Predictive modelling for coal abrasive index: Unveiling influential factors through Shallow and Deep Neural Networks

Abstract

The use of coal in applications such as power generation is influenced by its combustibility, sulphur content, Hardgrove grindability index (HGI), and abrasivity index (AI), including its volatile matter and ash content, etc. Collectively, these properties affect the quality and performance of coal for its intended use. In particular, the abrasive characteristics of coal from both the same and different coal fields pose challenges during milling operations, leading to accelerated wear and tear of machinery within the plant. In this study, the abrasive index (AI) of coal samples from three South African coalfields using the Yancey, Geer, and Price (YGP) method was conducted. A predictive model using artificial intelligence was developed using the AI obtained from individual samples and other characteristics of the coals. A reliable model utilizing Shallow Neural Network (SNN) and Deep Neural Network (DNN) was developed to relate the coal properties to AI across the different coalfields tested. It is noteworthy that the DNN demonstrated superior performance in modeling AI for all three coalfields, with ash content being the most influential factor. The research findings underscore the importance of other coal properties, including volatile matter, fixed carbon, calorific value, and HGI, in forecasting coal AI. Furthermore, it contributed to the knowledge of the use of artificial intelligence, specifically for a diverse range of coal samples collected from various South African coalfields.