Journal of Biomechanics 31 (1998) 963 967

Technical Note

An analytical model of the knee for estimation of internal forces

during exercise
Naiquan Zheng, Glenn S. Fleisig*, Rafael F. Escamilla, Steven W. Barrentine
American Sports Medicine Institute, 1313 13th Street South, Birmingham, AL 35205, U.S.A.
Received in final form 13 March 1998

Abstract
An analytical model of the knee joint was developed to estimate the forces at the knee during exercise. Muscle forces were estimated
based upon electromyographic activities during exercise and during maximum voluntary isometric contraction (MVIC), physiological
cross-sectional area (PCSA), muscle fiber length at contraction and the maximum force produced by an unit PCSA under MVIC.
Tibiofemoral compressive force and cruciate ligaments tension were determined by using resultant force and torque at the knee,
muscle forces, and orientations and moment arms of the muscles and ligaments. An optimization program was used to minimize the
errors caused by the estimation of the muscle forces. The model was used in a ten-subject study of open kinetic chain exercise (seated
knee extension) and closed kinetic chain exercises (leg press and squat). Results calculated with this model were compared to those
from a previous study which did not consider muscle length and optimization. Peak tibiofemoral compressive forces were
3134$1040 N during squat, 3155$755 N during leg press and 3285$1927 N during knee extension. Peak posterior cruciate
ligament tensions were 1868$878 N during squat, 1866$383 N during leg press and 959$300 N for seated knee extension. No
significant anterior cruciate ligament (ACL) tension was found during leg press and squat. Peak ACL tension was 142$257 N during
seated knee extension. It is demonstrated that the current model provided better estimation of knee forces during exercises, by
preventing significant overestimates of tibiofemoral compressive forces and cruciate ligament tensions. ( 1998 Elsevier Science Ltd.
All rights reserved.
Keywords: Knee model; Muscle force; Cruciate ligament; Knee exercise

1. Introduction
Understanding the biomechanics of the knee during
exercises, such as knee extension, leg press and squat, is
very important for therapists to design rehabilitation
programs and for trainers to strengthen and condition
athletes. Muscle force, ligament tension and joint contact
forces between the femur and tibia during exercise are the
main issues to evaluate exercises. For example, therapists
try not to overload the anterior cruciate ligament (ACL)
for ACL injured and reconstructed patients during their
rehabilitation programs. It has been controversial which
exercise should be utilized for a patient since the ACL
tension during different exercises are not fully understood (Wilk et al., 1997).

* Corresponding author. Tel.: 001 205 918-2138; fax: 001 205 9180800; e-mail: glennf@asmi.org
0021-9290/98/$19.00 ( 1998 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 2 1 - 9 2 9 0 ( 9 8 ) 0 0 0 5 6 - 6

Recently, resultant forces of the knee during different exercises have been studied (Lutz et al., 1993; Stuart
et al., 1996; Wilk et al., 1996). Some efforts were made
in these studies to estimate muscle forces and to compare muscle activity; however, further study of tensions
of cruciate ligaments and the bone-to-bone contact
force of the tibiofemoral joint was prevented by limits
of the biomechanical models utilized. Muscle forces
were estimated based only on the physiological crosssectional area and electromyographic activities, not
muscle fiber length at contraction. Thus, muscle forces
were overestimated at full extension for the knee extensors without taking into account the shorter muscle fiber
length.
The purpose of this study was to develop an analytical
model of the knee in the sagittal plane during quasistatic
(i.e. isometric or low speed) exercise. The tibiofemoral
compressive force and tensions of the anterior and posterior cruciate ligament were presented during the knee
extension, leg press and squat.

964

N. Zheng et al. / Journal of Biomechanics 31 (1998) 963967

2. Materials and methods

Resultant force and moment at the knee were the sums
of those due to individual ligament, muscle and joint
contact:
n.
nl
n#
F " + F # + Flj# + F ,
3%4
.i
#k
i/1
j/1
k/1
nl
n#
n.
M " + M # + Mlj# + M ,
.i
#k
3%4
i/1
j/1
k/1
(A"m, l, c; i"1, 2, 3, 2 , n ),
M "r ]F
Ai
Ai
A
Ai

(1)

(2)
(3)

where F and M were the resultant force and moment

3%4
3%4
at the knee; n , nl and n were the number of muscle
.
#
groups, ligament bundles and bone contact interfaces
used in the model; F and M were the force and
.i
.i
moment of the ith muscle group; and Fl and Ml were the
j
j
force and moment of the jth ligament bundle; F and
#k
M were the force and moment of the kth bone contact
#k
element; and r was the vector of a moment arm.
Ai
To simplify the model some assumptions were made:
(a) collateral ligaments were not included since they have
little effects on the mechanics of the knee in the sagittal
plane (Crowninshield et al., 1976), (b) the flexionextension moment of the cruciate ligaments were ignored since
they were located close to the rotation center of the knee,
(c) the tibiofemoral force was assumed to be applied in
the line of knee rotation center, therefore, the contribution of the tibiofemoral force to the resultant extension
torque was ignored; and (d) the tibiofemoral joint and
patellofemoral joint were assumed to be frictionless due
to the synovial fluid.
In this model, resultant forces and torques at the knee
were assumed to be known. Forces of the muscles were
estimated using surface EMG data from eight muscle
groups (rectus femoris, vastus medialis, vastus lateralis,
vastus intermedius, biceps femoris, semitendinosus,
semimembranosus, and gastrocnemius) and the following
equation:
EMG
i,
F "c k A p
.i
i i i .i MVC
i

cross-bridge model of muscle. The muscle length was

determined by the knee joint angle, hip joint angle and
muscle line of action and geometry of the lower extremity (Pierrynowski, 1991), which determined the
muscle fiber length and sarcomere length (Herzog et al.,
1990). For single joint knee muscles, the muscle
length factor k was determined as a segmental linear
function of knee flexion angle since the isometric tension
of the muscle is a segmental linear function of the sarcomere length (Gordon et al., 1966). For two joint
muscles, k was a segmental linear function of both knee
and hip flexion angles (Fig. 1). Since the model was
developed for quasistatic exercise, no force-velocity factor was included.
PCSA data from Wickiewicz et al. (1983) were used to
determine the ratios of PCSA between muscle groups.
According to Narici et al. (1988), the total PCSA of the
quadriceps (PCSA ) was approximately 160 cm2 for
Q
a 75 kg man. Total PCSA of the quadriceps was scaled
up or down by individual body mass (PCSA "
Q
160 * BW/75 cm2). The maximum voluntary contraction
force per unit PCSA was assumed to be 40 N cm~2 for
the quadriceps and 35 N cm~2 for the hamstring and
gastrocnemius muscles (Narici et al., 1988; Cholewicki et
al., 1995; Narici et al., 1992; Wickiewicz et al., 1984).
Moment arms of muscle forces and angles of the line of
action for muscle and ligaments were represented as
polynomial functions of the knee flexion angle using data
from Herzog and Read (1993) (Tables 1 and 2). Since lines
of action and moment arms for muscles and ligaments
from literature were limited to the sagittal plane, only the
forces in the sagittal plane were analyzed in this model.
Using the equilibrium equations the cruciate ligament
forces (F , F ) and tibiofemoral force (F ) were
PCL ACL
TF

(4)

where k was a muscle forcelength factor defined as

i
a function of knee and hip flexion angle; A was the
i
physiological cross-sectional area (PCSA) of the ith
muscle; p was maximum voluntary isometric contrac.i
tion force per unit PCSA of the muscle; EMG and
i
MVIC were window averages of ith muscle EMG during
i
exercise and maximum voluntary isometric contraction;
coefficient c was the weigh factor which was adjusted in
i
an optimization program to minimize the errors in
muscle force estimation.
The muscle forcelength factor k was determined by
the knee joint angle and hip joint angle based on the

Fig. 1. Forcelength factor k for two joint muscle of quadriceps vs knee

and hip flexion angle. Muscle fiber length has significant effect on the
muscle force output.

N. Zheng et al. / Journal of Biomechanics 31 (1998) 963967

965

Table 1
Muscle moment arm lengths (, m) as a polynomial function of knee angle (h, degree) based on Herzog and Read (1993)
Function

"B #B h#B h2#B h3#B h4

0
1
2
3
4

Coefficient
Patellar tendon
Biceps femoris
Semimembranosus
Semitendinosus
Gastrocnenius

B
0
4.71E-02
1.46E-02
2.84E-02
!4.11E-03
1.99E-02

B
1
4.20E-04
!9.26E-05
!1.61E-04
5.86E-04
!3.50E-04

B
2
!8.96E-06
8.55E-06
6.81E-06
6.90E-06
9.20E-06

B
3
4.47E-08
!8.78E-08
!8.80E-08
!5.31E-8
!1.03E-07

B
4
0.00
2.38E-10
2.77E-10
0.00
4.07E-10

Table 2
Lines of action for muscles and ligaments (/, degree, 0 to the anterior, 90 to the inferior) as a polynomial function of knee angle (h, degree, 0 for full
extension) based on Herzog and Read (1993)
Function

/"B #B h#B h2#B h3

0
1
2
3

Coefficient
Patellar tendon
Biceps femoris
Semimembranosus
Semitendinosus
Gastrocnenius
Anterior cruciate ligament
Posterior cruciate ligament

B
0
!74.40
275.00
260.00
255.00
!90.00
227.00
!66.00

B
1
!5.75E-02
!0.8720
!0.8880
!0.8160
0.00
!0.4880
0.7370

estimated. Eq. (1) can be rewritten as

n.
(5)
F #F #F "F ! + F .
.i
TF
PCL
ACL
3%4
i/1
When the resultant anteroposterior shear force was positive, the tension of the posterior cruciate ligament was
determined and tension of the anterior cruciate ligament
was assumed to be zero. Conversely, when the resultant
anteroposterior shear force was negative, tension of the
anterior cruciate ligament was determined and tension of
the posterior cruciate ligament was assumed to be zero.
The patellar tendon force was determined by the quadriceps tendon force and the ratio of the patellar tendon
force to the quadriceps tendon force (Van Eijden et al.,
1987).
Since the accuracy of the estimation of the muscle
forces depends on the accuracy of the estimation of the
physiological cross-sectional area, maximum voluntary
isometric contraction force per unit PCSA, EMG data
and muscle fiber length, coefficient c was used for each
i
muscle in estimation of its force. These coefficients were
determined in an optimization program by minimizing
the difference between the resultant moment of the knee
from kinetic analysis (M ) and that from the estimation
3%4
of the model (+ M ). The following objective function
.i
was used:
min
subject to

2
n.
n.
f (c )" + (1!c )2#j M ! + M
i
3%4
.i
i
i/1
i/1
c )c )c ,
(6)
-08
i
)*')

B
2
!4.75E-03
!7.12E-04
!8.52E-04
2.63E-04
0.00
0.00
!4.96E-03

B
3
3.09E-05
0.00
0.00
!6.19E-6
0.00
0.00

where c and c
were the lower and upper limit for
-08
)*')
c and j was a constant. The coefficient c was set at one at
i
i
beginning and adjusted by using the DavidonFletcher
Powell algorithm (Dennis and Schnabel, 1983).
To test the model, previously collected data were used
(Wilk et al., 1996). A three-dimensional motion analysis
system (Motion Analysis Corporation, Santa Rosa, CA,
U.S.A.) and force platform (Advanced Mechanical Technology, Inc., Newton, MA, U.S.A.) were used to collect
kinematic data, external loads and ground reaction forces for ten subjects performing three exercises: seated
knee extension, leg press and squat. The knee flexion or
extension angular velocities were found to be approximately 90 s~1 for peak value and approximately 60 s~1
for average, making data acceptable for the quasistatic
assumption in the model. The resultant force and moment at the knee were determined by using threedimensional rigid link models and principles of inverse
dynamics (Feltner and Dapena, 1989; Wilk et al., 1996).
Surface EMG data were collected from rectus femoris,
vastus lateralis, vastus medialis, biceps femoris, medial
hamstrings and gastrocnemius muscles with a Noraxon
system (Noraxon, Pheonix, AZ, U.S.A.). The average of
the rectus femoris, vastus lateralis and vastus medialis
EMG data was used to represent the EMG data of the
vastus intermedius. EMG data of the medial hamstrings
were used for both the semitendinosus and semimembranosus. The window average of EMG during exercises
was expressed as a percentage of the window average of
EMG during the maximum voluntary isometric contraction, which was recorded when the muscle forcelength

966

N. Zheng et al. / Journal of Biomechanics 31 (1998) 963967

factor was one, and then the average for three repetitions
of each exercise was taken based on the knee angle. In
order to consider the electromechanical delay window
averages were calculated only for past history with a window width of 100 ms. The c , c
and were set to 0.5,
-08 )*')
1.5 and 1.0, respectively, for optimization.

3. Results
The tibiofemoral compressive force during leg press
and squat increased with knee flexion angle, while the
tibiofemoral compressive force during knee extension
increased with decrease of knee flexion angle (Fig. 2).
Peak tibiofemoral forces were 3134$1040 N during
the squat, 3155$755 N during the leg press and
3285$1927 N during the knee extension.
There were no significant tension in the anterior cruciate ligament during the leg press and squat (Fig. 2). The
tension in the anterior cruciate ligament increased when
the knee approached full extension during the seated
knee extension with a peak value of 142$257 N. In
general, tension in the posterior cruciate ligament increased with knee flexion. Peak PCL tensions were
1868$878 N during the squat, 1866$383 N during the
leg press and 959$300 N during the seated knee extension. Optimization produced significant adjustments on
muscle forces (Fig. 3).

Fig. 2. Tibiofemoral compressive force and PCL/ACL tension vs knee

flexion angle during knee extension (circle), leg press (triangle) and
squat (square). There were no significant difference in peak tibiofemoral
compressive forces among three exercises, but there were significant
difference in peak ACL and PCL tensions.

4. Discussion
A planar quasistatic model of the knee in the sagittal
plane was developed, which can be expanded to threedimensional when lines of action and moment arms of
ligaments and muscles are determined in three dimensions. The model was used for a study of 10 subjects
performing the knee extension, leg press and squat
(Zheng et al., 1996). The model took into account muscle
activities, especially the co-contraction of the extensors
and flexors of the knee. Although co-contraction of the
hamstrings and quadriceps may have no effect on the
extension torque of the knee joint, it has a significant
effect on tibiofemoral compressive force and anteroposterior shear force. Co-contraction of the hamstrings and
quadriceps is an important factor in determining tensions
in the cruciate ligaments. The co-contraction has more
effect on the tension in the posterior cruciate ligament
when the knee flexed 90, and has more effect on the
tension in the anterior cruciate ligament when the knee is
near full extension. This is due to varying lines of action
for the patellar tendon and hamstrings muscle. Thus, it is
important to take into account the co-contraction of the
hamstrings and quadriceps.
The model provided better estimation of internal forces of the knee during exercises than those in previous

Fig. 3. Muscle force estimation of quadriceps before and after optimization for a squat of one subject. Optimization had significant adjustments on muscle forces.

studies which did not consider muscle length and optimization. Linear muscle activityforce relationship was
assumed at a given knee flexion angle (Bean et al., 1988;
Lawrence et al., 1983). The muscle forcelength factor
k was about 0.25 for vastus lateralis, medialis and intermedius when the knee was in full extension. By ignoring
the forcelength factor the muscle forces would be estimated as high as four times. When only EMG and PCSA
were considered, the tibiofemoral compressive forces
were overestimated with peak values of 6139$1709 N
during the squat, 5762$1508 N during the leg press and
4598$2547 N during the knee extension (Wilk et al,,
1996), much higher than those calculated with this model.
By ignoring the forcelength factor of the muscles, higher
tibiofemoral compressive force (3453$1313) at 30 knee
flexion was reported than that (2198$805) at 90 during

N. Zheng et al. / Journal of Biomechanics 31 (1998) 963967

isometric leg press (Lutz et al., 1993). No cruciate

ligament tensions during exercises have been previously
reported. Tibiofemoral shear forces were reported to
compare different exercises (Wilk et al., 1996; Lutz et al.,
1993). Wilk et al. reported maximal posterior shear force
of 1783$634 N during the squat, 1667$556 N during
the leg press and 1178$594 N during the knee extension.
After considering the orientation of the PCL at the knee
flexion angle when these maximal shear force occurred,
PCL tensions would be about 2318 N during the squat,
2167 N during the leg press and 1531 N during the knee
extension, which were respectively 24, 16 and 60% higher
than those calculated with this model.
The optimization program was used for adjusting the
muscle force calculations, based upon knee resultant
forces and torques. EMG assisted optimization has been
recently introduced in muscle force estimation
(Cholewicki and McGill, 1994). The objective function
allowed the existence of co-contraction of quadriceps and
hamstring muscles. Various objective functions from the
literature (Pedersen et al., 1987) were tried and found to
be unable to include the co-contraction of the extensors
and flexor of the knee. Smaller range of lower and upper
limit for c will allow less adjustment for muscle force
i
estimation and may lose some effect of optimization.
Smaller (0.1, 0.01) should be used if the accuracy of the
known resultant extension torque of the knee (M ) is
3%4
lower. Further study is needed to improve the muscle
forcelength relationship during knee exercises, especially to specify the difference between concentric and
eccentric contractions. The model does not include
a force-velocity relationship for muscle force estimation
since it was developed for low speed exercise. Forcevelocity relationships should be included in the model
before it can be used for high-speed knee activities. The
improvement of the model could be achieved with increased knowledge of the assumptions used in the model
including PCSA, EMGforce relationship, lines of action
of muscles and other data needed to expand to a threedimensional model.

References
Bean, J.C., Chaffin, D.B., Schultz, A.B., 1988. Biomechanical model
calculation of muscle contraction forces: a double linear programming method. Journal of Biomechanics 21, 5966.
Cholewicki, J., McGill, S.M., 1994. EMG assisted optimization: a hybrid approach for estimating muscle forces in an indeterminate
biomechanical model. Journal of Biomechanics 27, 12871289.
Cholewicki, J, McGill, S.M., Norman, R.W., 1995. Comparison of
muscle forces and joint load from an optimization and EMG assisted
lumbar spine model: towards development of a hybrid approach.
Journal of Biomechanics 28, 32131.
Crowninshield, R, Pope, M.H., Johnson, R.J., 1976. An analytical
model of the knee. Journal of Biomechanics 9, 39740.

967

Dennis, J.E., Schnabel, R.B., 1983. Numerical Methods for Unconstrained Optimization and Nonlinear Equation. Prentice-Hall, Englewod Cliff, NJ.
Feltner, M.E., Dapena, J., 1989. Three-dimensional interactions in
a two-segment kinetic chain: Part I: general model. International
Journal of Sport and Biomechanics 5, 40319.
Gordon, A.M., Huxley, A.F., Julian, F.J., 1966. The variation in isometric tension with sarcomere length in vertebrate muscle fibers. Journal
of Physiology (London) 185, 170192.
Herzog, W. Abrahams, S.K., Keurs, H., 1990. Theoretical determination of forcelength relations of intact human skeletal muscles using
the cross-bridge model. European Journal of Applied Physics 416,
113119.
Herzog, W., Read, L.J., 1993. Lines of action and moment arms of the
major force carrying structures crossing the human knee joint. Journal of Anatomy 182, 213230.
Lawrence, J.H., DeLuca, C.J., 1983. Myoelectric signal versus force
relationship in different human muscles. Journal of Applied Physiology 54, 16531659.
Lutz, G.E., Palmitier, R.A., An, K.E., Chao, E.Y.S., 1993. Comparison
of tibiofemoral joint forces during open-kinetic-chain and closedkinetic-chain exercises. Journal of Bone and Joint Surgery 75-A,
732739.
Narici, M.V., Roi, G.S., Landoni, L., 1988. Force of knee extensor and
flexor muscles and cross sectional area determined by nuclear magnetic resonance imaging. European Journal of Applied Physics 57,
3944.
Narici, M.V., Landoni, L., Minetti, A.E., 1992. Assessment of human
knee extensor muscles stress from in vivo physiological cross-sectional area and strength measurement. European Journal of Applied
Physics 65, 438444.
Pedersen, D.R., Brad, R.A., Cheng, C., Arora, J.S., 1987. Direct
comparison of muscle force predictions using linear and nonlinear programing. Journal of Biomedical Engineering 109,
192199.
Pierrynowski, M.R., 1991. Analytic representation of muscle line of
action and geometry. In: Allard, P, Stokes, I., Blanchi, J. (Eds.),
Three-dimensional Analysis of Human Movement , Human Kinetics.
pp. 215256.
Stuart, M.J., Meglan, D.A., Lutz, G.E., Growney, E.S., An, K.N., 1996.
Comparison of intersegmental tibiofemoral joint forces and muscle
activity during various closed kinetic chain exercise. American Journal of Sports Medicine 24, 792799.
Van Eijden, T.M., Kouwenhoven, E., Weijs, W.A., 1987. Mechanics of
the patellar articulation. Acta Orthopaedica Scandinavica 58,
160566.
Wickiewicz, T.L., Roy, R.R., Powell, P.L., 1983. Muscle architecture of
the human lower limb. Clinical Orthopedics and Related Research
179, 27583.
Wickiewicz, T.L., Roy, R.R., Powell, P.L., Perrin, J.J., Edgerton, V.R.,
1984. Muscle architecture and force-velocity relationships in humans.
Journal of Applied Physiology: Respiratory, Environmental and
Exercise Physiology 57, 435443.
Wilk, K.E., Escamilla, R.F., Fleisig, G.S., Barrentine, S.W., Andrews,
J.R., Boyd, M.L., 1996. A comparison of tibiofemoral joint force
and electromyographic activity during open and closed kinetic
chain exercises. American Journal of Sports Medicine 24,
518527.
Wilk, K.E., Zheng, N., Fleisig, G.S., Andrews, J.R., Clancy, W.G., 1997.
Kinetic chain exercise: implications for the anterior cruciate ligament
patient. Journal of Sports Rehabilitation 6, 125143.
Zheng, N., Fleisig, G.S., Escamilla, R.F., Barrentine, S.W., Wilk, K.E.,
Andrews, J.R., 1996. Forces at the knee during open and closed
kinetic chain exercises. Proceedings of 20th American Society of
Biomechanics Annual Meeting, Atlanta, GA, pp. 7576.