Abstract
M.Ing.(Electrical and Electronic Engineering)
The study investigated whether non-destructive impact testing, aided by supervised machine learning methods, could be used to identify improper roof bolt installations, related to insufficient grout coverage. The testing method involved the installation of four roof bolts, with varying installation properties, into a 1511 × 940 × 1350mm rock test block. Three fully grouted bolts served as examples of proper installations, with the fourth bolt grouted only up to half the length of the borehole serving as an improper roof bolt installation. The testing procedure involved placing sensors directly onto the bolts and mechanically impacting a chosen bolt while measuring the response on all the bolts. The focus was on gaining understanding of the working principle of the testing technique and how the measured response was influenced by the presence of signal-modifying factors of the physical test block geometry, such as changes in material properties, boundary changes, cracks or empty boreholes. It was shown that the roof bolt integrity testing method aided by supervised machine learning methods could identify and classify both properly and improperly grouted roof bolts on the small sample of test bolts, in a series of tests conducted at the CSIR Centre for Mining Innovation premises. The method was also shown to be robust enough to do so even in the presence of the signal-modifying factors of the physical test block geometry.