Trans. Nonferrous Met. Soc.

China 29(2019) 2392−2399

Implicit modeling of complex orebody with constraints of geological rules

De-yun ZHONG, Li-guan WANG, Lin BI, Ming-tao JIA

School of Resources and Safety Engineering, Central South University, Changsha 410083, China
Received 19 September 2018; accepted 27 August 2019

Abstract: To dynamically update the shape of orebody according to the knowledge of a structural geologist’s insight, an approach of
orebody implicit modeling from raw drillhole data using the generalized radial basis function interpolant was presented. A variety of
constraint rules, including geology trend line, geology constraint line, geology trend surface, geology constraint surface and
anisotropy, which can be converted into interpolation constraints, were developed to dynamically control the geology trends.
Combined with the interactive tools of constraint rules, this method can avoid the shortcomings of the explicit modeling method
based on the contour stitching, such as poor model quality, and is difficult to update dynamically, and simplify the modeling process
of orebody. The results of numerical experiments show that the 3D ore body model can be reconstructed quickly, accurately and
dynamically by the implicit modeling method.
Key words: three-dimensional geomodeling; implicit modeling; radial basis function; structural anisotropy; geological rules

dynamic updating [2−4].

1 Introduction In the past two decades, a variety of implicit
function interpolation methods including the discrete
Due to the limitation of geological conditions and smooth interpolation (DSI) method [5], the (radial basis
exploration technology, complete and regular geological function, RBF) RBF-based method [6,7], the Hermite
data cannot be obtained in geological exploration, and RBF (HRBF) method [8] and the moving least squares
there is great uncertainty in modeling sparse data. The (MLS) method [9] have been developed. These methods
method of interpreting and 3D modeling of the complex are mainly applied to dense point cloud data of
orebody by artificial experience and human-computer three-dimensional laser scanning and are difficult to fit
interaction is inefficient, arbitrary and subjective, and sparse and uneven sampling data of geological
difficult to update the orebody model. Therefore, it is of exploration with various geological rules and constraints.
great significance to dynamically reconstruct the Among them, RBF is a widely used interpolation method
three-dimensional orebody models close to the original with complete theoretical support, which has been
shape from the raw data and geological interpretation of integrated into Leapfrog Geo software [10] and widely
multi-source and sparse sampling. used in geological modeling. CARR et al [11] solved the
The methods of geology modeling can be divided problem of fast interpolation of large-scale point cloud
into two kinds, termed explicit modeling and implicit data by introducing the fast multipole method (FMM).
modeling [1]. The geometrical quality of the model built And the Leapfrog Geo software uses the similar method,
by the traditional explicit modeling method is not high FastRBF method [12,13], to interpolate large geological
enough, and there are a lot of degenerate triangles, which data sets efficiently. JONES and CHEN [14] transformed
are prone to reveal defects such as opening and the contours into the three-dimensional distance field
self-intersection. Implicit modeling recovers the using the distance transformation by the contours of the
three-dimensional orebody model through surface orebody and reconstructed the model by iso-surface
reconstruction using an implicit function. Compared with extraction. GUO et al [15] conducted a preliminary study
the traditional explicit modeling method, the implicit on geological interfaces modeling using HRBF method
modeling method has the advantages of high mesh with section constraints.
quality, repeatable process, global uncertainty and The traditional RBF methods are based on domain

Corresponding author: Lin BI; Tel: +86-731-88877665; E-mail: [email protected]

DOI: 10.1016/S1003-6326(19)65145-9
 De-yun ZHONG, et al/Trans. Nonferrous Met. Soc. China 29(2019) 2392−2399 2393
constrains, which are difficult to handle complex N
s ( x ) =  ω j λ jx ′Φ ( x , x ′) + p( x ), s ∈ H (1)
geological structural trends. More and more attention has
j =1
been paid to the study of the RBF-based methods with
generalized constraints recently. Several extended where λ jx ′ is continuous linear functional acting on a
interpolants based on the theory of Hermite−Birkhoff usual radial basis function Φ(x, x′) viewed as a function
interpolation with radial basis functions are developed to of x′=(x′, y′, z′). When conditionally positive definite
interpolate anisotropic and first-order Hermite data functions are used, it is often required to construct
(points with normals), such as anisotropic RBF low-order polynomials p(x) to ensure that the function
(ARBF) [16,17], generalized RBF (GRBF) [18] and converges.
generalized HRBF (GHRBF) [19,20]. More recently, a Four types of general constraints [18−22] can be
generalized interpolation framework using RBF from constructed by the different operations of the continuous
scattered multivariate structural data was presented by linear functional and the distance transformation, as
HILLIER et al [18] to generate continuous geological shown in Fig. 1.
surfaces. However, each interpolation method has its (1) Domain constraints: The domain constraints are
own advantages and drawbacks, and is applicable to locations at which we require the implicit function to
specific conditions. It is necessary to study interpolation take on the specific values. The given μ scattered data
methods that meet the constraints of specific geological points { xi , f ( xi )}iμ=1 satisfy
rules in different situations to ensure the accuracy of f(xi)=fi, i=1, 2, …, μ (2)
geological model. where fi is the function value of the geological domain.
The implicit modeling method is based on the The domain constraint can well control the internal
reconstruction of an implicit function from the drillhole and external relations of the geological domain and is the
datasets. The geology space constructed by the most basic constraint to define the geological interface.
combination of drilling samples is transformed into a According to the value of the function, the domain
signed distance field. The result of complex orebody constraints can be divided into three different types:
model is expressed as a mathematical function, and the on-surface constraints (f(xi)=0), interior constraints
implicit surface is represented as a zero-level set of the (f(xi)<0) and exterior constraints (f(xi)>0).
function. To consider various types of geological rules, (2) Gradient constraints: The gradient constraints
the GRBF interpolant was used as the implicit function. refer to the evaluation of some derivative at specific
One of the advantages of this approach is that it can locations. The given σ scattered data points
generate geometrically valid 3D ore body models { xi , ∇f ( xi )}iμ=+μσ+1 satisfy
directly from the raw drillhole data with or without the
constraints of interpretation. ∇f ( xi ) =ni, i=μ+1, μ+2, …, μ+σ (3)
where ni is the unit normal vector of the geological
2 Mathematical framework domain.
The sign of the gradient constraint represents
The GRBF interpolant is built upon the theory of different shape constraints. The normal direction of ni
generalized Hermite−Birkhoff interpolation [18−22] with points to the exterior of the shape, then the opposite
radial basis functions. We will first review the relevant direction points to the interior of the shape. The normal
theory loosely. constraints can be used to construct the trend surface
Given the known data (xi, λif ), i=1, 2, …, N, xi∈Rn, constraint in the domain.
where λi is a linearly independent set of continuous linear (3) Tangent constraints: The tangent constraints
functionals and f is some (smooth) data function, the refer to the sampling orientations tangent to the domain
generalized Hermite interpolation problem tries to at specific locations. The given τ scattered data
construct the interpolant s(x) satisfying λis(x)=λif as points { xi , ∇f ( xi ), ti }iμ=+μσ++στ1 satisfy

Fig. 1 Various types of interpolation constraints used for implicit modeling

2394 De-yun ZHONG, et al/Trans. Nonferrous Met. Soc. China 29(2019) 2392−2399
 ∇f ( xi ), ti  =0, i=μ+σ+1, μ+σ+2, …, μ+σ+τ (4) function can be formed by interpolating these constraints
and solving the corresponding linear system.
where ti is the unit tangent vector of the geological
During discretization, the distance values of
domain.
sampling points are initially computed via the distance
The tangent constraint does not have the polarity of
along the direction of the drilling trajectory. The
the orientation, but only affects the direction of adjacent
on-surface constraints are constructed at the two
points and changes the local curvature, so it is well suited
endpoints of the sample segments. The off-surface
for constructing trend line constraints in the domain.
constraints (including the exterior and interior constraints)
(4) Anisotropy constraints: The anisotropy constraints
are constructed in the sample and non-sample segments.
refer to the evaluation of the anisotropy distance by the
Along the drilling track, the sample segments are
distance transformation at specific locations. The given
sampled discretely from both ends to the middle to
N scattered data points { xi , λi f }iN=1 satisfy
construct the interior constraints according to a given
ΦT(x, x′)=Φ(||x−x′||T) (5) sampling interval. The initial function values of the
where interior constraints are computed according to the sample
distance to the nearest on-surface points. To ensure the
||x−x′||T =||(x−x′)·T || (6)
reliability of the solution, at least one sample point
and T is the anisotropic distance matrix which can be should be sampled in the sample segments. Similarly, the
constructed via the affine matrix transformation. The non-sample segments are sampled discretely from both
anisotropy of its three principal directions represented as ends to the middle to construct the exterior constraints
an ellipsoid. according to a given sampling interval. The initial
The main difference between the isotropic and function values of the exterior constraints are computed
anisotropic GRBF is the computation of the distance according to the non-sample distance to the nearest
field. Similar to the isotropic GRBF but with a modified on-surface points. As an example of discretization in
distance metric by a distance transformation, the Fig. 2, the on-surface constraints with zero values are
anisotropic general radial basis function (AGRBF) red, the off-surface point constraints with positive values
interpolant has the following form: are cyan and the off-surface point constraints with
N negative values are blue. After the discretization of
s ( x ) =  ω ′j λ jx ′Φ T ( x , x ′) + p( x ), s ∈ H , N=μ+σ+τ drillhole data, the on-surface and the off-surface
j =1
constraints are added to the interpolation equation to
(7)
solve the implicit function of the orebody model.
where the new set of weights ω ′j are different from ωj
computed from the isotropic GRBF interpolant.
3.2 Distance field correction
The interpolation equation can be obtained by
During the construction of the implicit model, a
acting the continuous linear functions on the radial basis
signed distance field should be formed in the process of
function. Similarly, the anisotropic interpolation equation
distance computation, and the implicit function can be
can be obtained by acting λ jx ′ on anisotropic kernel
regarded as a signed distance field function. Therefore,
ΦT ( x , x ′) . Similar to the RBF interpolant, using the the process of orebody implicit modeling can be
anisotropic GRBF interpolant s(x) and the general
regarded as the process of constructing a signed distance
constraints, the unknown weight coefficients can be
field in line with the trend of the geological domain
determined by solving a linear system such as Ax=b.
using sampling points.
To distinguish the internal and external field of the
3 Implicit modeling
orebody model, the relationship between the implicit
function value and the implicit surface can be expressed
3.1 Discrete drillhole
as
To construct the spatial interpolation conditions of
an implicit function, it is necessary to discretize the  x | f ( x ) = 0, x ∈ R 3 , on the surface
drillhole data sampled from geological exploration. The 
 x | f ( x ) = +dist( x , x ′) > 0, x ∈ R 3 ,
discretization of drillhole data refers to the process of 
obtaining the sampling points of sample segments and  exterior of the surface (8)
 3
non-sample segments based on the grade combination of  x | f ( x ) = −dist( x, x ′) < 0, x ∈ R ,
drillhole data. The geological domain containing the  interior of the surface

whole drillholes is viewed as a non-Euclidean distance
field. The sampling points of sample segments and where x=(x, y, z) is a three-dimensional sampling point,
non-sample segments can be transformed into on-surface and dist(x, x′) is the nearest distance from x to the closest
constraints and off-surface constraints. Then, the implicit point x′ on the surface. As shown in Fig. 3, the colors
 De-yun ZHONG, et al/Trans. Nonferrous Met. Soc. China 29(2019) 2392−2399 2395

Fig. 2 Discretization procedure of drillholes: (a) Grade combination; (b) Sampling point

Fig. 3 Signed distance field of drillholes: (a) Discretized points of drillholes; (b) Signed distance field

show different field values, for which blue denotes the The geology trend line can be used as an orientation
maximum negative value and red represents the constraint to guide the model extension trend of nearby
maximum positive value. domains. Taking the direction of the trend line as the
To construct a geological distance more consistent tangential direction, the trend line is sampled at a certain
with the trend distribution of the drilling trajectory, the sampling interval to construct tangent constraints. By
distance values of two points are computed as the length specifying trend lines in sparse regions, the reconstructed
of the drilling trajectory instead of the Euclidean model has a tendency to extend along the trend line.
distance. In the process of discrete sampling, the cross-
distributed drillholes tend to generate ambiguity
constraints with abnormal values, so the distance should
be modified according to the distribution of
mineralization field. To verify the distance of sampling
points, an iterative closest point correction (ICPC)
algorithm [23] is used to correct the distance, which
ensures the distance field of the implicit function
complying with the distribution characteristics of the
mineralization field.

4 Constraint with trend

4.1 Constraint rule Fig. 4 Four types of constraint rules constructed by general
The method of implicit modeling still requires constraints: (a) Geology trend line; (b) Geology constraint line;
the guidance from the structural geologists’ expert (c) Geology trend surface; (d) Geology constraint surface
knowledge, including the designation of multiple
geological constraints. Based on the general constraints (2) Geology constraint line
of the GRBF interpolant, four types of constraint rules The geology constraint line is discretized according
can be obtained, as shown in Fig. 4. to the specified sampling interval, and the discrete
(1) Geology trend line constraint points are added to the interpolation equation.
2396 De-yun ZHONG, et al/Trans. Nonferrous Met. Soc. China 29(2019) 2392−2399
By reconstructing the implicit surface, the orebody contour points along their normal direction [11]. If the
model can satisfy the model boundary represented by the cutting plane of the section orthogonally intersects with
constraint line. The geological constraint line can well the local surface of the orebody model, the normal vector
control the extrapolation boundary of the automatic of the boundary surface at pi can be computed as
interpolation model and change the local continuity trend t pi × c pi
of the model according to the geological rules. n pi = Sign pi × (9)
|| t pi × c pi ||
(3) Geology trend surface
The geology trend surface is constructed by the where t pi is the tangent vector of the contour at pi,
direction sampling (e.g., gradient constraints), but the c pi is the normal vector of the cutting plane, and Sign pi
constraints with specific distance values (e.g., domain is the sign of the normal vector determined by the side of
constraints) are not constructed. The stratified the contour. For other cases, the normals of sections are
resampling method is used to resample the trend surface, estimated via the method proposed by HECKEL
which can control the uniformity of the resample et al [24]. Moreover, the user is allowed to interactively
placement and the minimum distance between samples. add, remove or edit the local normals at discrete points.
(4) Geology constraint surface Then, the unknown normals can be automatically
The geology constraint surface represents a local interpolated by the given ones.
surface modeled by other methods, which can be used to The additional section constraints can be used to
recover the local implicit surface at the sparse locations guide the topological connection between drilling holes
with incomplete sampling. There are several ways that and form the orebody model to satisfy the mineralization
the constraint surface can be converted to general trend, as shown in Fig. 5. There are many section
constraints. One of the ways is to construct both gradient constraints shown as green polylines in Fig. 5(a) and
constraints and domain constraints at the sampling geology constraint lines shown as red polylines in
points. Fig. 5(c).

4.2 Section constraints 5 Fast modeling

To convert the additional section constraints
interpreted by structural geologists into the GRBF The implicit modeling based on GRBF can be
interpolation constraints, the contours should be finally transformed into the solution of the linear system
discretized to form domain constraints. To ensure the Ax=b. It consists of two processes, spatial interpolation
unique solution of the interpolation equation, the of an implicit function based on interpolation constraints
contours of the interpreted sections should be and three-dimensional surface reconstruction based on
preprocessed to avoid abnormal normal constraints or point evaluations. To ensure the rapid dynamic updating
contradictory domain constraints. Firstly, the redundant of the implicit model, in addition to the fast solution of
points in given tolerance should be removed, and the the interpolation equation, the implementation of the fast
abnormal segments in the angle tolerance of refraction evaluation of implicit function is also required.
should be cleared. Secondly, in a certain tolerance range, To improve the speed of solving large-scale linear
the sections are preprocessed to reduce the curvature of equations, an iterative method, GMRES, proposed by
contours, to build a smoother implicit model. Lastly, the SAAD and SCHULTZ [25], is recommended for
normals of the section will be estimated to construct implementation. This method belongs to Krylov
gradient constrains or off-surface constraints. The subspace iteration, which uses Arnoldi iteration to solve
off-surface constrains are formed by offsetting the the minimum residuals in the subspace to approximate

Fig. 5 Implicit modeling from raw drillhole data: (a) Drillholes and geology constraints; (b) Result without manual constraints;
(c) Result with manual constraints
 De-yun ZHONG, et al/Trans. Nonferrous Met. Soc. China 29(2019) 2392−2399 2397
the solution of the linear system. For the problem of this method, the results with the traditional radial basis
complex orebody modeling, it is a very effective method function without constraints were compared.
to solve large asymmetric linear equations, which can For sparse drillhole data with large intervals, the
greatly improve the speed of solving large-scale reconstructions without constraints are likely to produce
interpolation equations. discontinuities. Therefore, the modeling method still
The Marching Cubes method is often used for requires the knowledge of a structural geologist’s insight
surface reconstruction. The Marching Cubes algorithm and this input is made in the form of constraint rules,
has some ambiguities when extracting the triangular additional sections, and structural anisotropy. The
facets of the cubes in the spatial regular data field. To constraint rules were constructed to constrain the shape
resolve the ambiguity, the hyperbolic asymptote method of the implicit surface according to the actual geological
proposed by NIELSON and HAMANN [26] was conditions of the drillhole data and the trend of the
implemented. To make full use of the valid cubes in the mineralization domain. Figure 6(d) shows the orebody
evaluation process, the surface- following method based modeling results directly based on the drillhole data
on the Marching Cubes algorithm was implemented. The using the traditional radial basis function. It can
method constructs the initial voxel seed points near the automatically model without additional constraints. To
isosurface, using greedy voxel growth rules to track the make the modeling result more consistent with the
isosurface by searching the close cubes. It does not extension trend of the orebody, the implicit model
evaluate cubes over the whole volume, which can greatly can be dynamically modified by adding constraint lines
speed up the process of surface reconstruction. and trend surface constraints (Fig. 6(c)). The results
(Fig. 6(e)) of implicit modeling of orebody meet the
6 Results interpretation requirements of structural geologists.
The ore grade shells, geology interfaces or
The anisotropic GRBF method was implemented structural trends will be represented by the implicit
and tested on several non-trivial geological examples. functions, which can effectively handle the problems
These examples contain series of drillhole datasets in encountered in the traditional explicit modeling. Figure 7
sparse data environments. To validate the performance of shows the dynamic grade shell models established by

Fig. 6 Processes of implicit modeling from raw drillhole data: (a) Grade combination; (b) Discretization of non-sample and sample
segments; (c) User-defined constraint rules; (d) Result without manual constraints; (e) Result with manual constraints; (f) Compared
result
2398 De-yun ZHONG, et al/Trans. Nonferrous Met. Soc. China 29(2019) 2392−2399

Fig. 7 Dynamic grade shell models with different cut-off grades: (a) 2% and 3%; (b) 1% and 3%

Table 1 Running time of solution and reconstruction of algorithm on several examples

Time/s
Model Constraints Resolution Triangles
Solution MC PMC SF
Fig. 6(d) 760 4 17204 1.25 17.87 5.28 1.87
Fig. 6(e) 881 4 17792 2.69 21.83 7.23 2.13
Fig. 7 866 10 38016 2.18 223.52 75.72 4.14

raw drillhole data under different cut-off grades, which

can well reflect the spatial distribution trend of the 7 Conclusions
orebody model under different economic indicators. At
the same time, the orebody model constructed by implicit (1) Based on the anisotropic GRBF interpolant, an
modeling has the advantages of high quality and implicit modeling method of complex orebody
smoothness. It is easy to represent orebody models with conforming to geological rules is proposed, which can
complex topology and is convenient to perform Boolean convert the geological constraints into different
operations. interpolation constraints and constraint rules. It is not
The performance of the anisotropic GRBF method based on the grade interpolation but uses the distance
mainly depends on the number of constraints and the size field to calculate the implicit function by geological
of resolution. The algorithm was implemented in C++ constraints, so that it can be applied to the structural
language and tested on a Windows 64-bit PC with 3.20 modeling of the orebody. The results show that the
GHz Intel(R) Core(TM) i5-3470 and 4GB RAM. Table 1 reconstructed implicit models are guaranteed to be
lists the timings of the solution and reconstruction stages smooth, continuous and closed geology surface without
of the algorithm on these examples. mistakes such as intersections.
The implementation runs from a few seconds to less (2) As the RBF-based method has good
than one minute for the examples. The running time is extrapolation capability, it is very suitable for
dominated by the solution of large-scale linear systems constraint-based modeling of sparse drillhole data.
in the interpolation stage and the evaluation of sampling Moreover, it is known that the orebody model usually
grids in the reconstruction stage. To test the performance has strong local continuity and extension trend along the
of fast reconstruction using the improved method, the direction of the mineralization area.
running efficiency of multiple data sets was compared (3) Based on the anisotropy constraints, the
before and after the improvement. Table 1 shows that the anisotropic orebody model can be restrained by
surface following (SF) algorithm used has faster constructing geological trends in different directions
reconstruction efficiency than the traditional Marching according to the manual interpretation requirements of
structural geologists.
Cube (MC) and parallel Marching Cube (PMC)
extraction method. Moreover, as the size of the solution
equations becomes larger or the reconstruction resolution
References
decreases, the performance of the improved algorithm is
[1] JESSELL M, AILLÈRES L, de KEMP E, LINDSAY M,
more obvious. WELLMANN F. Next generation three-dimensional geologic
 De-yun ZHONG, et al/Trans. Nonferrous Met. Soc. China 29(2019) 2392−2399 2399
modeling and inversion [J]. Economic Geology, 2014, 18: 261−272. surfaces from contour data [J]. Computer Graphics Forum, 1994,
[2] COWAN J, BEATSON R, FRIGHT W R, MCLENNAN T J, 13(3): 75−84.
MITCHELL T J. Rapid geological modeling [C]//Applied Structural [15] GUO Jia-teng, WU Li-xin, ZHOU Wen-hui, LI Chao-lin, LI
Geology for Mineral Exploration and Mining International Feng-dan. Section-constrained local geological interface dynamic
Symposium. Kalgoorlie: Australian Institute of Geologists, 2002: updating method based on the HRBF surface [J]. Journal of
23−25. Structural Geology, 2018, 107: 64−72.
[3] COWAN E, SPRAGG K, EVERITT M. Wireframe-free geological [16] CASCIOLA G, LAZZARO D, MONTEFUSCO L B, MORIGI S.
modelling–An oxymoron or a value proposition [C]//Eighth Shape preserving surface reconstruction using locally anisotropic
International Mining Geology Conference. Melbourne: The
radial basis function interpolants [J]. Computers & Mathematics with
Australasian Institute of Mining and Metallurgy, 2011: 247−260.
Applications, 2006, 51(8): 1185−1198.
[4] MCLENNAN J A, DEUTSCH C V. Implicit boundary modeling
[17] CASCIOLA G, MONTEFUSCO L B, MORIGI S. Edge-driven
(boundsim) [R]. Edmonton: University of Alberta, 2006.
image interpolation using adaptive anisotropic radial basis functions
[5] FRANK T, TERTOIS A L, MALLET J L. 3D-reconstruction of
[J]. Journal of Mathematical Imaging and Vision, 2010, 36(2):
complex geological interfaces from irregularly distributed and noisy
125−139.
point data [J]. Computers & Geosciences, 2007, 33(7): 932−943.
[6] CUOMO S, GALLETTI A, GIUNTA G, MARCELLINO L. [18] HILLIER M J, SCHETSELAAR E M, de KEMP E A, PERRON G.
Reconstruction of implicit curves and surfaces via RBF interpolation Three-dimensional modelling of geological surfaces using
[J]. Applied Numerical Mathematics, 2017, 116: 157−171. generalized interpolation with radial basis functions [J].
[7] SKALA V. RBF Interpolation with CSRBF of large data sets [J]. Mathematical Geosciences, 2014, 46(8): 931−953.
Procedia Computer Science, 2017, 108: 2433−2437. [19] ETTL S, KAMINSKI J, HÄUSLER G. Generalized Hermite
[8] MACÊDO I, GOIS J P, VELHO L. Hermite interpolation of implicit interpolation with radial basis functions considering only gradient
surfaces with radial basis functions[C]//Proceedings of 2009 XXII data [C]//Curve and Surface Fitting: Avignon 2006. Avignon:
Brazilian Symposium on Computer Graphics and Image Processing. Nashboo Press, 2007: 141−149.
Rio de Janiero: IEEE, 2009: 1−8. [20] GOIS J P, TREVISAN D F, BATAGELO H C, MACÊDO I.
[9] FLEISHMAN S, COHEN-OR D, SILVA C T. Robust moving Generalized Hermitian radial basis functions implicits from
least-squares fitting with sharp features [J]. ACM Transactions on polygonal mesh constraints [J]. The Visual Computer, 2013, 29(6−8):
Graphics (TOG), 2005, 24(3): 544−552. 651−661.
[10] BIRCH C. New systems for geological modelling-black box or best [21] FASSHAUER G E. Meshfree approximation methods with
practice? [J]. Journal of the Southern African Institute of Mining and MATLAB [M]. London: World Scientific, 2007.
Metallurgy, 2014, 114(12): 993−1000. [22] WENDLAND H. Scattered data approximation [M]. Cambridge:
[11] CARR J C, BEATSON R K, CHERRIE J B, MITCHELL T J,
Cambridge University Press, 2004.
FRIGHT W R. Reconstruction and representation of 3D objects with
[23] ZHONG De-yun, WANG Li-guan, JIA Ming-tao, ZHANG Ju.
radial basis functions [C]//Proceedings of the 28th Annual
Orebody modeling from non-parallel cross sections with geometry
Conference on Computer Graphics and Interactive Techniques. New
constraints [J]. Minerals, 2019, 9(4): 1−17.
York: ACM, 2001: 67−76.
[24] HECKEL F, KONRAD O, HAHN H K, PEITGEN H O. Interactive
[12] COWAN E J, BEATSON R K, ROSS H J, FRIGHT W R,
3D medical image segmentation with energy-minimizing implicit
MCLENNAN T J. Practical implicit geological modeling
[C]//Proceedings of Fifth International Mining Geology Conference. functions [J]. Computers & Graphics, 2011, 35(2): 275−287.
Victoria: Australian Institute of Mining and Metallurgy, 2003: 89−99. [25] SAAD Y, SCHULTZ M H. GMRES: A generalized minimal residual
[13] STOCH B, ANTHONISSEN C, MCCALL M, BASSON I J, algorithm for solving nonsymmetric linear systems [J]. Siam Journal
DEACON J. 3D implicit modeling of the Sishen mine: New on Scientific & Statistical Computing, 1986, 7(3): 856−869.
resolution of the geometry and origin of Fe mineralization [J]. [26] NIELSON G M, HAMANN B. The asymptotic decider: resolving the
Mineralium Deposita, 2018, 53(6): 835−853. ambiguity in marching cubes [C]//Proceeding Visualization '91.
[14] JONES M W, CHEN M. A new approach to the construction of California: IEE, 1991: 83−91.

融合地质规则约束的复杂矿体隐式建模方法
钟德云，王李管，毕 林，贾明涛

中南大学 资源与安全工程学院，长沙 410083

摘 要：为了根据结构地质学家的经验对矿体形状进行动态控制，基于广义径向基函数插值方法提出一种可以直
接从原始钻孔数据进行矿体隐式建模的方法。发展多种约束规则来动态控制地质趋势，包括地质趋势线、地质约
束线、地质趋势面、地质约束面和各向异性等，均可以转化为插值约束条件。结合约束规则的交互工具，该方法
可以避免基于轮廓线拼接的显式建模方法存在的模型质量差、难以动态更新等缺点，对具有几何边界约束的矿体
结构模型进行建模，简化矿体建模过程。数值试验结果表明，该隐式建模方法可以快速、准确、动态地建立三维
矿体模型。
关键词：三维地质建模; 隐式建模; 径向基函数; 结构各向异性; 地质规则
(Edited by Xiang-qun LI)