Discrete Fracture Network modelling applied to caving problems: importance of accounting for data uncertainty and variability Davide Elmo NBK Institute of Mining, University of British Columbia, Vancouver, Canada Erik Eberhardt Department of Earth and Ocean Sciences, University of British Columbia, Vancouver, Canada Doug Stead Simon Fraser University, Vancouver, Canada Steve Rogers Golder Associates, Vancouver, British Columbia, Canada Slide 1
Outline The following presentation is organised in two sections: Introduction to uncertainty and variability applied to discrete fracture network modelling. Uncertainty and variability: implications for DFN based analysis of in-situ fragmentation. Objectives of the presentation: To discuss the importance of data collection for managing data uncertainty and better characterise data variability. To introduce key aspects of DFN modelling. To show examples of how data uncertainty may affect the characterisation of natural fragmentation. Slide 2
Introduction to Uncertainty & Variability Applied to Discrete Fracture Network Modelling Slide 3
Uncertainty in Rock Engineering Design There are various forms of uncertainty in rock engineering: Geological Uncertainty: unpredictability associated with the identification, characterization and interpretation of the site geology and hydrogeology. Parameter Uncertainty: absence of data for key parameters, spatial variability in rock/soil properties, and scale effects (e.g. intact rock vs. rock mass properties). Model Uncertainty: gaps in the scientific theory that is required to make predictions on the basis of causal inference. Human Uncertainty: subjectivity and measurement error, differing professional opinions. Slide 4
Managing Uncertainty and Variability Field Data Rock Mass Characterisation Discrete Fracture Network Continuum Discontinuum Hybrid, Particle Flow Geological Uncertainty Parameter Uncertainty Human Uncertainty Geological Uncertainty Parameter Uncertainty Human Uncertainty + Uncertainty due to processing & transfer of data Geological Uncertainty Parameter Uncertainty Model Uncertainty Human Uncertainty + Uncertainty due to processing & transfer of data Data collection provides the means to manage uncertainty, but it does not eliminate uncertainty Knowledge is often revised by simulations and testing, but computing models have limitations, including poor input data. Slide 5
Introduction to DFN Modelling Discrete fracture network (DFN) models have been increasingly applied to cave mining problems. When used correctly, a DFN approach offers the opportunity to: Maximise the use of fracture data collected from mapping of boreholes/rock exposures. Generate more realistic synthetic rock mass models. Provide estimates of natural fragmentation. Slide 6
Field Data and Primary Properties Used in DFN Models Field Data Source DFN Input Comments/Limitations Orientation Fracture Intensity Fracture Length Fracture Terminations 1D: Geotechnical or geological oriented boreholes 2D: rock exposures (surface/underground) 1D: Geotechnical or geological boreholes 2D: rock exposures (surface/underground) 2D: rock exposures (surface/underground) 2D: rock exposures (surface/underground) Orientation Distributions (set-by-set definitions or overall bootstrap distribution) 1D: P10 - Linear intensity (number of fractures per metre) 2D: P21 - Area intensity, fracture length per unit area 3D: P32 - Volumetric intensity, fracture area per unit volume Fracture Radius Distribution Termination Percentage (T-types terminations) 2D data from rock exposures generally not available at the earlier stages of a project P32 intensity defined based on linear relationships with P10 and P21 P21 data are generally not available at the earlier stages of a project due to lack of suitable rock exposures Need to convert mapped trace length to fracture radius to account for censoring biases Important parameter required to define spatial model and hierarchy of fracture sets Slide 7
DFN Modelling: Important Questions Are current methods for fracture mapping suitable for DFN modelling? What are the critical parameters that we do need to measure and what do we need to change in the current practice? To what extent natural fractures should be sampled for DFN analysis and which limitations are inherently introduced in the analysis by the sampling methods? What would be the impact of the lack or limited field data on the fragmentation/stability results provided by DFN models? Slide 8
DFN Modelling: Uncertainty and Variability The value of the DFN model directly depends on the quality and quantity of available field data. It requires collecting sufficient structural data at the required engineering scale. The lack of appropriate data manifests itself in the form of increased uncertainty and the difficulty to characterize variability in an objective manner. Partial Knowledge Degree of Knowledge Sufficient Knowledge Data Uncertainty Data Variability Slide 9
Scale Dependence of Uncertainty and Variability Geological data is often chartacterised and interpreted in isolation, without reference to the overall geological model. Small scale Fractures in a box DFN model. Mine scale DFN model. The choice of the distributions used to characterize the data often depends on the modelling scale. For instance, small scale fractures and large scale faults could be related according to power law distributions, whereas other forms of distributions (e.g. log normal) could provide a better form of length characterisation for small scale DFN models. Slide 10
Data Characterisation and Perceived Variability Subjective Data Characterisation Objective Data Characterisation 80 25 70 60 20 Frequency 50 40 30 Frequency 15 10 20 10 0 <1 1-3m 3-10m 10-20m >20m Fracture Length Intervals 5 0 Elmo et al. (2015) It is important to consider the perceived variability of the data, which is related to the characterisation process ( parameter & human uncertainty) Slide 11
DFN: Sources of Uncertainty and Variability Data Collection: Fracture length data is either seldom available due to a lack of exposures, or engineers have access to limited length data collected along exploratory drifts. For stability analysis the assumption of fully persistent fractures would generally yield a conservative answer. This does not hold true for fragmentation analysis. Distribution of mapped fracture lengths TRACE LENGTH Data Characterisation: Fracture length does not correspond to fracture radius since trace lengths observed on tunnel walls or bench faces are not the diameter of a circular fracture. Distribution of fracture radius FRACTURE RADIUS Slide 12
DFN Modelling: Managing Uncertainty and Variability Uncertainty Variability Randomness of natural geological processes Complete Ignorance Degree of Knowledge Complete Knowledge Data Collection & Characterisation Need to manage geological, parameter, and human uncertainty DFN modelling Subjective DFN Model Significant assumptions required to define input parameters Objective DFN Model Input parameters selected based on statistical analysis Slide 13
Uncertainty and Variability: Implications for DFN Based Analysis of In-Situ Fragmentation Slide 14
DFN Modelling and In-Situ Fragmentation Volumetric fracture intensity (P32) used as way for mapping zones of poorer rock quality. From a cave perspective, P32 can be used to identify those parts of the cave where residual blocks may be an issue. This information could be then used as a tool for guiding pre-conditioning to aid caving propagation and block mobilisation. Slide 15
Example 1: Uncertainty & DFN Analysis of Fragmentation Field mapping carried out in a room-and-pillar mine. Data source included detailed 2D mapping of rock exposure using a window mapping technique and information was collected for fracture orientation, fracture intensity, fracture length and terminations. To demonstrate the impact of data uncertainty, the analysis was repeated by artificially assuming the data were collected using 1D sampling methods (boreholes) instead of 2D window mapping. Slide 16
Example 1: Uncertainty & DFN Analysis of Fragmentation Data used as input for DFN modelling. Assumed uncertainty level indicated by indicated by colour coding (green = low; orange = moderate; and red = high) Data type Original field data source Spatial Analysis - Orientation Intensity Length Terminations 2D rock exposures 2D rock exposures 2D rock exposures 2D rock exposures Model A Model B Model C Enhanced beacher set-by-set analysis Yes (disaggregate approach) Yes (P21 converted to P32 for each set separately) Yes (for each set separately) Yes (T-types, for each set separately) Enhanced beacher set-by-set analysis Yes (disaggregate approach) Yes (P21 converted to P32 for each set separately) No (same for all data except bedding planes) No (0% assumed) Enhanced beacher Only bedding planes as separate features Yes (bootstrap method) Yes (P21 converted to P32 for all data combined, except bedding planes) No (same for all data except bedding planes) No (0% assumed) Slide 17
Example 1: Size Distribution Curves Uncertainty in the value used for fracture radius (fracture length) has a significant impact on the characterization of the coarser sizes. P70 passing size of 2m 3 for Model A (low uncertainty in the input data). P55 passing size of 2m 3 ; for models B and C with larger uncertainty in the input data. Slide 18
Example 2: Uncertainty & DFN Analysis of Fragmentation Data collected as part of the pre-feasibility study for an undisclosed mine included primarily boreholes data (geotechnical oriented core). Cumulative Fracture Intensity (CFI) plots were generated to establish variation of fracture intensity with depth. Data used as input for DFN modelling. Assumed uncertainty level indicated by indicated by colour coding (green = low; orange = moderate; and red = high) Data type Original field data source Base Case DFN Model Spatial Analysis - No attempt made to identify spatial and temporal relationship between fractures sets Orientation 1D Yes Intensity 1D Yes (P10 converted to P32 for all data combined) Length Very limited 2D rock exposures Very limited data available (one single fracture radius distribution) Terminations Not available No (0% assumed) Slide 19
Example 2: Size Distribution Curves (P32 of 2m 2 /m 3 ) Slide 20
Example 2: Size Distribution Curves (P32 of 4m 2 /m 3 ) Slide 21
Example 2: Size Distribution Curves (P32 of 8m 2 /m 3 ) Slide 22
Example 2: Summary of Size Distribution Curves Cumulative frequency curves for varying P32 plotted for a relative weight calculated over a given range [P32i, P32i+1]. Slide 23
Example 2: Fragmentation Analysis The range of P50th percentile passing decreases as the P32 fracture intensity increases rock mass blockiness increases and smaller blocks are being formed. Decreased variability for increasing P32 fracture intensity. Slide 24
Conclusions The DFN method relies on quantifiable field parameters (fracture orientation, intensity, length and terminations) and if used correctly it provides realistic geometric models of fracture networks. DFN methods can provide an alternative and effective approach for studying rock mass fragmentation. There is the need to include more appropriate set of guidelines bridging the gap between the current methods for collection of geotechnical data and the requirements imposed by the use of new technologies. Development of field methods for mapping of fracture data more suitable to DFN modelling, and developing techniques to integrate deterministic fractures within stochastic models. Accounting for spatial variation and structural geological models. There are known knowns. These are things we know that we know. There are known unknowns. These are things that we know we don't know. but there are also unknown unknowns. These are things we don't know we don't know. Slide 25
Acknowledgments This research is being supported by NSERC as part of a Collaborative Research and Development grant between UBC and Golder Associates Ltd. (NSERC CRDPJ 453374-13). Natural Sciences and Engineering Research Council of Canada Slide 26