You are on page 1of 9

Development of an LS-DYNA nonlinear finite element model for use in damage detection and health monitoring of highway bridges

Dhafer Marzougui*a, Shuang Jinb, Richard A. Livingstonb


a

FHWA/NHTSA National Crash Analysis Center, Ashburn, VA 20147 b Federal Highway Administration, McLean, VA 22101
ABSTRACT

As part of a program to apply stochastic system analysis to structural heath monitoring of highway structures, a detailed Finite Element (FE) model of a typical highway bridge has been developed. The model was created for use with the nonlinear explicit FE code, LS-DYNA, and consists of 144 parts and approximately 40,000 elements. The model represents a standard two-lane bridge with a span length of 40 meters. It consists of 4 girders and 21 cross frame sections. This paper discusses some important practical aspects involved in the modeling of such highway bridges including connections, material properties, boundary and dynamic loading conditions. Extensive simulations were conducted using a SGI supercomputer at the FHWA sponsored National Crash Analysis Center at the George Washington University to determine the bridge structural response under dynamic loadings. The resulting data sets from these simulations are used as the basis for chaotic system invariant spectrum analysis described in related papers in this conference. Keywords: Nonlinear FEM, Computer Simulation, Highway Bridge, Damage Detection, Structural Health Monitoring

1. INTRODUCTION
Highway bridges require continuous monitoring and maintenance in order to keep them operational and to avoid catastrophic structural failures. As these bridges getting older, it is essential to develop more sophisticated structural heath-monitoring techniques to provide early damage detection and to predict critical weak areas in the bridge. Several new structural heathmonitoring methods have been evolved in recent years [1,2,3]. Research at the Federal Highway Administration and the FHWA/NHTSA National Crash Analysis Center has focused on to develop advanced nonlinear stochastic dynamics analysis techniques applied to structural health monitoring of highway bridges. To verify the effectiveness of these new methods, highly realistic and detailed numerical models of typical highway bridges were needed. This paper specifically describes the development of such models and reports our current effort on the development of a nonlinear finite element model of standard two-lane highway bridge. In particular, we first introduce detailed information about the bridge structure and then we describe the methodology used in developing a nonlinear finite element representation of bridge structures. These include: the model geometry, constitutive material models, structural connections, complex boundary conditions, and dynamical loadings. Subsequent sections also briefly present some information about the different configurations of the bridge models developed in this study. Some of the simulation results obtained using these models have been used in related papers reported in this conference.

2. BRIDGE DESCRIPTION
Figure 1 presents portions of the design drawing of a typical two-lane standard full-size highway bridge. The bridge has a span length of 40 m and a width of 9.3m. The bridge deck is made by reinforced concrete of 0.25m-thick covered with a 0.05m-thick asphalt wearing surface. The deck is supported by four I-shaped steel girders, which are positioned 2.44m apart along the lateral direction. In the longitudinal direction, each girder consists of three sections. The two-end sections have a 12.7mm thick top flange, a 17.5mm-thick web, and 15.9mm-thick bottom flange. The middle section has a thicker, 25.4mm, bottom flange. Plate stiffeners are welded to the web and flanges of the bridge girders and located 6.6m apart along the
*

Correspondence: Email: dmarzoug@ncac.gwu.edu; Telephone: 703-726-8366; Fax: 703-726-8359


Health Monitoring and Management of Civil Infrastructure Systems, Steven B. Chase, A. Emin Aktan, Editors, Proceedings of SPIE Vol. 4337 (2001) 2001 SPIE 0277-786X/01/$15.00

432

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/16/2014 Terms of Use: http://spiedl.org/terms

longitudinal direction. Except the first and the last stiffeners, which are 17.5mm-thick, all other stiffeners are 9.5mm-thick. Also, there are cross frames placed between the girders to increase their lateral buckling stiffness. The bridge has a total of 21 cross frames, seven between each pair of girders. Each cross frame consists of three L-shaped steel beams. Two of them are placed diagonally and one is placed horizontally. The two diagonal beams are bolted at the ends to four steel plates, which in turn are bolted to the girder stiffeners. Two T-shaped steel gussets are connected to the lower portion of the girders to support the cross frame. The horizontal beam of the cross frame is connected to the lower steel plates and to the T-shaped gussets. The bridges deck-girders superstructure is supported by piers through the bearing and roller devices respectively at each end of the girders.

(a)

(b) Figure 1: Details of a typical bridge structure [4] (a) Deck and cross fame details (b) Girder details

3. FINITE ELEMENT MODELING METHODOLOGY


The finite element program selected for this study is LS-DYNA, an explicit three-dimensional code [5,6]. It was chosen because of its powerful capabilities in advanced modeling of nonlinear dynamics behavior, which are essential to our current studies and analyses. In order to develop an accurate, reliable finite element representation of a standard full size bridge structure, several key features in such a bridge structure need to be addressed, examined, and carefully modeled. In particular, these include:

Proc. SPIE Vol. 4337

433

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/16/2014 Terms of Use: http://spiedl.org/terms

1. 2. 3. 4. 5.

Component geometry involved in the bridge structure, Constitutive material model invoked in the analysis of such structures, Structural connections among various components in the bridge structural system, Complex boundary conditions, and Dynamical loading conditions involved in the analysis of the bridge structures.

It is noted that although many simplified structural models have been developed in the past to describe the basic behavior of such a bridge structure, many detailed structural contents have been ignored. This is a major source that causes significant discrepancy between the analytical results of the bridge model and the real behavior of the bridge structure. For example, the precise geometry of the concrete deck, girders, and cross members has a direct impact on the overall dynamic response of the bridge. As main structural components of the bridge, they affect the bridges mass and inertia properties as well as its stiffness. Consequently it is important to include accurate geometry representation of these components. Another important aspect regarding the dynamics behavior of the bridge structure is the material properties involved. In particular, unlike as commonly assumed being linear in design and dynamics analyses, a real bridge structure can undergo significant nonlinear behavior either locally or globally due to plastic deformation, time varying dependencies of the materials, and aging degradation of the structural properties. In order to ensure that a finite element bridge model can capture this nonlinear behavior in dynamics analyses, adequate constitutive material models needed to be assigned to each structural component in the bridge. The structural connections, such as bolts and welds, add both local and global rigidity to the bridge structure. Therefore, they also have to be correctly incorporated in the computer model. Furthermore, since we are dealing with the dynamics analyses of the nonlinear bridge structures and the bridge subjects to not only gravity loading but also complex traffic loading, a correct modeling of the dynamical loading to a real bridge structure is also critical. The following provides an overall description of the methodology in creating a nonlinear finite element model for the bridge structure investigated. A brief overview of the finite element program used in this study is given first. Following this discussion, several key features of the bridge structure, including how they have been incorporated into the nonlinear finite element model, and the techniques used for modeling damage due to crack are presented.

3.1 Finite element program LS-DYNA is an explicit Lagrangian finite element program that solves three dimensional, dynamic, nonlinear and large displacement problems. It has several features that allow for the analysis of a verity of engineering problems. In the past 20 years, it has been widely used in several research fields such as aerospace structures, automotive crash analysis, metal stamping, etc. It has a library of over eighty constitutive material models ranging from the very basic elastic material to complex orthotropic composite materials. Several material models are available to model plasticity as well as time dependent nonlinear material behaviors. LS-DYNA has a large selection of finite element types which include: one node lumped mass, two-node discrete spring and damper elements, beam elements, three and four-node shell elements, and eight-node solid and thick shell elements. These elements can be used together to model different geometry of the structural components. The code allows for simulating different dynamic loads. Some of the options for modeling loadings include concentrated load, pressure, and base accelerations. Connections such as bolts and welds can also be modeled in the code using features like nodal set constrains, rigid body constrains, and Spotwelds. LS-DYNA also has the capability of modeling fixed boundary conditions. Translational and rotational degrees of freedom of the nodes can be constrained in three-dimensional space along one or more of the global axes as well as arbitrary axes. The most advanced capability of LS-DYNA is its contact algorithm. Several types of contact interfaces can be defined in LS-DYNA. These include surface-to-surface, nodes to surface, nodes tied to surface, and surface tied to surface contacts. Because of the above-mentioned features, LS-DYNA was chosen for this study. A trade-off limiting factor, however, in using this code is computational time. Due to the nature of the explicit time integration scheme used in the code, a small time step has to be used to obtain a convergent solution to the problem. The time step is controlled by the element size and material properties. Since the problem is solved at every time step, many cycles have to be computed before reaching the end time. This leads to large computation time especially for long duration problems. Typical time steps for automotive cash worthiness problems are in order of 10-6 seconds and typical simulation time period is 0.1 second. For the bridge model, however, the time step is in the order of 10-5 seconds (the bridge has larger structural components and therefore can be modeled with larger elements) and the simulation time is in the order of 100 seconds.

434

Proc. SPIE Vol. 4337

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/16/2014 Terms of Use: http://spiedl.org/terms

3.2 Model geometry In this study, the detailed geometry of the main bridge components, such as concrete deck, girders, cross frame members, is incorporated into the finite element model. All these components are modeled with shell and solid elements rather than with beam elements. The detailed geometry presentation has significant effect on the accuracy of the bridge response. Figure 2 shows sample plots of the detailed finite element geometry modeling for the main structural members of the bridge. Without such a detailed geometry description, the overall physical characteristics of the bridge components, i.e. mass, inertia, center of gravity, and stiffness of the structural members, were not correctly represented and the areas of stress concentration caused by particular geometrical profiles of the bridge might be misrepresented. Another reason for using the real geometry to model main structural components is to ensure that bucking and torsional deformations of these components can be properly predicted. For example, using beam elements to model the girders cannot account for either the torsional deformation in the longitudinal direction or the out-of-plane deformations of the web. Compared to those much simplified bridge models, a detailed geometry representation allows for a significantly improved ability to provide accurate response prediction of the bridge structure. The description of the bridge geometry is based on the Standard Plans of Highway Bridges [4]. Some of these drawings are shown in Figure 1. The geometry and mesh of the bridge are created using PATRAN, a sophisticated pre-processor compatible with LS-DYNA. The girders and cross frame beams are represented in the finite element model using shell elements with the given thicknesses. The concrete deck and the wearing surface are represented in the model by using solid elements. Figure 2 also shows finite element mesh of some of the components in the bridge. The size of the mesh is carefully chosen such that the computation time is minimized without sacrificing accuracy.

(a)

(b) Figure 2: Detailed geometry incorporated into the bridge model (a) Cross frame (b) Girder

3.3 Constitutive material models Another important characteristic in developing a robust finite element model is to select the appropriate constitutive material formulation for the bridge components. When subjected to gravity and traffic loads, the bridge components can undergo nonlinear deformations. These nonlinearities can be attributed by plastic deformations or time varying dependency of the structural material. To capture these nonlinearities in the finite element model, each component of the bridge structure has been cautiously considered and given an appropriate constitutive model. For example, the wearing surface on the bridge deck is considered to have time rate dependent behavior. This material is assumed to behave as a visco-elastic material. The formulation of the visco-elastic constitutive model in LS-DYNA requires the following material parameters: the short term shear modulus, the long-term shear modulus, the bulk modulus, and a decay constant. These material parameters are obtained from the literature as listed in Table 1. Time dependency is not critical for the girders and cross frame members. However, they can experience plastic deformation or fracture. Consequently, these components are assigned by the isotropic piecewise linear elastic-plastic material model. This is the most commonly used material model in LS-DYNA and is suitable for modeling most engineering structural materials such as steel and aluminum. The parameters needed for this material

Proc. SPIE Vol. 4337

435

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/16/2014 Terms of Use: http://spiedl.org/terms

model are the mass density, elastic modulus, Poisons ratio, yield stress, and plastic strain at failure. In addition, the stress/strain relation is characterized using a curve of effective true stress versus effective plastic true strain. The material parameters used for the steel bridge girders are listed in Table 1. The true-stress/true-strain relation is shown in Table 2.

Table 1: Material properties used in bridge model Concrete Mass Density (kg/m3) Young's Modulus (Pa) Poisson's Ratio Yield Stress (Pa) Stress/Strain Relations 2.3E+03 2.76E+10 0.15 Wearing Surface 2.1E+03 3.1E+09 0.35 Steel 7.9E+03 2.1E+11 0.3 2.98E+08 Table 2

Table 2: Stress-strain relation used for steel material model


Plastic Strain (m/m) 0.000 0.020 0.040 0.080 0.160 0.320 0.640 Yield Stress (Pa)

2.98E+8 3.40 E+8 3.70 E+8 4.15 E+8 4.80 E+8 5.75 E+8 8.10 E+8

3.4 Structural connections One of the important aspects of the bridge model assembling is the manner by which the parts are interconnected. Several connections are introduced to model the bolts and the welds within the bridge structure. These connections can introduce localized areas of stress concentration and discontinuities in the structure. Choosing the correct connection type and location are very critical in ensuring appropriate behavior of the bridge structure. LS-DYNA has provided several options for choosing types of connections. These include: joints, spotwelds, nodal set constraints, rigid body constrains, welds, and various types of tied contact interfaces. Rigid body constraints and spotwelds are the most frequently used connections in LS-DYNA. They create a rigid body link between two or more nodes. These connections are primarily used to replicate the bolts in the finite element model of the physical bridge and they are incorporated here to represent the connections among the different cross frame members, the stiffeners, and the girders. Over 750 rigid body constraints have been used in our bridge model. Some of these connections are illustrated in Figure 3.

Figure 3: Rigid body constraints used to represent bolt connections

3.5 Dynamic loadings During everyday service, the bridge subjects to two types of loading: a static loading associated to its own weight (gravity loading) and a dynamic loading due to the moving vehicle traffic flow. In our finite element model of the bridge structure, gravity loads are incorporated by invoking the base acceleration option in LS-DYNA. Using this option, the global coordinate system of the bridge model is subjected to an acceleration of equal magnitude and in opposite direction to the gravity field (g). Each nodal point in the model is then subjected to an acceleration of the same direction and magnitude as the gravity field. This acceleration is introduced at the initial time of the simulation hence causes some initial oscillations. Note that in realty this gravity load is static and the bridge is already at equilibrium. Therefore, care must be taken in the analysis of the observed response data from such simulations since the dynamic effects are solely caused by moving traffic

436

Proc. SPIE Vol. 4337

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/16/2014 Terms of Use: http://spiedl.org/terms

loads. To eliminate these oscillations two treatments were investigated. The first was to use the dynamic relaxation feature available in LS-DYNA. This feature is often used to bring a system to equilibrium before starting the dynamical simulation. The dynamic relaxation analysis works in a very similar manner to the standard transient analysis except that damping forces are applied to the nodes to reduce the inertial oscillation and bring the system to equilibrium faster. The second was to allow, before applying the moving traffic loads, a period of time for the oscillations until the effect due to application of gravity loads to die out. Both treatments produced very similar results. However, the second treatment takes slightly more computation time but it is more systematical in nature and was used in this study. The loads due to dynamical moving traffic flow are incorporated into the finite element bridge model using concentrated nodal forces with appropriate load curves. The pressure due to the tire contacts with the road surface is assumed to act at the centerline of the tire and to move at the same speed as the vehicle. To simulate this moving load, load curve time histories are assigned to the nodes in the path of the vehicle motion. In order to ensure that the load is transferred smoothly from one node to the next and the load to the road surface is continuous, triangular pulse load curve time histories are used. Figure 4a shows portions of typical load curves assigned to two adjacent nodes. The triangular pulse load for a certain node n is chosen such that the load starts at zero when the tire is located halfway between the (n-1)th node and the (n)th node. The magnitude of the load increases and reaches maximum when the tire is located right on the top of the (n)th node. The load magnitude then decreases to zero when the tire reaches the halfway between the (n)th and the (n+1)the nodes. This method was found to be very effective in modeling the moving traffic loads. Examining the simulation results, it is clearly seen that pressure contours representing the loads of moving tires on the surface of the bridge deck are moving smoothly. Figure 4b shows a sample plot of the pressure distribution on the wearing surface of the bridge. The curves in Figure 4a show a portion of load curve in which the two triangular pulses represent respectively the load applied by the tire on two adjacent nodes. The magnitude of the pulse is determined by the vehicle weight and its distribution among the tires. The width of the triangular pulse is calculated based on the vehicle speed and the distance between adjacent nodes. The curves in Figure 4a show a situation of a load applied by a two-axle vehicle. Loads for vehicles with more than two axles can also be easily incorporated into the model in the same way by simply adding more triangular pulses to the loading curves. Similarly, situations of multiple vehicle moving loads can be invoked in the simulation of the finite element bridge model by introducing multiple triangular pulses with different arrival times.

Load

Vehicle moving direction

Load curve for node n

Load curve for node n+1 Time

(a) Figure 4: Load curves used for simulating dynamic loads (a) Curves of load vs. time (b) Pressure distribution on road surface

(b)

3.6 Boundary conditions The bridge is connected to the piers through rollers and bearing devices. Each roller device limits the relative motion between the girder and the pier to only one rotation about the lateral axis of the bridge girder. The bearing device, on the other hand, allows for translation motion along the longitudinal direction of the bridge girder. These devices support both the bridge weight and the moving traffic loads. It is therefore crucial to accurately represent these devices in the bridge structure. This was achieved in the finite element model by including the real geometry of these devices. Because of space limitation,

Proc. SPIE Vol. 4337

437

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/16/2014 Terms of Use: http://spiedl.org/terms

Figure 4 only shows the front and side view drawings of a typical roller device and the corresponding finite element representations. The bearing device is modeled in similar way.

(a)

(b) Figure 5: Roller device modeling (a) Design drawings of typical roller device (b) Finite element representation of roller device

3.7 Modeling of damage As previously mentioned, the main reason for developing these finite element bridge models is to provide significantly improved ability to use numerical models in dynamics analyses and simulations in order to verify the effectiveness of approaches for structural health-monitoring of such highway bridges. The goal is to provide improved understanding about the detailed mechanisms of structure damage in a bridge. To test this ability, damage is introduced at certain locations in the finite element bridge model and the results are compared against the situation when the bridge structure was intact. The damage is introduced by weakening the overall stiffness in certain region or location in the bridge girder. This was achieved by introducing anyone of the following methods. The first method consists of reducing the thickness of a few elements at a selected damage location. Since these elements have smaller cross-section area, their ability to support the load is diminished dramatically and a localized weakened area is realized. The second method is very similar to the first except that rather than to reduce the thickness the elastic modulus is reduced to weaken these elements. In the third method, a crack is introduced in the model. The nodal connections of the elements were modified in such a way that a crack is formed between the edges of elements. Simulations with these three methods were conducted and it was observed that all three methods produced very similar overall results.

4. FINITE ELEMENT MODEL DESCRIPTIONS


Several finite element models of standard highway bridges were developed to accommodate the needs of current research efforts on examining the effectiveness of various structural heath monitoring methods. The models varied in size and complexity. Three of these models are presented in this section. These models are shown in Figure 6 and a summary of the information related to these finite element models is given in Table 3. The first model as shown in Figure 6a is a detailed representation to a standard two-lane highway bridge. This model has the most detailed description of the bridge structure and was used to simulate the bridge response under dynamic loading due to a single-unit truck. The truck used in the simulation weights 18.1Tons (40,000lbs) and has a steady moving speed of 45 MPH. The weight of the truck was distributed in such a way that the front axel loads were 3.6Tons (8,000lbs.) and the rear axle loads were 14.5Tons (32,000lbs.). The bridge model consists of 26 different parts to represent the different bridge structural components. The geometry of the

438

Proc. SPIE Vol. 4337

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/16/2014 Terms of Use: http://spiedl.org/terms

bridge was represented using 26,000 nodes, 20,000 shell elements, and 15,000 shell elements. The simulation was performed on a SGI Origin 2000 multi-processor machine. Using 6 processors of this machine, it required 65 hours of computational time to simulate a 5 seconds period of highly accurate response of the bridge. In the second finite element model, as shown in Figure 6b, only one quarter of the bridge structure was included. Nonlinear springs were used to account for the remaining section of the bridge. The nonlinear properties of the springs were extracted from the full bridge model simulation. This model was used to simulate the response of the bridge in different situations. Some of these situations include: the bridge has no crack damage, the bridge has a single crack at different locations, and the bridge girder has multiple cracks at different locations. This model consists of 19 parts, 7,000 nodes, 2000 shell elements, 3000 solid elements, and 21 beam elements. The computational time required to simulate the same 5 seconds period of the bridge response was only 4 hours. The third finite element bridge model, as shown in Figure 6c, was used to simulate the bridge response under multi-vehicle traffic loads. The model consisted of 21 parts, 13,000 nodes, 8000 shell elements, 6000 solid elements, and 105 beam elements. This model is a reduced version of the first one. Some details were omitted from the model to decrease computational time. For example, the cross frame members were represented by beam elements rather then shells. However, special care was taken in this process not to diminish the overall integrity of the model. The traffic flow used in the simulation was modeled by using the Poisson distributed pulse process [7]. A total of 500 vehicles were used to simulate a period of 120 seconds twolane traffic flow. The vehicles varied in weights, geometries, speeds and arrival rates. Two cases were analyzed, in the first case the bridge was assumed to be free of damage and in the second case damage was introduced at different locations in the girders. The computational time required to simulate a period of 122 seconds bridge response was 160 hours.

(a)

(b)

(c)

Figure 6: Standard bridge finite element models (a) Detailed full bridge model (b) Quarter bridge model (c) Reduced full bridge model

Proc. SPIE Vol. 4337

439

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/16/2014 Terms of Use: http://spiedl.org/terms

Table 3: Summary of bridge models information

Number of Parts Number of Nodes Number of Shells Number of Solids Number of Beams Simulation Time Duration CPU Time

Detailed Full Bridge Model 26 26,000 20,000 15,000 0 5 sec 65 hrs

Quarter Bridge Model 19 7,000 2,000 3,000 21 5 sec 4 hrs

Reduced Full Bridge Model 21 13,000 8,000 6,000 105 122 sec 160 hrs

5. CONCLUDING REMARKS
In this paper, a detailed nonlinear finite element model for a typical, standard two-lane highway bridge is developed for use in advanced dynamics analyses and simulation studies applied to damage detection and heath monitoring of highway bridges. Several important practical aspects involved in the modeling of such bridge structures, including structural connections, material properties, boundary and dynamical loading conditions, have been addressed. This nonlinear finite element model has also been used in nonlinear dynamics analysis and extensive computer simulations to determine the bridge structural response under dynamical traffic loadings. The resulting response data sets from the simulations are used in advanced nonlinear and chaotic system invariant spectrum analysis for study of the problems in damage detection of aging highway bridge structures. It is shown that in addition to those much simplified bridge structural models, the advanced nonlinear finite element modeling, analysis and simulation can be invoked to develop bridge models using LS-DYNA code to effectively reveal detailed dynamical behavior of bridge structural system. This provides a significantly improved ability in advanced dynamics analysis in damage detection and structural health monitoring of highway infrastructures.

ACKNOWLEDGMENT
This work was performed while the second author held a National Research Council Research Associateship at the TurnerFairbank Highway Research Center of the Federal Highway Administration.

REFERENCES
1. Jin, S., Livingston, R.A., and Marzougui D., (2000), Energy Index Approach for Damage Detection in Nonlinear Highway Structures, SPIEs 5th Annual International Symposium on NDE for health Monitoring and Diagnostics, Newport Beach, Ca. Tseng, S.S. and Saleeb, A. F., (1998), Defect energy for global damage detection method, Journal of the Chinese Institute of Civil and Hydraulic Engineering, Vol. 10(2), pp.389-396. Yao, G. C., Chang, K.C. and Lee, G. C., (1992), Damage diagnosis of steel frames using vibrational signature analysis, Journal of Engineering Mechanics, ASCE,118(9), pp.1949-1961. Standard Plans for Highway Bridges, (1982) Vol. 2, Structural Steel Superstructures, U.S. Department of Transportation, Federal Highway Administration. Hallquist, J.O., (1997), LS-DYNA Theoretical Manual, Livermore Software Technology Corporation. Hallquist, J.O., (1997), LS-DYNA User's Manual, Livermore Software Technology Corporation. Livingston, R.A., Jin, S., and Marzougui D., (2001), Stochastic Modeling of Ambient Traffic Loadings in LS-DYNA Nonlinear FE Analysis, SPIEs 6th Annual International Symposium on NDE for health Monitoring and Diagnostics, Newport Beach, Ca.

2. 3. 4. 5. 6. 7.

440

Proc. SPIE Vol. 4337

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/16/2014 Terms of Use: http://spiedl.org/terms

You might also like