Unlike most other rocks, coal is a sedimentary rock composed primarily of organic matter derived from plant debris that accumulated in peat mires during different geological periods. Coal is also an essential economic resource in many countries, having been the main driving force behind the industrial revolution. Coal is still widely used industrially for many different purposes: carbonization and coke production, iron/steel making, thermal coal to generate electricity, liquefaction, and gasification. The utility of the coal is dictated by its properties which are commonly referred to as its rank, type, and grade. Coal composition, in terms of its macerals, and its rank determination are determined manually by a petrographer due to its complex nature. This study aimed to develop an automatic method based on machine learning capable of maceral segmentation at group level followed by a module for rank reflectance determination on petrographic images of coal that can improve the efficiency of this process and decrease operator subjectivity. Firstly, a Mask R-CNN-based architecture deep learning approach was developed to identify and segment the vitrinite maceral group, which is fundamental for rank analysis, as rank is determined by collotelinite reflectance (one of its individual macerals). Secondly, an image processing method was developed to analyze the vitrinite segmented images and determine coal rank by associating the grey values with the reflectance. For the maceral (group) segmentation, five samples were used to train the network, 174 images were used for training, and 86 were used for testing, with the best results obtained for the vitrinite, inertinite, liptinite, and collotelinite models (89.23%, 68.81%, 37.00% and 84.77% F1-score, respectively). Those samples were used alongside another eight samples to determine the rank results utilizing collotelinite reflectance. The samples ranged from 0.97% to 1.8% reflectance. This method should help save time and labor for analysis if implemented into a production model. The root mean square calculated between the proposed method and the reference reflectance values was 0.0978.
Santos, Richard Bryan Magalhaes; Paciornik, Sidnei (Advisor); Augusto, Karen Soares (Co-advisor). An image analysis system for the characterization of sinter feed microclusters. Rio de Janeiro, 2018. 89p.The ores, once extracted, undergo several stages of processing before they can be properly used. The fines of ores that, at the end of this stage, do not have the granulometry required to feed the reduction furnaces, pass through agglomeration processes to reach it, such as pelletizing and sintering. The material produced in one of the stages of the latter process is the focus of this work. These fines first go through a micro-agglomeration stage, which is fundamental to the process because many of the characteristics and properties of the sinter are function of the structure of the pre-heat treatment microcluster. It consists of a mixture of the sinter feed, water, fluxes and solid fuel (coke). There are 3 typical structures for a microcluster: quasiparticle, micropellet, and non-agglomerated particles. The present dissertation has developed an automatic routine in the FIJI image-processing program, based on optical microscope image processing and analysis, which is able to identify the particles of different granulometry that compose the sample, and classify them in the 3 classes mentioned above. After classification, the routine is able to extract attributes of the identified objects (percentage of each class, average circularity, average thickness), and to analyze the quasiparticle nuclei, classifying them as to the phase (hematite, magnetite, goethite and others). In addition, the routine presents all the data in the form of a pdf report, which also contains a listing of quasiparticles and micropellets in increasing order of size. This automatic classification eliminates the lack of reproducibility and subjectivity of the human operator, provides measures that would be untenable manually, allowing the forecast of the future characteristics of the sinter in a fully automatic fashion.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.