Research On Statistical Parameters Of Rock Mass Fracture System In High-level Radioactive Waste Repository Beishan Candidate Site

The small-scale fractures which are ubiquitously distributed in fractured rock mass often constitute the flow and migration path of underground fluid and radionuclide, thus have significant impacts on the seepage properties and stability of rock mass. This thesis mainly focuses on the small-scale fractures in Gansu Beishan Jijicao, which are determined to be one of the pre-selected areas for Chinese high-level radioactive waste repository. The geometrical parameters of fractures are statistically analyzed based on measured samples by using probability theory and statistical methods. The final results are presented as the statistical characteristics of fracture geometrical parameters. The main conclusions are as follows:(1)The traditional methods of delineating the major fracture sets are prone to be subjective and are difficult to define the sets unambiguously when dealing with large quantity of data. Three objective clustering algorithms, namely the Shanley-Mahtab method, the spectral clustering method and the fuzzy cluster algorithm, are adopted to assist delineating regional and local sets. The applicability of each method is analyzed and compared. The Shanley-Mahtab method is based on an objective function, which consists of sum of distances between data within single set and between set centers. The method of assigning non-dense point to dense point set proposed by this algorithm needs further investigation. A new assigning criterion is proposed which assign non-dense point to the set containing the nearest set center. The spectral clustering method could not provide in advance an approach of determining the optimized set number, and the determination of scaling parameter is ambiguous. The fuzzy cluster algorithm is capable of determining the optimized set number based on specific indexes, and of dealing with different kinds of data. An improvement for this algorithm is proposed based on a particular threshold. The improved algorithm is capable of finding out the discrete data which apparently does not belong to any sets. Using the proposed algorithm improves the accuracy of delineating fracture sets.(2)Two kinds of stereological relationships (Zhang’s moment relationship and Tonon’s closed-form relationship) developed for statistical inference of the fracture size (Baecher disk model) from trace length are investigated in detail using Monte-Carlo simulation technique. The reliability and sensibility of these relationships are also thoroughly checked out. Results show that the probability distribution forms of trace length and the diameter of disk are not consistent. The optimized diameter distribution calculated from Zhang’s formulae is basically the same as that of trace length, whereas the diameter distribution calculated from Tonon’s formulae could not be categorized into any known theoretical distribution. The minimum diameter is also found using Tonon’s method. The estimation of diameter expectation determined by these formulae fluctuates around the true value. The expectation of trace length is equal or larger than that of diameter. The estimation accuracy which is convinced to be influenced by the quality of trace length samples can be estimated by the Bootstrap method. Parameter sensitivity analyses show that the impacts of each parameter on both kinds of formulae are intrinsically different. The estimations of diameter expectation, variance, and probability distribution by these formulae are proved to be approximate to some extent. A smaller expectation value compared with the true value is generally determined. A relatively accurate value with respect to the diameter variance is usually determined by Zhang’s formulae. The stability of Tonon’s formulae is thought to be better than Zhang’s formulae with respect to the diameter expectation. Finally a diameter sample which follows the Tonon’s calculated distribution can be generated using Veneziano’s algorithm.(3)The influences of point patterns on trace length estimation are investigated using different point pattern models. Results show that the homogeneity is the most important factor of affecting the accuracy of trace length estimation using convex window. If the points are inhomogeneously distributed in the window, the estimation will gradually be smaller than the true value with the increased window size. The estimation will reach the true value if the point pattern is completely homogeneous, regardless of the correlation between points. The Pearson chi-square test provides an effective approach of point pattern homogeneity inference. Different kinds of descriptive functions can be defined to further investigate the correlation between points. The midpoint distribution of traces on outcrops can be simulated using different point pattern models.(4)Two regional major sets are found in Jijicao based on in-situ observations and clustering methods. The strikes of these two sets are NNE-NEE and NWW-NW, respectively. A local set whose strike is approximately NS is also identified. The mean dip angles of fracture sets in A and B investigation areas are less than 70°, whereas the mean dip angles in other areas are mainly between 70°-80°. The trace lengths of all major sets can generally be represented by a log-normal distribution, except for a small part which can be represented by a gamma distribution. The mean-standard deviation and mean-coefficient of variance, as well as trace length-orientation relationships, are identified to be correlative to some extent. For example, the sets with mean trace length over 3.5m tend to have a dip direction between 200°-360°, whereas the sets with dip angle over 70°basically own a mean trace length less than 3.0m.(5)A complete fracture density-depth relationship curve has been drawn based on the geological maps from cores of BS16 borehole. Results show that fracture density between 300-465m below surface is apparently larger than other parts of cores. Geostatistical analyses imply that the fracture density distribution along cores exhibits spatial correlation characteristics. Fractures with trace length over 10m are ubiquitous lineaments in Jijicao, which should be treated as independent objects according to their unique properties.