Improved empirical hard rock pillar strength predictions using unconfined compressive strength as a proxy for brittleness

• Published in: International journal of rock mechanics and mining sciences (Oxford, England : 1997) England : 1997), 2021-12, Vol.148, p.104934, Article 104934
• Main Authors: Walton, G., Sinha, S.
• Format: Article
• Language: English
• Subjects: Brittleness; Compressive strength; Critical components; Ductile-brittle transition; Empirical database; Failure mechanisms; Hard rock; Mechanics; Pillar strength; Rock brittleness; Rock mechanics; Rocks; Spalling; Underground mining
• ISSN: 1365-1609; 1873-4545
• DOI: 10.1016/j.ijrmms.2021.104934

Pillars are a critical component of underground mining systems, and for this reason, pillar mechanics has historically been a major topic of rock mechanics research. In the late 20th century, many empirical studies on hard rock pillar strength were conducted, culminating in the development of a pillar strength formula by Lunder & Pakalnis( Lunder and Pakalnis, 1997) 1 based on a large database of pillar cases with different performance outcomes. In this study, we re-evaluate this database with the benefit of an improved understanding of hard rock pillar mechanics based on recent findings in the literature. Specifically, pillar strength can be governed by spalling-dominated, shear-dominated, or transitional failure mechanisms depending on pillar W/H (and associated confining stress conditions). Based on this, we identify that the W/H ratios at which these transitions in strength trends occur vary as a function of brittleness, which can be approximated using the unconfined compressive strength (UCS). In contrast to existing empirical hard rock pillar strength models, this implies a non-linear relation between pillar strength and UCS. We then develop and compare a series of models that account for this behavior against the empirical database compiled by Lunder & Pakalnis( Lunder and Pakalnis, 1997) .11 Our final model is shown to be both more accurate and more conservative than that of Lunder & Pakalnis( Lunder and Pakalnis, 1997) ,11 while still requiring the same basic input parameters (UCS and W/H); specifically, our model accurately classifies 98.4% of “Stable” and “Failed” cases within the database, as compared to 90.6% for the Lunder & Pakalnis( Lunder and Pakalnis, 1997) 1 model.