The application of artificial intelligence for the identification of the maceral groups and mineral components of coal

Highlights

• The automatic classification of maceral groups and mineral matter in hard coal using AI was proposed.

• The efficiency of pattern recognition and artificial neural networks was compared.

• A high rate of compatibility of automatic classification with petrographers' decisions was achieved.

• The results prove the usefulness of AI in supporting of quantitative analyses of hard coal.

Abstract

The correct and consistent identification of the petrographic properties of coal is an important issue for researchers in the fields of mining and geology. As part of the study described in this paper, investigations concerning the application of artificial intelligence methods for the identification of the aforementioned characteristics were carried out. The methods in question were used to identify the maceral groups of coal, i.e. vitrinite, inertinite, and liptinite. Additionally, an attempt was made to identify some non-organic minerals. The analyses were performed using pattern recognition techniques (NN, kNN), as well as artificial neural network techniques (a multilayer perceptron – MLP). The classification process was carried out using microscopy images of polished sections of coals. A multidimensional feature space was defined, which made it possible to classify the discussed structures automatically, based on the methods of pattern recognition and algorithms of the artificial neural networks. Also, from the study we assessed the impact of the parameters for which the applied methods proved effective upon the final outcome of the classification procedure. The result of the analyses was a high percentage (over 97%) of correct classifications of maceral groups and mineral components. The paper discusses also an attempt to analyze particular macerals of the inertinite group. It was demonstrated that using artificial neural networks to this end makes it possible to classify the macerals properly in over 91% of cases. Thus, it was proved that artificial intelligence methods can be successfully applied for the identification of selected petrographic features of coal.

Introduction

Coal is not a homogeneous substance. It is built of elementary components known as macerals, which can be distinguished only with a microscope. The term “macerals” was first defined by M. Stopes for the constituent of coal isolated by maceration (Stopes, 1935). They are identified across three main groups: vitrinite, inertinite and liptinite (ICCP, 1998, ICCP, 2001), that differ with respect to their genesis – and, what follows, to their optical and strength properties (O’Keefe et al., 2013). The three maceral groups are, to a certain degree, characterized by their chemical composition. If one compares isometamorphic maceral groups, i. e. groups of the same rank, the vitrinite contains relatively more oxygen, the liptinite - more hydrogen, and the inertinite - more carbon (Stach et al., 1982).

Vitrinite is the most frequent maceral group occurring in bituminous coals. It is derived from the tissues of organic components, i.e. plants. In reflected light, the color of the macerals from the vitrinite group changes with the degree of coalification, from dark gray to light cinerous. Still, it is much lighter than the macerals of the liptinite group. As to the inertinite group, it originally comprised the microelements of coal that are inactive in the coking process. Later, it was demonstrated that some macerals from this group do possess coking properties. However, the most important role in this process is played by micrinite, formerly termed „fine micrinite”. The characteristic optical property of inertinite macerals is their high reflectance. The third group of macerals – the liptinite group - consists of sporine, cutine, suberine, resins, waxes, fats, and oils of vegetable origin. This group is characterized by the lowest reflectance. An important optical property of the macerals belonging to this group is the fluorescence phenomenon, which occurs when they are exposed to blue or ultraviolet light. The intensity of this phenomenon decreases as the degree of coalification increases, until it disappears completely (Stach et al., 1982, Diessel, 1992).

Petrographic coal analyses are used by geologists in order to obtain information about investigated coal seams. Additionally, the petrographical composition of coal has a significant influence on its technological properties. Most frequently, petrographical analyses of coal are helpful in evaluating its quality from the perspective of using it in the coking process. Such analyses are also used to monitor the quality of the mining output, as well as to detect and identify impurities.

Microscopic quantitative analyses of coal macerals are usually performed manually - and one of the key issues is to identify macerals correctly and consistently. From previous research we know that this is not a trivial question (Bodziony et al., 1986). In the case of manual analyses, the difficulties result from substantial diversification of the petrographical properties of hard coal, which is the consequence of differences in the genesis of particular components. Such research requires taking into consideration both the optical properties and the diversified structure of macerals. Differences concerning the results of analyses conducted by various specialists in petrography can be substantial. Researchers point out that, in such studies, an important role is played by the subjective factor. Thus, measurements need to be objectivized (Bodziony et al., 1986).

Automated methods constitute an alternative to manual methods. They are based mainly on image analysis methods. Automatic methods are increasingly used in the geological sciences, giving many interesting results (e.g. Obara, 2007; Sharif et al., 2015; Sochan et al., 2015; Campaña et al., 2016). One should also mention the innovative approach to image analysis – Object-based Image Analysis (OBIA). In the object-based classification approach, we do not analyze separate pixels, but the so-called objects which fit the required criteria of homogeneity (spectral characteristic, texture, shape, size, relationships between adjacent objects, etc.). These methods are successfully used in petrographical issues (e.g. Marschallinger and Hofmann, 2010; Hofmann et al., 2013; Leitner et al., 2014).

It should be noted that using the image analysis algorithms, a mathematical description of maceral groups and a precise procedural algorithm for microscope analyses may be difficult to implement. This is due to the multidimensional nature of such measurements and the fact that particular macerals sometimes differ only subtly. In such cases, the AI methods may prove a helpful instrument for analysis, as they do not require the knowledge of exact relationships in the data set. Instead, they make use of learning sets, which contain information about the patterns available for the issue being investigated. The advantage of such a solution is also the fact that they may yield repeatable measurement results, which neutralizes the impact of the subjective factor on conducted analyses.

The purpose of the discussed research was to develop a universal and repeatable body of methods based on the AI methods, by means of which it may be possible to identify maceral groups. The aim was to achieve the state in which the identification process is based solely on the knowledge obtained by the computer from sample images and information as to their content. Pattern Recognition is the field of AI-related research that has the longest tradition (Poole and Mackworth, 2010). The aim of pattern recognition is to classify various types of objects into certain, previously defined categories. The recognition process is carried out when there is no a priori information regarding an object's affiliation to a group, and the only information that we have is included in a learning set consisting of objects for which the right type of classification is known (Duda et al., 2001; Theodoridis and Koutroumbas, 2003). The use of pattern recognition methods for automatic classification of the maceral groups has been proposed among others such as D.P. Mukherjee et al. (1994). This method had two modules, an off-line training module and an on-line classification module and maceral class was determined based on the gray value (or the reflectance properties) in the RGB color space. The ways of using pattern recognition in geology and mining sciences are the subject of continuous investigation. Among works exploring this topic, there are the following: Marschallinger (1997), Wei-hua and Hong-wei (2009), Młynarczuk (2010), Młynarczuk et al. (2013), Eberle et al. (2015).

More recent methods used in AI studies are artificial neural networks. They are inspired by the way the human brain works, and, as a matter of fact, they constitute its simplified mathematical model (Tadeusiewicz et al., 2014). The first model of a neuron was proposed in 1943 by McCulloch and Pitts. These days, artificial neural networks are an advanced field of study – but, at the same time, it is a field that is constantly being developed and applied to new ends, also in the area of mining, geology and environmental protection (Atakan and Ayşe, 2010). Some of the applications of artificial neural networks include: prediction of coal and gas outburst (Ruilin and Lowndes, 2010); identifying lithofacies in petroleum and gas mining, identification of geochemical anomalies (Ziaii et al., 2009), predicting the pace of drillings (Aalizad and Rashidinejad, 2014), or rock and mineral identification (Koujelev et al., 2010, Aprile et al., 2014, Jamróz and Niedoba, 2015).