A statistical approach for the integrated analysis of mine-induced seismicity and numerical stress estimates, a case study - Part 1: developing the relations

Integrated analyses at a number of underground operations have suggested that there may be qualitative relations between microseismic variable values estimated in real-time from mine seismic systems and the stress estimates provided by numerical models. In this study, multivariate statistical techniques were used to evaluate whether quantitative relations could be formed using data from the Creighton Mine in Sudbury, Ontario. It is believed that if sufficiently reliable relations could be developed, they might then be exploited to better characterize the state of the rock mass and provide insight beyond what could be obtained from either data set in isolation. This insight could then be used to eventually impact both short- and long-term mine design decisions. This paper forms the first part of a two part series and describes the methodology that was developed to form a relatively small number of meaningful and statistically significant relations of interest between variables within these two complementary databases. The methodology developed consists of obtaining representative values for variables within both databases and a number of data manipulation steps. In particular, the data set was manipulated by filtering events generated through non-random effects, transforming the variable values, removing outlying values and data smoothing. In the end, through the application of this methodology, a relatively large number of multivariate relations were formed and a five-part strategy was needed to identify the relations of greatest interest. The results of this part of the study indicated that it was possible by following this methodology to form meaningful and statistically significant relations between variables within these two databases. However, the results also suggested, at least for the data set analysed, that similar relations were formed regardless of whether rock mass failure was incorporated into the numerical models and roughly independent of the level of data set filtering. Elimination of these two aspects of the methodology will substantially simplify future analyses, although weaker relations will likely be formed. (C) 2004 Elsevier Ltd. All rights reserved.