Data-driven optimization approach for mass-spring models parametrization based on isogeometric analysis
Introduction
The mechanical behavior of elastic objects can be described by continuum elasticity theories that model how the objects deform under applied forces [1]. In the case of computer graphics applications, the search for models that enable real-time dynamic simulation of deformable objects is an active area of research [2], [3], [4], [5], [6], [7]. To perform computational simulation of the deformable body, the 3D/2D object's geometry is represented in computer aided design (CAD) files that must be translated into geometries suitable to be meshed in order to generate a compatible description for finite element (FEM) analysis [8]. Hence, as pointed out in [9], the geometric modeling and FEM communities realized the need to unify CAD and FEM descriptions which nowadays is known as Isogeometric Analysis (IGA) [10]. The idea of IGA analysis is that the functions used for the geometry description in CAD are adopted by the solution field representation (isoparametric concept).
In the specific case of non-uniform Rational B-Splines (NURBS) based IGA, the geometric description and field variables are represented via the same NURBS basis functions. Therefore, the control points associated with the NURBS basis function define the geometry and the degrees-of-freedom of field variables are the coefficients of the basis functions, i.e. control variables. The NURBS-based IGA framework [10] and extensions have been applied for vibrations, wave propagation, fluids, fluid–structure interaction, heat conduction, and elasticity problems [11], [12].
Other possibility for elastic objects simulation is to apply discrete approaches, based on Mass Spring Models (MSM), that represent the object's geometry by a mesh whose nodes are treated like mass points and edges act like springs connecting adjacent nodes. MSM approaches are preferable for real time applications because they are in general faster than the continuous ones besides being simple to implement [13]. Therefore, MSM techniques have been used to model deformable objects [2], for woven cloth simulation [13], [14], soft organic tissues, like muscles, face and abdomen in virtual surgery applications [15], [16], [17], [18], [19], [20].
Notwithstanding of the MSM advantages, it is known that methods that are based on the continuum mechanics and FEM are, in general, more realistic than their MSM counterparts [21], [22], [23]. That is the motivation to develop a general physically based or systematic method to determine the mesh topology or MSM parameters from the continuum set up. However, to the best of our knowledge, works in MSM derivation from FEM models have been restricted to linear interpolation functions [24], [25], [26]. On the other hand, NURBS based IGA enables accuracy of geometry representation and deals with higher-order polynomial basis functions [9]. These features can be exploited to improve the quality of MSM derivation processes.
Hence, in this paper, we propose a new derivation method for setting NMSM parameters (mass, spring stiffness coefficient and damping constants) using a data-driven strategy with a NURBS-based IGA model acting as continuum counterpart model. Due to this new approach, the mass spring model inherits NURBS parameterization in which the control points are treated like mass points linked by springs. For this reason, we have called our new scheme as NURBS Mass-Spring Model, abbreviated to NMSM. Generally speaking, this method computes the spring coefficients by solving an optimization problem based on the static equilibrium of NURBS-based IGA model and NMSM; calculates particles mass following [27], [28] and derives the damping constant using a technique based on [29].
The framework proposed in this work determines the NMSM parameters so that it behaves like IGA (its continuum counterpart), eliminating the trial-and-error approach for the parameters tuning. The NMSM adds relevant contributions in the state-of-the art in MSM parameterization approaches: (a) It allows to increase the geometric accuracy in the representation of elastic objects, since we have a mass-spring model based on NURBS; (b) it can simulate deformations of curved shapes by exploring NURBS refinement techniques or manipulation of higher order polynomials; (c) the method combines the simplicity and computational efficiency of the MSM with NURBS to achieve deformation simulation with realism similarly to the one offered by NURBS-based IGA model but with reduced computational cost.
We validate the proposed method by several experiments in the context of 2D and 3D IGA models. The obtained results show that the derivation methodology has the lowest percentage of error when compared with related ones. Also, we verify the robustness of the algorithm against different discretizations and NURBS geometry, optimization problem setups and material properties. Respect to the polynomial order, the method achieves outstanding results for first and second orders. However, we notice a degradation in the ability of the derived NMSM to mimic the IGA model when increasing the NURBS polynomial order.
The paper is organized as follows. Section 2 describes the related literature. The NMSM approach is presented in Section 3. Experimental results are presented in Section 4. Next, Section 5 discusses the experimental results and related features of NMSM. Then, in Section 6, the conclusions and future works are presented. Appendix A offer some details about the implementation, which was carried out in MATLAB, software architecture, and parameters/options used in the simulated annealing function. Finally, Appendix B presents some background in NURBS based IGA models.
Section snippets
Related works
Two categories of methods can be identified in the estimation of the parameters for MSM in order to guarantee a realistic behavior: data-driven and model driven [22]. The first category is composed by methods that use a minimization procedure to find the model that shows the closest behavior to that of the observed (or simulated) deformable object [30], [31], [32]. The techniques in this category share the same basic principle: applying random values to different springs properties and correct
NURBS mass-spring model parametrization
Through the remaining sections, we use the following notation: bold lower cases, like ’x’, for vectors and particle positions, which are represented in column arrays, and capital letters, like ’A, M, …, etc’, for matrices. The NURBS Mass-Spring Model (NMSM) proposed in this work combines IGA, NURBS and MSM in an innovative way. The NMSM uses IGA model exploring its geometry accuracy, ability to deal with complex topologies and global refinement mechanisms of control mesh. This characteristic is
NMSM experiments
In order to evaluate the proposed NMSM parametrization process we have carried out experiments with different elastic models (2D and 3D objects). The elastic objects used in our experiments have the Young's modulus given by E = 15 kPa which is a usual value for soft body [22], [24], the Poisson's ratio of ν = 1/3, and mass density ρ = 1.143 kg/m3 unless otherwise stated.
In all the experiments, boundary conditions were specified through fixed control points, that are colored as red in the control mesh.
Discussion
In the experiments for the sensitivity to grouping springs (subsection 4.4), at first glance, one may notice that Table 8(a) and (b) show an enhancement of the solution when the group size (the number of springs groups) becomes higher. Nonetheless the running III in Table 8(b) is contrary to this tendency. In the MSM, the mechanical properties are consequences of the mesh topology, the stiffness coefficients and rest lengths, which control the degrees of freedom of the system. Grouping springs
Conclusion and future works
The development of a systematic procedure to set up Mass-Spring Models (MSM) parameters remains an open problem because the model parameters are not related to the constitutive laws of elastic material in an obvious way. The recent applications of isogeometric analysis open another possibility to face this problem from perspectives not yet explored. So, in this article we address this problem through a data-driven optimization methodology that can be used for determining stiffness constants in
Josildo Pereira da Silva received his PhD degree in Computer Science from Federal University of Bahia, Brazil, in 2015. He is currently Professor at Federal Intitute of Bahia. His current research investigates algorithms for modeling and simulation deformable objects, with application in interactive mechanical simulations. His research interests are primarily virtual environment, surgical simulation, games, physics-based deformable models and isogeometric analysis.
References (50)
- et al.
Isogeometric analysis: CAD, finite elements, NURBS, exact geometry and mesh refinement
Comput. Methods Appl. Mech. Eng.
(2005) - et al.
Determination of elasticity parameters in lumped element (mass-spring) models of deformable objects
Graph. Models
(2010) - et al.
Evolutionary optimization of mass-spring models
CIRP J. Manuf. Sci. Technol.
(2009) - et al.
Isogeometric analysis using T-splines
Comput. Methods Appl. Mech. Eng.
(2010) - et al.
THB-splines: the truncated basis for hierarchical splines
Comput. Aided Geometr. Des.
(2012) - et al.
Isogeometric analysis using LR B-splines
Comput. Methods Appl. Mech. Eng.
(2014) - et al.
Isogeometric analysis: an overview and computer implementation aspects
Math. Comput. Simul.
(2015) Elasticity: Theory, Applications, and Numerics
(2009)- et al.
Physically based deformable models in computer graphics
Comput. Graph. Forum
(2006) - et al.
A survey of physically based simulation of cuts in deformable bodies
Comput. Graph. Forum
(2015)
A survey of computer-based deformable models
Machine Vision and Image Processing Conference, 2007. IMVIP 2007. International
Skinning: real-time shape deformation
ACM SIGGRAPH 2014 Courses
Physics-based Animation (Graphics Series)
Animation of deformable bodies with quadratic Bézier finite elements
ACM Trans. Graph. (TOG)
The Finite Element Method: Its Basis And Fundamentals, The Finite Element Method
Isogeometric Analysis: Toward Integration of CAD and FEA
Adaptive Hierarchical Isogeometric Finite Element Methods
Isogeometric analysis of plane-curved beams
Math. Mech. Solids
Virtual Clothing: Theory and Practice
Realistic and stable animation of cloth
Comput. Anim. Virtual Worlds
Calibration of mass spring models for organ simulations
IEEE/RSJ International Conference on Intelligent Robots and Systems, 2007. IROS 2007
A non-linear mass-spring model for more realistic and efficient simulation of soft tissues surgery
Stud. Health Technol. Inform.
Maxillofacial surgery simulation using a mass-spring model derived from continuum and the scaled displacement method
Int. J. Comput. Assist. Radiol. Surg.
Designing Deformable Models of Soft Tissue for Virtual Surgery Planning and Simulation Using the Mass-Spring Model
Modeling and simulation of skeletal muscle for computer graphics: a survey
Found. Trends. Comput. Graph. Vis.
Cited by (0)
Josildo Pereira da Silva received his PhD degree in Computer Science from Federal University of Bahia, Brazil, in 2015. He is currently Professor at Federal Intitute of Bahia. His current research investigates algorithms for modeling and simulation deformable objects, with application in interactive mechanical simulations. His research interests are primarily virtual environment, surgical simulation, games, physics-based deformable models and isogeometric analysis.
Gilson A. Giraldi received his PhD degree in Computer Graphics from Federal University of Rio de Janeiro, Brazil, in 2000. Since then he has been with the National Laboratory for Scientific Computing, Brazil, where he is responsible for academic research projects in the areas of image segmentation and data visualization.
Antonio L. Apolinário Jr. receives his bachelor's degree in Mathematics in 1991 and completed his PhD in Computer and Systems Engineering from Federal University of Rio de Janeiro, Brasil in 2004. He is currently Associate Professor at Federal University of Bahia. His current research investigates algorithms for modeling and register deformable objects, with application in medicine. His research interests are primarily computer graphics, augmented reality, real time rendering for games and deformable object modeling.