Aplicações de processamento digital de imagens no estudo da microestrutura de materiais

Abstract

Studying the material microstructure can become a broad and confusing task due to the large amount of existing materials and the similarity they present to each other. Usually, the analysis is carried out by experts who uses visual inspection; some softwares allow the automation of the process, however the cost of acquiring the license is high and it is required the use of specialized equipment to get high quality images. The information provided by the metallographic analysis can be used to predict the mechanical characteristics of the material, including the life time and its behavior at different environments. The present work started as a requirement of the GiMat research group of the Politecnica Salesiana University located in Cuenca-Ecuador, GiMat is a research group working on the generation of new materials and analysis alternatives, with a vision not only in the academic area but also in the industrial area. In this way, the present work is divided into two parts. In the first part, metallographic images are analyzed for the identification of the corresponding steel type and in the second part the geometrical properties and orientation tensors of biocomposites images are determined. Obtaining metallographic images, it is proposed to develop an automatic and efficient method which allows engineering students to identify the type of steel and verify the effects of the use of thermal treatments on the material by replacing traditional analysis methods and industrial-oriented. This project will contribute to the on-site inspection of machine parts or structures without altering its state. For this purpose, five of the most common types of steel for machine and tool fabrication were used, such as: AISI 4340, AISI 01, AISI D6, AISI 01 (rectified) and AISI 1018. First of all, the images to be used are subdivided, secondly, Laws's texture filters and Haralick's extraction features as well as Wavelet descriptors are applied to obtain the classification attributes. It was verified, that the best results are obtained with ELM and MLP, with a success rate higher than 90%, which evidences the effectiveness of the proposed procedure. In the second part, the aim is to obtain the geometrical characteristics of the composite material fibers and its orientation tensors. For this purpose, images of the compound PP-GAK (polypropylene with Guadua Angustifolia Kunth) at 30% of concentration by weight are used. In order to obtain the geometric characteristics, the ellipses method is used. Initially, standard images with known information are used to evaluate the performance of the proposed technique, obtaining a systematic error of 0.3% for A11. Subsequently, the orientation tensors of the images are calculated. These results are compared with an analysis obtained from Stream Essential software, the comparison shows a similarity rate of the data obtained. The most important aspect evaluated is the A11 component of the tensor matrix which shows a similarity percentage of 98%. Finally, alternative analysis for this type of images are presented considering the automation and optimization of the process for the new researchers.