Abstract—In this work, Neural Networks (NN) approach was proposed to deal with abrasiveness behavior of thermal coal. Back-propagation neural network (BPNN) and Generalized regression neural network (GRNN) techniques were employed to assess the Abrasive index (AI) of coal to mineral. The multivariate statistical results revealed that the BPNN and GRNN models were successfully developed to model the abrasiveness characteristics of thermal coal with the coefficient of determination, R 2 = 0.9003 for BPNN and R 2 = 0.937 for GRNN. These good results indicated that the NN techniques were capable of accurately modeling the abrasiveness characteristics of coal. Keywords: Abrasive index, back-propagation neural network, coal, generalized regression neural network. I. INTRODUCTION OAL is a combustible, organic rock, which is composed mainly of carbon, hydrogen and oxygen. Coal is primarily used as a solid fuel to produce electricity and heat through combustion. When coal is used for electricity generation, it is usually grinded to an efficient burnable size in a mill and then burned in a furnace with a boiler. During grinding, frictions occur, cause abrasive wear or erosion on the critical components and thereby affect the performance of power plant. It is therefore important to assess the relative abrasion characteristics of thermal coal by selecting the right type of materials for grinding and burning of the coal [1].To study the abrasion of coals, an index namely abrasion index (AI) which has been used to assess the abrasive nature of the thermal coal was established first by Yancey, Geer and Price (YGP) in 1951 using YGP test rig. Over the years there have been some modifications to this method by both mining houses and coal users which have resulted in inconsistent and conflicting results [2]. AI was then measured using different methods, which include the AI tester pot using four iron blades as cutting elements [3]. Reference [4] revealed that two and three-body abrasive wear is not only concerned with hard material, as material of less hardness than the concerned metal blades can still cause material wear. Reference [3] revealed that the effects of Manuscript received July 07, 2016; revised July 22, 2016. This work was supported by the Centre for Renewable Energy and Water, Vaal University of Technology, South Africa. J. Kabuba is with the Centre for Renewable Energy and Water, Department of Chemical Engineering, Vaal University of Technology, Private Bag X021, Vanderbijlpark 1900, South Africa. Tel. +27 16 950 9887; fax: +27 16 950 6491; e-mail: johnka@ vut.ac.za. particles that are less hard than the cutting blades are inconsistent in comparison to harder minerals. To improve the prediction of abrasiveness of thermal coal, it is important to understand the nature and properties of the mineral matters in a coal that would contribute to abrasive wear. Most of the empirical equations available in literature [5]-[9] for predicting AI of coal, are based on linear assumptions which may lead to erroneous estimations and do not take into consideration most of the relevant factors. To achieve this, Neural Network (NN) based predictive techniques was suggested to understand the nonlinear relationships and thereby achieving ability to predict accurately. Reference [10] used non-linear multivariable regression and NN to find the correlation between Hardgrove grindability index and the proximate analysis of chemise coals. Reference [11] also used NN to studies the relationship between petrography and grindability for Kentucky coals. To our knowledge, this is the first time that NN has been used to predict the abrasive index of the coal. The objective of this study is to investigate the possibility for the prediction of abrasiveness characteristics of thermal coal abrasive index using neural network. II. MATERIALS AND METHODS A. Experimental Data The coal samples used in this study was sourced from different colliers in South Africa. These samples were analyzed for both their chemical and physical characteristics based on the premise that they can contribute information that can make it possible for this study to adequately reveal coal constituents that cause abrasion during grinding. The abrasion index tester pot was used to determine AI of the coal samples. A Perkin-Elmer simultaneous thermogravimetric analyzer (STA 6000) equipped with Pyris manager software was used to determine the proximate analyses that gives information about moisture and ash percentage, and petrographically determined minerals in the coal samples. A multivariate statistical analysis was conducted, and from this analysis it was determined that four variables (Ash, Quartz, Pyrite and Moisture) were significant contributors to AI models. B. Back-Propagation Algorithm (BPNN) In this study, the first step was done to scale the inputs and targets within the range 0 and 1 in case the higher values would drive the training process and mask the Application of Neural Networks Technique for Predicting of Abrasiveness Characteristics of Thermal Coal John Kabuba, Member, IAENG C Proceedings of the World Congress on Engineering and Computer Science 2016 Vol II WCECS 2016, October 19-21, 2016, San Francisco, USA ISBN: 978-988-14048-2-4 ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online) WCECS 2016