The 1st Joint International Conference on Multibody System Dynamics May 2527, 2010, Lappeenranta, Finland

Volumetric Contact Models and Experimental Validation

Michael Boos , John McPhee Department of Systems Design Engineering University of Waterloo 200 University Ave W., N2L 3G1 Waterloo, Canada e-mail: mboos@uwaterloo.ca, mcphee@uwaterloo.ca

ABSTRACT A volumetric contact dynamics model has been proposed for the purpose of generating reliable simulations of space-based manipulator contact dynamics rapidly. Forces and moments between two bodies in contact can be expressed in terms of the volume of interference between the undeformed geometries of the bodies. The model allows for the modelling of contact between complex geometries and scenarios where the contact surface is relatively large, while being computationally less expensive than nite element methods. However, the volumetric model requires experimental validation. Models for simple geometries in contact (e.g. cylinder-on-plane) have been developed for stationary contact and for contact with motion normal and tangential to the contact surface; an apparatus has been developed to experimentally validate these models. This paper focuses on validation of the normal contact models. Experimental measurements of forces and displacements will be used to identify the parameters related to the normal force, i.e. the volumetric stiffness and hysteretic damping factor for both soft and hard surfaces. The experimental measurements are compared with simulated results to assess the validity of the volumetric contact model. Keywords: contact dynamics, validation, parameter identication, normal forces, volumetric model.

1 1.1

INTRODUCTION Background and Motivation

The Mobile Servicing System (MSS), Canadas main contribution to the International Space Station (ISS), consists of two manipulators, the Space Station Remote Manipulator System (SSRMS) and the Special Purpose Dextrous Manipulator (SPDM). The MSS is used for assembly and maintenance of the ISS [1]. Tele-operated space-based robotic operations requires careful task planning, verication and training on the ground. The complexity and risk of these operations means that accurate real-time contact dynamics models are required for on-earth simulation and training. Many point contact models are unsuitable for situations involving complex or conforming contacts, which may occur when handling orbital replacement units (ORUs). More complex models, such as nite element models, are far too computationally intensive for real-time simulation. The Canadian Space Agency has applied hardware-in-the-loop simulations (HLS) to determine contact dynamics; however HLS can be expensive [5]. A volumetric contact model has been proposed for use in generating reliable simulations of space-based manipulator operations. This model can be shown to be applicable to complex and conforming geometries, and accounts for angular dynamics ignored by many point contact models such as rolling resistance and spinning friciton torque. However, experimental validation of this model is required. 1.2 Point Contact Models

Contact is often modelled using point-contact models - that is, the region of contact is assumed to be very small relative to the geometries of the bodies, such that contact may be assumed to be occurring at a single point. A common model used in relatively stiff contact for robot modelling is the Kelvin-Voigt model, in which the contacting materials are represented by a spring and damper in parallel [7]. The contact model

proposed by Hunt and Crossley is a normal force model similar to spring-damper models, where the spring force is based on Hertz theory for linear elastic solids and the damping is adjusted to avoid the limitations incurred by other models. The Hertz model for contact force is the best known model for contact between two spheres of isotropic material [9]. In order to use this model, a number of assumptions must be made [3]. First, it must be assumed that the contact patch is small relative to the geometries of the bodies. This allows contact to be considered from a single point, where the depth of penetration, can be measured. Consequently, the surfaces must be non-conforming (i.e. not concave relative to each other where they come into contact). Otherwise, contact might occur at more than one location or be spread out over a wide area. Finally, it is assumed that the bodies are homogenous isotropic, linearly elastic solids. According to Hertz theory, the contact normal force is given by fN = k n (1)

The generalized stiffness k is dependent on the shapes and material properties of the surfaces in contact. 3 For two spheres in contact, n = 2 . The Hertz theory formulation is applicable for spheres in contact under static conditions. However, if pure Hertz theory were to be applied for dynamic contact situations (ignoring friction), no energy would be dissipated in the process of contact. Thus some damping is necessary [9]. Hunt and Crossley [6] proposed the following model for the contact normal force: fN = K() + B(, ) = k n + ( n )
3 which is consistent with Hertzian theory for contacting spheres under static conditions for n = 2 .

(2a) (2b)

For low impact velocities and most linear elastic materials [2], the coefcient of restitution can be approximated for a limited range of values by eemp = 1 vi (3) where is an empirically determined value and vi is the initial impact speed. It can then be shown [10] that 3 fN k n (1 + ) 2 (4)

Once normal forces are determined via Hunt-Crossley contact, sliding friction forces can be modelled by ft = fN where is the dynamic friction coefcient. The Hunt-Crossley contact model is fast and efcient; little effort is required to compute the penetration depth and rate for simple geometries, as well as the contact forces. In addition, the model is continuous in velocities and accelerations, which aids numerical integration and should not introduce discontinuous disturbances for control. The static contact force in the model is based on physical theory, yielding high delity for low-speed impact and where assumptions about the size and shape of the contact region hold. However, not all contact scenarios may involve relatively small contact patches or simple, non-conforming geometries. In these cases, other models are required to accurately simulate contact. In addition, point contact models tend to ignore the rotational effects of contact. A complete contact model would need to account for torques derived from rolling resistance and spinning friction. 1.3 A Volumetric Contact Model (5)

A exible volumetric contact dynamics model has been proposed [3]. This model allows for more complex and conforming geometries where point contact models may be inadequate because contact surfaces are

relatively large. It can be shown that the model also accounts for angular dynamics such as rolling resistance and spinning friction torque. For larger or conforming contact surfaces, a Winkler elastic foundation model [8] has been used. The Winkler model assumes a pressure distribution from one surface deforming as a bed of springs to comply with the contacting surface. This model has been adapted to contact dynamics; the forces and moments between two bodies in contact can be expressed directly in terms of the volume of interference, V , between the undeformed geometries of the bodies.

fn

s S (s)

V
Figure 1. Volume of interference between two contacting bodies.

The volume of interference is given by V =

S

(s)dS =
V

dV

(6)

where S is the contact surface and (s) is the depth of penetration at point s on the contact surface, as depicted in Figure 1. The centroid of this volume is pc = 1 V pdV
V

(7)

where p is the position vector to a point in the volume. 1.3.1 Normal force model

In this model, the contact normal force is related directly to the size of the volume of interference through a volumetric stiffness kV , given in units of force per unit volume. The normal force is given by [3] fN = kV V (1 + avcn )n (8)

where a is a hysteretic damping parameter and vcn is the speed of the centroid in the normal direction, n. This force acts normal to the contact surface, through the centroid of the volume. It should be noted that Equation (8) is very similar in form to Equation (4) for the point-based Hunt-Crossley model. The parameter a can be shown to be dependent on only the coefcient of restitution and the initial impact velocity [5]. 1.3.2 Rolling resistance torque

As motion in the normal direction is damped through the normal contact force, there will also be resistance to tangential rolling, as tangential rolling causes parts of the contact surface to move in the normal direction. By integrating the contact pressure distribution over the contact surface, the rolling resistance torque is r = kV aJS t (9)

where t is the component of the relative angular velocity between the bodies tangent to the contact surface, and Js is the contact surface second moment of area, weighted by penetration depth: Js =
S

((s s )I s s )fs (s)dS

(10)

where s is a vector from the centroid pc to the point s on the surface. This can be approximated by the volume inertia tensor Js JV = ((V V )I V V )dV (11)
S

where V is a vector from the centroid pc to the point p in the volume. 1.3.3 Friction model

Both tangential friction forces and spinning friction torque can be determined for the volumetric model [4]. The Contensou effect is included, where spinning friction can cancel out some friction forces that would normally resist tangential motion (as with a oor polishing machine). 1.4 Approach

While this volumetric model presents numerous benets, it is yet to be validated experimentally. An adaptable experimental apparatus has been designed and constructed for validating the friction and normal force elements of the model and for identifying major contact parameters between objects. Of special interest are the contact properties between metals used in space (e.g. aluminum, titanium, magnesium). This paper focuses on the validation and parameter identication pertaining to the normal forces between bodies in contact. Simple geometries are used in order to easily estimate volume of interference from measured displacement. Separate quasi-static and dynamic experiments are presented for determining volumetric stiffness and damping factors, respectively.

2 2.1

EXPERIMENTAL SETUP Experiments

The purpose of the experiments is to determine parameters for the volumetric contact model and validate that model. Parameters to be determined include a volumetric stiffness constant, a hysteretic damping factor, and friction parameters. Of interest for validation are the volume-normal force relationship, bristle-friction model, spinning friction torque model, and the Contensou effect. The experiments have been divided into those pertaining to the normal contact forces and the friction forces. For the purposes of this discussion, only the normal force experiments are described. For the contact normal force model, measurements of the displacement and forces in the normal direction are required. Two experiments are used to determine and validate both the volumetric stiffness and damping parameters separately. For the initial experiments, the at surface of a cylindrical payload is brought into contact with a planar force plate. A cylindrical payload was selected because it provided a relatively large conforming contact surface that would demonstrate the usefulness of the model in cases where point contact could not be used. Also, because contact pressure is spread over a larger surface area than with say sphere-on-sphere or sphereon-plane, there is less risk of plastic deformation from a highly concentrated point load. Finally, the volume of interference between the cylinder and the plane can be expressed in terms of the depth of penetration, : V = r2 (12)

where r is the radius of the cylinder. Thus, the volume of interference and consequently, the normal force in the model, have a linear relationship with the displacement.

2.1.1

Volumetric stiffness

Starting from rest, with the payload touching the contact surface with no forces between them, the force driving the payload is gradually increased so that the force sensors are loaded quasi-statically. The rate in displacement was limited to 12 m/s such that the effect of damping was negligible. The measured displacement was used to nd the volume of interference, so that a volumetric stiffness constant, kV , could be estimated through a linear t of force to volume measurements. 2.1.2 Damping

i The hysteretic damping factor a depends on the initial normal velocity at impact, vn , as well as a coefcient of restitution, eef f . With a known volumetric stiffness and measured displacements, velocities, and forces, the damping factor and coefcient of restitution can be estimated.

The payload is brought into contact with the force plate at different velocities and the subsequent forces and displacements are measured. For each case, the damping factor, a is estimated. 2.2 Experimental Apparatus

An apparatus has been developed to experimentally validate these models. The apparatus has two congurations, one for validating the normal contact models and the other for the friction models. The normal conguration uses a linear actuator to drive a rigidly mounted payload into a normal contact surface, which is mounted to a force transducer with a congured resolution of 0.1 N . Payload position relative to the contact surface is measured through a linear encoder. For the friction conguration, the same linear actuator and encoder are employed for translational motion of the payload across a contact surface mounted to a 5 DOF force sensor, and a second motor rotates the payload about the normal axis. The normal conguration is depicted in Figure 2. A cylindrical titanium payload with a radius of 5 mm is rigidly clamped to the linear actuator. Also rigidly attached to the actuator is a linear encoder which measures the position of the payload relative to a glass reference grating. This reference is rigidly mounted to the force sensor. The linear encoder has a resolution of up to 1.22 nm.

Force sensor Cylindrical payload

Linear encoder Contact surface Encoder reference

Figure 2. Normal force conguration of the experimental apparatus.

Different contact surfaces can be mounted in front of the piezoelectric force sensor. An aluminum (6061) plate and a soft elastomer are used as contact surfaces. These materials are more compliant than the payload, so it is assumed that the majority of the deformation will take place in the contact surface rather than the payload.

2.3 2.3.1

Parameter Estimation Volumetric stiffness

Under quasi-static loading, the magnitude of Equation (8) can be approximated as fN = kV V which can be re-expressed in terms of displacement fN = r2 kV (14) (13)

Using only measurements where contact is detected by the force sensor (normal force is above the noise threshold), a best-t line of the form fN = m(x b) (15) relating force to measured displacement (x) is found using least-squares. The x-intercept b is determined to be the point of contact giving =xb (16) The slope m (or displacement stiffness) is then used to nd the volumetric stiffness in units of N/m3 : kV = 2.3.2 Hysteretic damping factor m r2 (17)

Once the volumetric stiffness, kV is known, the damping factor a can be determined. Equation (8) can be rearranged to separate the normal velocity from the other measurements fN 1 = avcn kv V (18)

Velocities are estimated from rst-order central differences of the displacement measurements. Since Equation (18) is linear in terms of a, the damping factor can be solved using simple least squares.

3 3.1

RESULTS AND DISCUSSION Quasi-static Experiments Elastomer contact

3.1.1

In order to remain within the linear elastic range of the elastomer, loading forces on the payload were limited to 4 N . Displacement and normal force measurements are depicted in Figure 3. As expected, there is a linear relationship between the contact force and displacement and volume of interference. The elastomer was found to have a linear displacement stiffness of 2.12 104 N/m and a volumetric stiffness of 2.71 108 N/m3 . 3.1.2 Aluminum contact

The aluminum contact surface was loaded from 0 to 30 N . From the measurements shown in Figure 4 it would appear that the force-displacement relationship for forces below 5 N is non-linear, but linear above 5 N . This may be the result of either a misalignment between the payload and the contact surface or contact between surface asperities at low pressure. The aluminum was found to have a linear displacement stiffness of 2.48 107 N/m and a volumetric stiffness of 3.16 1011 N/m3 .

4 Measured data Model fit 3.5

3 Contact force (N)

2.5

1.5

0.5 150

100 Displacement (m)

50

Figure 3. Force and displacement measurements for quasi-static loading of elastomer contact surface.

35 Measured data Model fit 30

25 Contact force (N)

20

15

10

0 2

1.5

1 Displacement (m)

0.5

Figure 4. Force and displacement measurements for quasi-static loading of aluminum contact surface.

3.2 3.2.1

Dynamic experiments Elastomer contact

The elastomer contact surface was impacted by the payload at various speeds between 1.7 and 3.1 mm/s and the damping factor was estimated for each case. Measured data for the 2.25 mm/s impact case along with the predicted and measured damping force is shown in Figure 5.

Displacement (mm)

0.2

0.1

0 5

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

Measured Model with damping Model no damping

Force (N)

0.01

0.02

0.03

0.04

0.05 0.06 Time (s)

0.07

0.08

0.09

0.1

Figure 5. Force and displacement measurements for elastomer impact at 2.25 mm/s with modelled contact forces.

The damping factors for each speed are shown in Figure 6. Also shown is the theoretical curve for a where = 21.9 s/m, according to the model described by Gonthier [3]. The mean damping factor was 44.8 s/m. 3.2.2 Aluminum contact

Impact velocities between the payload and the stiff aluminum contact surface were limited to less than 1 mm/s in order to remain within the force limits of the experimental apparatus. Under these low velocities, damping with the aluminum plate could not be observed - almost all of the normal force could be related to the volumetric stiffness alone. This is to be expected from Equation (3), which suggests that for very low initial velocities, impacts should be nearly elastic. 3.3 Sources of Error

It is likely that the non-linearity in the force-displacement relationship for the initial contact between the payload and the aluminum plate are the result of mechanical imperfections. A misalignment of the apparatus by three hundredths of a degree would mean that the relationship between the volume and displacement should be non-linear for the rst half-micron of contact, for the given payload. Surface asperities are also a plausible cause for the non-linearity, since the surfaces have not been smoothed to a mirror nish. It is also possible that there is some small compliance between surface asperities of the aluminum contact surface and the plate of the force sensor when the load is applied. The range of contact loads presented in the aluminum case is limited by the experimental apparatus. When

60

50

Damping factor (s/m)

40

30

20

10 Estimated factors Model value for = 21.9 s/m 0 1.6 1.8 2 2.2 2.4 2.6 Impact velocity (mm/s) 2.8 3 3.2

Figure 6. Estimated damping factor a plotted against impact velocity with theoretical curve for a where = 21.9 s/m.

higher loads are achieved, slip may occur between the payload and the clamp. Additionally for higher loads, there is compliance in the plate holding the encoder reference to the force sensor. These limitations should be addressed for future experiments. The motor driving the linear actuator was a source of vibration to the payload. Since damping is the result of vibrations dissipating away energy, the vibration of the motor may have had some impact on the accuracy of the dynamic experiments. Further investigation, including performing dynamic experiments with the payload de-coupled from the motor (e.g. drop-tests), is required.

CONCLUSIONS

A volumetric contact dynamics model based on the Winkler elastic foundation model is presented for validation. Forces are expressed in terms of the properties of the volume of interference between the solid geometries of the bodies in contact. A series of experiments and an apparatus have been presented to validate the model of the normal contact force in static and dynamic conditions and to identify volumetric stiffness and hysteretic damping factors. Experiments were performed using a cylindrical payload on a planar surface in order to provide a relatively large contact surface area and so that the relationship between volume of interference and measured displacement should be linear according to the model. Soft (elastomer) and hard (aluminum) contact surfaces were used. Quasi-static experiments were used to determine and validate volumetric stiffness. Stiffnesses of 2.71 108 N/m3 for the elastomer and 3.161011 N/m3 for the aluminum were determined. A linear relationship between force and volume of interference was found for the elastomer surface. Non-linearity between force and volume was observed for low contact forces with the aluminum surface, likely the result of small mechanical imperfections. A linear relationship was found for forces above 5 N . Dynamic experiments could not be performed to measure the damping factor for the stiff aluminum contact surface within the safe limits of the apparatus. Damping parameters over a limited range from 1.7 to 3.1 mm/s impacts with the elastomer surface were estimated and found to fall within the range of 44.8 15 s/m.

Future efforts include improving the acceptable force range of the experimental apparatus, validating the model for other simple geometries (such as sphere-on-plane), and extending the apparatus for friction experiments.

ACKNOWLEDGEMENTS

We would like to acknowledge the support of Dr. Yves Gonthier of the Canadian Space Agency (CSA). Funding for this research was provided by the Natural Sciences and Engineering Research Council of Canada (NSERC), the CSA, and Parallel Geometry Inc.

REFERENCES [1] DE C ARUFEL , J., M ARTIN , E., AND P IEDBUF , J.-C. Control strategies for hardware-in-the-loop simulation of exible space robots. IEEE Proceedings-D: Control Theory and Applications 147, 6 (2000), 569579. [2] G OLDSMITH , W. Impact: The Theory and Physical Behavior of Colliding Solids. Edward Arnold Ltd., London, U.K., 1960. [3] G ONTHIER , Y. Contact Dynamics Modelling for Robotic Task Simulation. PhD thesis, University of Waterloo, 2007. [4] G ONTHIER , Y., M C P HEE , J., AND L ANGE , C. On the implementation of coulomb friction in a volumetric-based model for contact dynamics. In Proceedings of ASME IDETC (Las Vegas, Nevada, September 2007). [5] G ONTHIER , Y., M C P HEE , J., L ANGE , C., AND P IEDBUF, J.-C. A regularized contact model with asymmetric damping and dwell-time dependent friction. Multibody System Dynamics 11, 3 (2004), 209233. [6] H UNT, K., AND C ROSSLEY, F. Coefcient of restitution interpreted as damping in vibroimpact. Journal of Applied Mechanics 7 (June 1975), 440445. [7] JANABI -S HARIFI , F. Collision: Modeling, simulation and identication of robotic manipulators interacting with environments. Journal of Intelligent and Robotic Systems 13, 1 (May 1995), 144. [8] J OHNSON , K. Contact Mechanics. Cambridge University Press, London, 1985. [9] L ANKARANI , H. M., AND N IKRAVESH , P. E. Continuous contact force models for impact analysis in multibody systems. Nonlinear Dynamics 5 (1994), 193207. [10] M ARHEFKA , D., AND O RIN , D. A compliant contact model with nonlinear damping for simulation of robotic systems. Systems, Man and Cybernetics, Part A: Systems and Humans 29, 6 (November 1999), 566572.