Lithology identification during rotary percussion drilling based on acceleration waveform 1D convolutional neural network

Abstract

Identifying the lithology of underground formations is of great importance in the mining industry as it provides useful information about the geometry of rocks under the surface. Current methods are based on core logging and deploying an array of down-the-hole instruments to determine the lithology and structure of the subsurface. Although these methods are highly dependable and accurate, they result in serious time lag. In recent years, various methods have been applied to lithology identification based on acoustic data, vibration data, and two-dimensional convolutional neural networks (CNN). This study presents a system that uses one-dimensional CNN with time acceleration as input data to identify the lithology of an area in real-time. 18 m3 rocks of granite and marble were drilled horizontally with similar drilling parameters, using a rock drill and an intact tungsten carbide drill bit. Time acceleration of drill vibration was measured using acceleration sensors mounted on the guide cell of the rock drill. Model accuracy verification and prediction were carried out. The lithology identification model achieved a verification and classification accuracy of 98.55% and 98.9%. The proposed model was compared to state-of-the-art (SOTA) deep learning neural networks. The model outperformed SOTA methods in terms of validation and classification accuracy. The proposed model gave satisfactory results when evaluated with time acceleration data from defective and abrasion drill bits. It was therefore proven that using drill vibration as input data to a 1D CNN algorithm, provides an efficient method for obtaining the lithology of the subsurface.