Chiranth Hegde

Acoustic fingerprinting for rock identification during drilling

During the process of mining, it is imperative to know the type and properties of the rocks being handled. The current technology for this involves core drilling, and subsequently subjecting the drilled cores to various tests in the laboratory, to identify the rocks and establish their properties. In many cases, obtaining a sample may be cumbersome and/or non-profitable. This paper presents a novel method to monitor and evaluate the sounds produced as undesirable by-products, at the drill-bit and rock interface, to predict the type of rock being drilled. A rotary drill was fabricated in the laboratory and vertical drilling was carried out on cubical rock samples, keeping various drilling parameters constant. The results obtained are promising and reinforce that it may be possible to extend the proposed methodology in the field as well, with appropriate modifications. This method may be extrapolated further in the estimation of rock properties as well.

A Critical Comparison of Regression Models and Artificial Neural Networks to Predict Ground Vibrations

Blasting is important and an essential prerequisite in any opencast mine for fragmenting hard deposits. Blasting always produces unwanted effects like ground vibrations, noise and fly rock; among which ground vibrations effect is more on surrounding structures. Propagation of ground vibrations can lead to destruction of surrounding structures. Prediction of ground vibrations especially in terms of peak particle velocity is beneficial as opposed to conventional data monitoring techniques which can be expensive as well as time consuming. This paper uses predictors to estimate the intensity of ground vibrations and compares different methods of prediction methods like linear regression, multiple linear regression, non linear regression (NLR) and artificial neural networks. Intensity of ground vibrations generated from blasting operations was monitored in three different mines of limestone, dolomite and coal; obtaining about 168 ground vibration recordings in total. The statistical modelling or data-driven modeling has shown promise in the prediction of blast vibrations. Proposed a system of introducing site specific rock parameters like poison’s ratio, uniaxial compressive strength of rock and Young’s modulus to improve the correlation coefficient using statistical modelling (commonly called feature engineering in machine learning circles).

Prediction of peak particle velocity using multi regression analysis: case studies

ABSTRACT Ground vibrations produced from blasting operations cause structural vibrations, which may weaken structure if it occurs at the resonant frequency. Measurable parameters associated with ground vibrations are peak particle velocity (PPV), amplitude and dominant frequency (frequency of highest PPV amongst translational, vertical and horizontal vibrations). In this paper, an attempt is made to correlate measurable parameters associated with ground vibrations with scaled distance. Using the correlated data, it was found that a predictor equation can be determined for the amplitude and PPV, but not for dominant frequency as it is dynamic and depends upon infinitesimal changes that occur within a number of other parameters. Another analysis of the same is made using multiple linear regression analysis. This included predicting the PPV using scaled distance, maximum charge per delay, amplitude as predictors. A considerable improvement is seen in the prediction on adding the interaction of the predictors in multiple regressions. A comparison of different combination of predictors is made so as to assess the best combination giving the best R2 value for the given mine. Frequency is also plotted using the aforementioned method. However, it was found that the dominant frequency cannot be predicted with high accuracy even with this method.

Classification of Stability of Highwall During Highwall Mining: A Statistical Adaptive Learning Approach

AbstractThe depleting coal deposits day by day required the introduction of novel methods of mining like highwall mining. Highwall mining is a method of extraction of coal blocked in the highwall. The method involves considerable challenges in the area of roof control and most importantly the stability of the highwall itself. Highwall mining has gained considerable importance all over the world, owing to the fact that the coal otherwise would not be extracted forever. This paper aims to assess the influence of varying conditions which can affect the stability of the highwall during highwall mining. The effect of gallery length, width of pillar and number of galleries are systematically studied through field investigations where a highwall mining was adopted first time in India. Initially, assessment was carried out using a numerical modelling approach and then the stability of the highwall is classified using multilinear regression, logistic regression and naive Bayes classifier. This will provide a mechanism to predict the stability of the highwall in future cases of similar conditions. The classification is done using statistical adaptive learning methods and a comparison of the methods is done.