHNIC-POLITEKNIK DERGISI, 2025 (ESCI)
The properties of materials can be definitively determined by examining their microstructure or characterisation. Microstructural imaging provides essential insights for both the characterization of novel materials and the optimization of manufacturing processes existing materials. The analysis of these images is economically prohibitive and demands a high level of material-specific expertise. Despite expert analysis, the interpretation of microstructural images is susceptible to subjective bias, leading to erroneous conclusions. The precise, timely, and optimal assessment of microstructural images is of paramount importance in this field. Through implementation of sophisticated artificial intelligence algorithms, the evaluation of microstructural images can be expedited, thereby reducing the likelihood of errors. Deep learning constitutes a sophisticated artificial intelligence algorithm. Deep learning models have exhibited a high degree of accuracy in image processing applications. The purpose of this research is to examine different microstructures of AISI 4340 steel by employing artificial intelligence algorithms. In order to produce bainitic, martensitic and pearlitic microstructures in AISI 4340 steel, austempering, quenching and normalization heat treatments were applied, respectively. Optical microscopy was employed to image diverse microstructures and inclusions resulting from different heat treatment processes, and obtained images were compiled into a dataset. The VGG16 model was employed for microstructure classification, and the U-Net model was utilized for inclusion identification. The performance metrics of these models are as follows: The VGG16 model exhibited accuracy of 93.33% in microstructure classification tasks. The U-Net model achieved an accuracy of 98.50% and a dice score of 73.59% for inclusion segmentation.