Underground » Geology
Coal quality parameters such as ash content, density, volatile matter and specific energy are important to the coal mining industry from mine planning, design, extraction and beneficiation through to utilisation. These parameters are traditionally obtained through laboratory analysis conducted on drill-core samples. Currently, obtaining coal quality information requires the collection of bore cores, which are then subjected to pre-treatment to simulate the size reduction and liberation that can be expected during the mining process. This process is expensive and time consuming. In addition to this, most boreholes are drilled with limited or with no coring due to costs. Therefore, only a limited number of coal samples can be tested and analysed and this largely limits the ability to appropriately map the spatial variability of the coal quality in both horizontal and vertical directions. Obtaining estimates of these coal quality parameters from non-cored holes would complement this information and thus provide a better estimate of the resource.
The objectives of this project are to:
· Develop new methods for enhanced estimation of routinely measured coal quality parameters (i.e. proximate analysis properties, specific energy and relative density) from routinely acquired geophysical logs such as density, gamma and sonic, thus significantly improving the spatial definition of coal deposits;
· Establish whether the use of the multiple geophysical logs significantly enhances the reliability of the predicted coal quality parameters.
In this project, we reviewed the coal quality estimation from geophysical logs. It was found that the commonly-used approach for determining coal quality from the geophysical logs is mainly based on simple cross-plots. However, the relationships between coal quality parameters and geophysical logs are not always represented by simple linear trends and may instead be curved trends generated by complex equations. This suggests that instead of using a simple two-variable correlation approach, a multi-variable data analysis approach has a better chance of dealing with the complexity of coal quality parameters and thus improve the estimation accuracy of the these parameters. To perform coal quality parameter estimations using multiple geophysical logs, we proposed and implemented a multi-variable data analysis algorithm based on the use of a particular type of neural network known as Radial Basis Functions (RBF). We chose this approach because of its two important advantages:
1) It estimates coal quality parameters from parameters and relationships derived within the data set without a need for pre-existing assumptions or models;
2) It can easily accommodate the coal rank variations by simply adding the representative samples into the control data base. We also developed data pre-processing algorithms to extract the geophysical logging data corresponding to the coal samples.
The RBF-based coal quality parameter estimation algorithms were tested by using the data sets provided by Anglo American and BMA. In both cases, routinely-acquired geophysical logs such as density, gamma ray and sonic logs have been used to estimate the coal quality parameters such as relative density, ash content, fixed carbon and volatile matters. This has been demonstrated on both self-controlled training data sets and an independent data set. It is observed that the density logs play a key role in coal parameter estimation. However, the use of multiple types of geophysical logs, including logs with different resolutions, such as short spaced density log DENB and long spaced density log DENL, improves the estimation accuracy. It is therefore expected that more accurate coal quality parameters can be estimated if more geophysical logs such as photo-electric factor (PEF), SIROLOG and PGNAA which provide geochemical constituents are acquired.