Statistical Analysis of the Capabilities of Various Pattern Recognition Algorithms for Fracture Detection Based on Monitoring Drilling Parameters

Ground conditions, including characteristics of fractures, joints, bed separations, and strengths of rock layers, are critical factors for proper design of openings in underground mining and construction projects. Correct understanding of geologic conditions allows for improvement and optimization of ground support design and minimizing incidents of ground failure and instabilities in underground workings. Rock bolts have been widely accepted as the preferred method of ground support in almost all forms of rock excavation applications. The concept of monitoring drilling data to evaluate characteristics of geological features of interest in the rock surrounding the underground opening is a very attractive option for developing the geological model of the ground on real-time basis. This includes information on distributions of joints and bed separations, locations of voids, and strengths of rock layers, which enables the automated and rapid evaluation of ground conditions while drilling is in progress. Several smart drilling systems have been developed and proposed to detect joints; however, they offer limited capabilities and have exhibited difficulties in identifying joints with small apertures. The current study was focused on developing a more sensitive method to locate joints with smaller apertures along the hole being drilled with an instrumented roof bolter. A series of full-scale drilling tests were carried out in samples which contained simulated joints with different inclined angles in controlled laboratory conditions. New joint detection programs, with improved capabilities based on various pattern recognition algorithms, have been developed and used for analysis of data recorded in the full-scale drilling tests. To precisely locate joints, composite parameter was also used to offer more accurate detection. This paper reviews the laboratory testing program, data analysis, logic/algorithms used in the programs, statistical analysis of the detection results, and comparison of the various algorithms for this application.