You are on page 1of 6

Journal of Biomechanical Engineering

Technical Brief
anterior cruciate ligament ACL during walking and stair ascent 1, the inuence of Q angle on knee kinematics during a squat 2, and the affect of posterior cruciate ligament PCL resection on patello-femoral and tibio-femoral kinematics during a squat 3. In addition, a dynamic knee simulator has been used to compare joint kinematics of implanted mobile bearing prosthetics and xed implants to original intact cadaver knees 4. A variety of methods have been employed to reproduce the loading and motion of ambulatory activities on dynamic knee simulators. Paviovic et al. utilized a plastic knee model when programming their machine to imitate walking. Estimated muscle forces were applied to simulated quadriceps and hamstring actuators and then the hip and ankle were iteratively moved until the desired exion angle and relative spatial positions were achieved 5. McLean and Ahmed imitated walking on their simulator by using published data of knee exion, hip and ankle exion, vertical and fore/aft ground reaction, and center of pressure location under the foot to calculate the necessary inputs to the actuators of the machine 6,7. A dual-limb knee squat was imitated on the John Hopkins Knee Simulator by applying a constant 100-N vertical load to a simulated hip and a constant 150 N of force to a simulated hamstring. Flexion angle was then controlled through a simulated quadriceps by linearly extending the actuator at 0.5 mm/ s 24. A two-dimensional computational model of the Purdue Knee Simulator: Mark II was developed to predict the required simulator input proles to produce desired sagittal-plane knee loading 8,9. This study documents the rst step in achieving the overall goal of reproducing desired three-dimensional net knee loading or motion on a dynamic knee simulator. The overall goal will be accomplished through development of a computational model that translates three-dimensional net knee loading estimated in a gait lab, or through other methods, into control proles that drive the actuators of the dynamic simulator. The specic aims of this study were: 1 construct a constrained and simplied analog knee instrumented to measure joint forces, 2 develop a threedimensional computational model of a dynamic knee simulator and the analog knee, 3 verify the capability of the computational model to predict knee loading by comparing predicted and measured analog knee forces through squat and laxity tests performed on the simulator, 4 verify the capability of the model to generate control proles to the controllable axes of the simulator by reproducing the loading and motion of a three-dimensional walking prole on the analog knee.

Computational Modeling of a Dynamic Knee Simulator for Reproduction of Knee Loading


Trent M. Guess
e-mail: guesstr@umkc.edu Department of Mechanical Engineering, University of Missouri-Kansas City, 350F Robert H. Flarsheim Hall, 5100 Rockhill Road, Kansas City, MO 64110 Phone: 816 235-1252

Lorin P. Maletsky
e-mail: Maletsky@ku.edu Department of Mechanical Engineering, The University of Kansas, 1530 W. 15th St., 3138 Learned Hall, Lawrence, KS 66045 Phone: 785 864-2985

As a rst step towards reproducing desired three-dimensional joint loading and motion on a dynamic knee simulator, the goal of this study was to develop and verify a three-dimensional computational model that generated control proles for the simulator using desired knee loading and motion as model inputs. The developed model was veried by predicting tibio-femoral loading on an instrumented analog knee for given actuator forces and the ability to generate simulator control proles was demonstrated using a three-dimensional walking prole. The model predicted axial tibia loading for a sagittal-plane dual-limb squat within 1% of measured peak loading. Adding out-of-sagittal-plane forces decreased the accuracy of load prediction. The model generated control proles to the simulator that produced axial tibia loading within 16% of desired for walking. Discrepancies in predicted and measured quadriceps forces inuenced the accuracy of the generated control proles. Future work will replace the analog knee in both the model and machine with a prosthetic knee. DOI: 10.1115/1.2073676

Introduction
Dynamic knee simulators attempt to reproduce the estimated forces, moments, and motions of both the patello-femoral and tibio-femoral joints during dynamic activities. These machines have been used to evaluate the sagittal plane restraining role of the
Contributed by the Bioengineering Division of ASME for publication in the JOURBIOMECHANICAL ENGINEERING. Manuscript received by the Bioengineering Division June 19, 2003; revision received July 25, 2005. Associate Editor: Marcus G. Pandy.

Materials and Method


Kansas Knee Simulator (KKS). Modeled after the Purdue Knee Simulator: Mark II 8, the Kansas Knee Simulator Fig. 1 is a ve-axis dynamic simulator designed to reproduce the loading from ambulatory activities using cadaveric knees or total knee prostheses. Each of the ve axes is actuated with a hydraulic cylinder with servo valve control. Both position and force are measured at each axis allowing for control in either position or load. The machine includes a femur and tibia that can independently ex and that are respectively grounded through a hip and Transactions of the ASME

NAL OF

1216 / Vol. 127, DECEMBER 2005

Copyright 2005 by ASME

Downloaded From: http://www.asmedigitalcollection.asme.org/ on 11/16/2013 Terms of Use: http://asme.org/terms

Fig. 2 KKS computational model with analog knee and controllable axes

hydraulic cylinders. The mass properties of the rigid bodies were derived from empirical measurements of assembled components of the simulator. Analog Knee. The analog knee included simplied articulations and was instrumented to measure tibio-femoral loading. The tibio-femoral joint was designed as a simple revolute joint and the patello-femoral joint was designed as a simple pulley mechanism to accommodate loading applied by the quadriceps actuator. Compressive forces were measured at four locations, medial and lateral tibia axial compression and medial and lateral posterior force. The analog knee was designed using a standard computer aided drafting program and component geometries were imported into the computational model of the KKS. Three-dimensional deformable contacts represented the load cells used to measure tibiofemoral force on the analog knee. A series of stiff nonlinear springs connected by small spheres and contacts between the spheres and surfaces of the analog knee represented the pulley mechanism of the patello-femoral joint. Control Prole Generation. In the computational model, linear feedback algorithms were used to control the magnitudes of the axis force vectors in order to obtain desired model motion or loading. For the walking prole demonstrated, hip exion angle was maintained through PID control of the force vector representing the KKS quadriceps axis hydraulic cylinder. In addition, desired tibia axial compression was maintained through PID control of a KKS axis applying a vertical load to the hip sled vertical load axis and desired tibial axis internal/external torque was maintained through PID control of a KKS axis applying a torque about a vertical axis at the ankle vertical torque axis. Model inputs were the time histories of desired loading at respective

Fig. 1 Photograph of the Kansas Knee Simulator with analog knee adduction/abduction actuator below picture

ankle sled. The resultant knee loads and motions are reactions to forces applied at the simulated hip and ankle and from a simulated quadriceps muscle. In general, the simulated quadriceps is used to control femur exion at the hip and the remaining four actuators operate in load control to apply dynamic loading at the hip and ankle. Computational Model of KKS. A three-dimensional computational model of the KKS was developed in the computer aided engineering package MSC.ADAMS MSC Software Corporation, Santa Ana, CA. MSC.ADAMS formulates and solves the dynamic equations of motion for a given system of constraints that include lumped parameter rigid bodies, vector forces, torques, joints, springs, friction, and contacts. The KKS model Fig. 2 was intended to provide a computational platform that replicated the motion and loading produced by the hydraulic actuators, mass, and inertia of the machine. To replicate the proper line of action of the ve hydraulic actuators; the basic three-dimensional structures of the machine were modeled. Four revolute joints and three translational joints were used to describe the degrees of freedom of the KKS and vector force elements represented the forces generated by the ve

Table 1 Verication Tests Axis Control Mode Squat Laxity 1,2,3 Laxity 4,5,6 Quadriceps (hip exion angle) Position deg 23 13 @ 0.1 Hz 10, 23, 36 10, 23, 36 Vertical Force N 200 44.5 44.5 Ad / Ab Force N 0 0 44.5 @ 0.05 Hz 0 Vertical Torque Torque N-mm 0 0 0 6778 @ 0.05 Hz Ankle Flexion Force N 0 0 0

Journal of Biomechanical Engineering

DECEMBER 2005, Vol. 127 / 1217

Downloaded From: http://www.asmedigitalcollection.asme.org/ on 11/16/2013 Terms of Use: http://asme.org/terms

locations or axes about the knee and model outputs were KKS axis input proles needed to reproduce the desired knee loading. Model Verication. To verify that the model accurately represented the forces generated by the actuators, mass, and inertia of the KKS and the transfer of loading to the analog knee from these forces, a squat prole and six laxity tests were performed Table 1. For all tests, the quadriceps axis was in position control of hip exion angle and the remaining KKS axes were in load control. Hip exion angle was dened as 0 deg when the femur was parallel to a vertical axis and 90 deg when the femur was parallel to the ground. During the squat prole, the quadriceps axis followed a sinusoidal prole of 23 deg 13 deg hip exion angle at 0.1 Hz and the vertical load axis maintained a constant 200-N hip load. The remaining axes followed zero load proles. For the six laxity tests the quadriceps axis maintained constant hip exion angles of 10, 23, and 36 deg and the axis loading shown in Table 1. Control Prole Verication. A walking prole was used to verify the capability of the computational model to develop simulator control proles that reproduced the desired loading and motion at the analog knee. The walking prole included knee exion/ extension angle, compressive force along the tibial axis, and internal/external rotational torque about the tibial axis from ISO specication 14243-1. In addition, an adduction/abduction load from Hersh 10 scaled for a body mass of 86 kg 190 lb was applied at the ankle. The walking cycles were for a right knee and were run at a frequency of 0.05 Hz.

Results
The root-mean-square rms errors between predicted and measured loading for the squat prole were 49, 44, 71, and 28 N, respectively, for the medial axial compression, lateral axial compression, medial posterior, and lateral posterior positions. Figure 3 displays one cycle of measured and predicted loading for the squat at all four measurement positions. The difference between predicted maximum load and measured maximum for one cycle of testing was 4, 2, 69, and 59 N, or 0.2%, 0.1%, 13.3%, and 11.5% of maximum predicted loading. In both the KKS and computational model of the KKS, the quadriceps axis was used to maintain desired hip exion angle. Root-mean-square error between predicted and measured quadriceps axis loading was 151 N. Measured and predicted quadriceps load is shown in Fig. 4 along with vertical load tracking for one cycle. The measured quadriceps load lags the predicted load with a greater force than predicted during the extension phase of the squat and a lower force than predicted during the exion phase. A constant 200 N was the desired vertical load for the KKS to maintain during the exion-extension cycle. The actual force applied to the hip sled varied from a maximum of 227 N during extension to a low of 168 N during exion. The laxity tests were used to conrm the ability of the model to predict knee loading produced by the out-of-plane axes. Figure 5 provides a graph of rms errors between predicted and measured loading for the six laxity verication tests as well as the squat test. Axial compression errors were greater when out-of-plane loading was applied compared to the squat prole. For the walking prole Fig. 6, rms errors were 154, 107, 35, and 41 N, respectively for the medial axial compression, lateral axial compression, medial posterior, and lateral posterior positions. The difference in peak loading between measured and predicted axial compression was 241, 80, 37, and 61 N, or 15.8%, 7.4%, 12.8%, and 25.3% for the four measurement positions. The one-cycle rms error between predicted and measured quadriceps load was 119 N. During stance phase, the model consistently overestimated the quad load required to maintain the desired exion angle and consequently overestimated the tibia axial knee loading. The predicted vertical load prole required to generate compressive loading was underestimated as the quadriceps load provided the additional compressive load in the model. Input proles gen1218 / Vol. 127, DECEMBER 2005

Fig. 3 Predicted and measured loading for one cycle of the squat prole at the medial axial compression a, lateral axial compression b, medial posterior c, and lateral posterior sensor positions d

erated by the model and KKS tracking for one walking cycle are shown in Fig. 7. Except for the initial peak during stance and swing phase in the vertical load axis, the KKS closely follows the reference signals. Transactions of the ASME

Downloaded From: http://www.asmedigitalcollection.asme.org/ on 11/16/2013 Terms of Use: http://asme.org/terms

Discussion
As stated, this work represents a rst step in reproducing threedimensional loading on a ve axis dynamic knee simulator. Future work will replace the analog knee with geometries and structures more representative of joints that will be placed in the simulator for experimental testing. The instrumented analog knee used in this study represents the patello-femoral joint as a simple pulley that only redirects forces. In natural and prosthetic knees the patello-femoral joint redirects forces and in addition the ratio of quadriceps to patellar tendon force varies with exion angle 1113. The KKS model with analog knee reasonably predicts the quadriceps force required to reproduce desired hip exion for the verication tests. Quadriceps rms error was less than 41 N for all laxity tests. Quadriceps rms error during the squat was 151 N for

a cycle that varied from approximately 650 to 2900 N, and it is believed that much of this error was due to variation in simulator force at the vertical load axis Fig. 4. Quadriceps rms error increases as the force applied at the vertical load axis increases. The model overestimated the quadriceps force required to maintain the desired exion angle during the stance phase of walking and as a result produced control proles that overestimated tibia axial load. During the stance phase, vertical load axis force exceeded 900 N compared to a constant 200 N for the squat and 44.5 N for the laxity tests. It is believed that unmodeled forces, possibly from coiled hydraulic lines or unmodeled friction in the pulley mechanism, caused the model to overpredict the quadriceps force when loading from the vertical axis was large. The model and method of generating control proles is sensitive to load discrepancies at the controllable axes. In both the walking prole and laxity tests, predicted abduction/ adduction moment at the knee for a given medial/lateral load at the ankle was greater than measured. In addition, greater motion was measured at the ankle sled during tests that included medial/ lateral ankle forces than predicted in the model. With the exception of the deformable contacts simulating the load cells in the analog knee, the components and assemblies of the computational model were perfectly rigid. Although small, there was compliance in the assemblies and connections of the KKS and analog knee. It is believed that unmodeled compliance in the frontal plane accounts for some of the error seen between predicted and measured tibia axial compression. Adding compliance to the KKS model, through a spring or exible body, may increase the accuracy of frontal plane load prediction. In addition, these ndings highlight the need for accurate representation of frontal plane compliance in future versions of the model. The ability to translate estimated three-dimensional net knee loading into machine inputs for reproduction of that loading is a desired capability of the KKS. To reach this goal, a computational model of the simulator was developed and an instrumented analog knee constructed. The generation of machine inputs to reproduce desired loading for a three-dimensional walking prole on the analog knee was demonstrated. In addition, model parameters that are important to accurate reproduction of joint loading were identied. This information will assist the next phase of the project, replacing the analog knee with a prosthetic knee.

Fig. 5 One-cycle rms error between predicted and measured loading at the analog knee along with rms error between predicted and measured force on the quadriceps axis for all verication tests

Journal of Biomechanical Engineering

DECEMBER 2005, Vol. 127 / 1219

Downloaded From: http://www.asmedigitalcollection.asme.org/ on 11/16/2013 Terms of Use: http://asme.org/terms

Fig. 6 Predicted and measured loading for one cycle of walking at the medial axial compression a, lateral axial compression b, medial posterior c, and lateral posterior sensor positions d

Fig. 7 Control axis reference and measured feedback for one cycle of walking at the quadriceps axis a, vertical load axis b, vertical torque axis c, and adduction/abduction axis d

1220 / Vol. 127, DECEMBER 2005

Transactions of the ASME

Downloaded From: http://www.asmedigitalcollection.asme.org/ on 11/16/2013 Terms of Use: http://asme.org/terms

Acknowledgments
Funding for this project was provided by The University of Kansas New Faculty General Research Fund, Award No. 2301811-RPSGNF.

References
1 Ahmed, A. M., and McLean, C., 2002, In Vitro Measurement of the Restraining Role of the Anterior Cruciate Ligament During Walking and Stair Ascent, ASME J. Biomech. Eng. 1246, pp. 768779. 2 Mizuno, Y., Kumagai, M., Mattessich, S. M., Elias, J. J., Ramrattan, N., Cosgarea, A. J., and Chao, E. Y., 2001, Q-angle Inuences Tibiofemoral and Patellofemoral Kinematics, J. Orthop. Res. 195, pp. 834840. 3 Kumagai, M., Mizuno, Y., Mattessich, S. M., Elias, J. J., Cosgarea, A. J., and Chao, E. Y., 2002, Posterior Cruciate Ligament Rupture Alters In Vitro Knee Kinematics, Clin. Orthop. Relat. Res. 395, pp. 241248. 4 Elias, J. J., Kumagai, M., Mitchell, I., Mizuno, Y., Mattessich, S. M., Webb, J. D., and Chao, E. Y., 2002, In Vitro Kinematic Patterns are Similar for a Fixed Platform and a Mobile Bearing Prosthesis, J. Arthroplasty 174, pp. 467 474. 5 Paviovic, J. L., Kirstukas, S. J., Touchi, H., Bechtold, J. E., and Gustilo, R. B., 1994, Dynamic Simulation Machine for Measurement of Knee Mechanics and Intra-articular Pressures, ASME Advances in Bioengineering, M. Askew Ed.,

ASME, New York, Vol. 28, pp. 277278. 6 McLean, C. A., and Ahmed, A. M., 1993, Design and Development of an Unconstrained Dynamic Knee Simulator, ASME J. Biomech. Eng. 1152, pp. 144148. 7 McLean, C. A., and Ahmed, A. M., 1994, Design and Performance of an Unconstrained Dynamic Knee Simulator, ASME Advances in Bioengineering, M. Askew, Ed., New York, Vol. 28, pp. 285286. 8 Maletsky, L. P., and Hillberry, B. M., 2000, Loading Evaluation of Knee Joint During Walking Using the Next Generation Knee Simulator, ASME Advances in Bioengineering, T. A. Conway ed., ASME, Orlando, Florida, Vol. 48, pp. 9192. 9 Maletsky, L. P., and Hillberry, B. M., 2005, Simulating Dynamic Knee Activities Using a Five-Axis Knee Simulator, ASME J. Biomech. Eng., 1271, pp. 123133. 10 Hersh, J. F., 1980, Laboratory Evaluation of Knee Prostheses, M.S. thesis, Mechanical Engineering, Purdue University, West Lafayette, IN. 11 Hefzy, M. S., and Yang, H., 1993, A Three-Dimensional Anatomical Model of the Human Patello-femoral Joint, for the Determination of Patello-Femoral Motions and Contact Characteristics, J. Biomed. Eng., 154, pp. 289302. 12 Hirokawa, S., 1991, Three-Dimensional Mathematical Model Analysis of the Patellofemoral Joint, J. Biomech., 248, pp. 659671. 13 van Eijden, T. M., Kouwenhoven, E., Verburg, J., and Weijs, W. A., 1986, A Mathematical Model of the Patellofemoral Joint, J. Biomech., 193, pp. 219229.

Journal of Biomechanical Engineering

DECEMBER 2005, Vol. 127 / 1221

Downloaded From: http://www.asmedigitalcollection.asme.org/ on 11/16/2013 Terms of Use: http://asme.org/terms