Вы находитесь на странице: 1из 11

ES 205

Analysis and Design of Engineering Systems


Dr. Daniel Kawano
Spring 20132014

Identifying unknown parameters


for a mass-spring-damper system

April 18, 2014

Team Damper
Cody Bressler
Joe Kaltenthaler
Austin Schulthies

Rose-Hulman Institute of Technology


Terre Haute, IN 47803

Introduction
The goal of this experiment was to determine the physical parameters of a mass-spring-damper
system. This was accomplished using a semi-iterative process in which we collected data,
created a model, compared the model to the data, and then refined the model until it matched
the data with acceptable accuracy.

Experimental system
Figure 1 is a schematic of the experimental system. It consisted of a cart (mass), attached to a
spring and a damper. The cart rested on a level surface (table) for the entirety of the experiment.
A computer running MATLAB and Simulink controlled the magnitude of the input force and
recorded the displacement of the cart. The force which initially displaced the cart was provided
by an electric motor via a connecting rod attached to the rack and pinion. Power was supplied
to the motor and other system equipment through the ECP control box. As the mass oscillated,
its displacement was measured by the encoder and transferred to a computer via the ECP
control box to be plotted in MATLAB.

computer

control box

encoder
motor

rack and pinion

connecting rod
mass
spring
damper
table

Figure 1: General schematic of the experimental setup.

System model
To develop a model of the system, it was necessary to examine the general schematic and
simplify it to a small number of key components which could be analyzed independently of the
rest of the system. Since the goal of this experiment was to find the physical parameters of the
mass-spring-damper system, the only components necessary in developing the model were the
cart (mass), the spring and the damper.
Figure 2 shows the modeling schematic used to analyze the system when Coulomb friction was
not taken into account; where is mass of the cart, is the coefficient of viscous damping, is
the spring stiffness, and is the input force and is the horizontal displacement of the cart.

Figure 2. Initial modeling schematic of a mass-spring-damper system, when Coulomb friction is


not present.
Figure 3 shows the kinetic and free-body diagrams of the system when Coulomb friction was
assumed absent.
FBD

KD

in

Figure 3: Kinetic diagram and free body diagram for the initial model of the system.
From the principle of Conservation of Linear Momentum Rate, an initial second order
differential equation (1) was developed to model the horizontal displacement of the block in
centimeters.

+ + =

(1)

Figure 4 shows the modeling schematic used to analyze the system when Coulomb friction
acting between the cart and the table was taken into consideration; where is the coefficient of

Coulomb friction between the cart and the table, and is the constant of gravitational
acceleration.

Figure 4. Refined modeling schematic of a mass-spring-damper system with Coulomb friction


present.
Figure 5 shows the kinetic and free body diagrams of the system when Coulomb friction
between the cart and table was accounted for.

FBD

KD

in

mg
Figure 5: Kinetic and free body diagrams of the refined model of the system which includes the
presence of Coulomb friction.
Again, from the principle of Conservation of Linear Momentum Rate, a second order
differential equation (2) was developed to model the refined system; where the ( ) function
accounts for the sign of the carts velocity, and changes the sign of the Coulomb friction term
such that it always opposes the carts motion.

+ + + ( ) =

(2)

Experimental procedure
To run the system we first turned on the ECP box and then loaded the following files with
MATLAB and Simulink: lab_1_210.mdl and ECPDSPResetmdl.mdl. The ECPDSPResetmdl.mdl
program was run first to reset the encoder and define the carts position at that time to be zero.
The file lab_1_210.mdl was then run to drive the motor and provide a force to the cart causing

horizontal oscillatory motion. The encoder recorded the displacement of the cart and
transferred it to the computer where it was plotted in MATLAB.

Data
Figure 6 shows the data collected during the experiment.
3

Displacement (cm)

2.5

1.5

0.5

0.5

1
1.5
Time (seconds)

2.5

Figure 6. Experimental step response for a mass-spring-damper system.

Table 1 shows information extracted from the data which was useful for calculating the
standard form parameters, as well as , a value from the lab_1_210.mdl file.
Table 1. Useful data extracted from the experimental step response.

First peak
value, 1 (cm)

Second peak
value, 2 (cm)

Time to first
peak, 1 (s)

Time to
second peak,
2 (s)

Steady-state
value, ss (cm)

Software step
amplitude, step

2.554

2.150

0.1120

0.3440

1.502

0.15

Method of analysis
Identifying standard form parameters from the step response data
To estimate the systems natural frequency n , damping ratio , and static gain , we assume
the system is an ideal, linear, second-order system. This assumption is eventually shown to be
insufficient for obtaining a high-fidelity model whose response matches the data, but it is
sufficient for beginning the analysis.
For an ideal second-order system that is underdamped, the logarithmic decrement may be
calculated according to

1 ss
= ln (
) ,
2 ss

(3)

where 1 and 2 are the first and second peak values, respectively, and ss is the steady-state
value. The data reported in Table 1 are substituted into Eq. (3) to compute the logarithmic
decrement. Knowing , the damping ratio is determined using

2 + 4 2

(4)

The time interval between two peaks in the response data is used to calculate the observed
(damped) period of oscillation, d :

d = 2 1 ,

(5)

where 1 and 2 are the times that correspond to the first and second peaks, respectively. These
times are reported in Table 1. Knowing the period, the damped frequency d is computed
using

d =

2
.
d

(6)

Once and d are known, the natural frequency n is calculated according to

n =

d
1 2

(7)

The step force in applied to the cart is related to the software step amplitude step (which is
recorded in Table 1) by

in (69.5 N)step .

(8)

The static gain may then be calculated from in and the steady-state value ss using

ss
.
in

(9)

Identifying physical parameters assuming viscous friction only


Assuming all friction in the system may be modeled as viscous friction, the relationships
between the cart mass , viscous damping coefficient , and spring stiffness and the standard
form parameters n , , and are determined by comparing the original viscous friction model
(1) to the standard form of the model:

2
+
+ = in .
2
n n

(10)

Comparing the coefficients of Eqs. (1) and (10), if we know the static gain , then we may
estimate the spring stiffness according to

1
.

(11)

Similarly, after calculating and knowing n and , estimates for the cart mass and damping
coefficient are obtained from

(12)

2
= ,
n

(13)

n =
and

respectively.
Quantifying the goodness of fit
We compute the standard error of estimate (SEE) to quantify the goodness of fit between the
data and model response after running a simulation of the original model (1), whose parameters
are calculated using Eqs. (11), (12), and (13). The SEE is given by

(model, data, )
SEE = = 1
2

(14)

where is the number of data points, and model, and data, are the response values of the
model and data, respectively, at the ith instant in time.
Refining the model to obtain a better fit of model to data
When plotted on the same axes as the data, the simulation results for the original model (1)
reveal that the response of this model does not quite match the response data. The model is
then refined, as shown in Eq. (2), to include the effect of Coulomb friction (using a dry friction
coefficient ) in addition to viscous friction.
Beginning with the estimates of , , and for the original model (1), we set = 0 in the
refined model (2). We compare the simulation results for the refined model (2) to the
simulation results for the original model (1) to confirm that the refined model gives credible
results.
We then increase until we achieve both a reduced SEE value (calculating the SEE using Eq.
(14) after each iteration) and a good visual fit between the model response and the data. The
final values for , , , and represent our best estimates of the unknown physical
parameters for the mass-spring-damper apparatus.
Comparing the original and refined models
To quantitatively demonstrate the improvement in model fidelity upon refinement, we compute
the percent reduction in the SEE:

% reduction in the SEE =

SEEoriginal SEErefined
100 ,
SEEoriginal

where SEEoriginal and SEErefined are the SEE values for the original and refined models,
respectively.

Results and discussion


Table 2 displays the standard form parameters calculated in Eqs. (4), (7), (8) and (9).
Table 2. Standard form parameters for a mass-spring-damper system.
Natural
frequency, n
(rad/s)

Damping ratio,

Static gain,
(cm/N)

Input force, in
(N)

27.16

0.0769

0.1441

10.43

(15)

Table 3 compares the physical parameters, calculated using Eqs. (11), (12) and (13) and data in
Table 2, for the initial and refined models. Table 3 also includes SEE, calculated in Eq. (14), and
the Coulomb damping constant which was determined through trial and error.
Table 3. Comparing calculated physical parameters for initial and refined model of a massspring-damper system.

Case

Mass, (kg)

Viscous
damping,
(N-s/m)

Spring
stiffness,
(N/m)

Coulomb
damping,

SEE (cm)

Original
model

0.9407

3.927

694.1

N/A

0.0829

Refined
model

0.9407

3.927

694.1

0.0133

0.0606

Figure 7 compares displacement, in centimeters, with respect to time, in seconds, for the step
responses of the initial model with the experimental data from the mass-spring-damper system.
The model response was generated using the parameters listed under Original model in
Table 3.
3
Model
Data

Displacement (cm)

2.5

1.5

0.5

0.5

1
1.5
Time (seconds)

2.5

Figure 7. Comparing the step responses of the initial model for a mass-spring-damper system
with the experimental data.

For the model to be usable, its response should reasonably match the data. Based on Figure 7,
the model does a poor job of representing the data past the second period of oscillation.
Graphically, the model and data appear to have very similar steady-state displacements and
peak times. However, settling time for the data is conspicuously less than for the model. In
addition, the model produces peaks which are noticeably higher than those of the data. This is
because the model does not take into account the Coulomb friction which was present when the
data was collected.
Figure 8 compares displacement, in centimeters, with respect to time, in seconds, for the step
responses of the refined model, which takes into account Coulomb friction, and the
experimental data. The model response was generated using the parameters listed under
Refined model in Table 3.

3
Model
Data

Displacement (cm)

2.5

1.5

0.5

0.5

1
1.5
Time (seconds)

2.5

Figure 8. Comparing the step responses of the refined model for a mass-spring-damper system
with the experimental data.
Based on Figure 8, the refined model response is a markedly better fit to the data. Graphically,
the model and data have very similar steady-state displacements, peak times and settling times.
The peak heights for the refined model and the data are acceptably matched over the entire
response. Compared to the initial model, the refined model reduces the SEE by 27%, calculated
in Eq. (15). These improvements indicate that taking into account Coulomb friction is sufficient
to generate an acceptably accurate model.

Conclusions
The goal of this experiment was to determine the physical parameters of the mass-springdamper system for which we collected data. This goal was achieved by incorporating estimates
of these parameters into a model and then improving the model until it fit the experimental
data. Our final physical parameters were a mass of 0.9407 kg, a stiffness 694.1 N/m, a viscous
damping factor of 3.927 N-s/m, and a Coulomb damping factor of 0.0133. The model based on
these parameters provided an SEE reduction of 27% over a model which neglected Coulomb
Friction. The final iteration of the model was a very close match to the data and provides an
excellent approximation of this systems physical behavior.

Вам также может понравиться