Você está na página 1de 10

Mechanical Systems and Signal Processing 68-69 (2016) 207–216

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing


journal homepage: www.elsevier.com/locate/ymssp

Low uncertainty method for inertia tensor identification


J.P. Barreto a, L.E. Muñoz b,n
a
Universidad de los Andes, Mechanical Engineering Department, Carrera 1 Este No. 19A-40, Bogotá, Colombia
b
Universidad de los Andes, Mechanical Engineering Department, Carrera 1 Este No. 19A-40, Bogotá, Colombia

a r t i c l e i n f o abstract

Article history: The uncertainty associated with the experimental identification of the inertia tensor can
Received 5 May 2013 be reduced by implementing adequate rotational and translational motions in the
Received in revised form experiment. This paper proposes a particular 3D trajectory that improves the experi-
11 April 2015
mental measurement of the inertia tensor of rigid bodies. Such a trajectory corresponds to
Accepted 19 April 2015
Available online 28 August 2015
a motion in which the object is rotated around a large number of instantaneous axes,
while the center of gravity remains static. The uncertainty in the inertia tensor
Keywords: components obtained with this practice is reduced by 45% in average, compared with
Inertia tensor identification those calculated using simple rotations around three perpendicular axes (Roll, Pitch, Yaw).
Stewart–Gough platform
& 2015 Elsevier Ltd. All rights reserved.
Uncertainty quantification

1. Introduction

The modeling, simulation, design and control of many mechanical systems require an adequate knowledge of the inertial
properties (i.e. mass, center of gravity and inertia tensor) [1,2]. For simple systems with few elements and simple shapes, the
center of gravity and inertia tensor can be estimated by means of CAD or theoretical models with an acceptable uncertainty.
However, for complex systems like ground vehicles, satellites, airplanes or ships, the use of CAD or theoretical models may
not be feasible. The large number of elements and uncertainty in their dimensions and composition increases the global
uncertainty as well as the time and effort required for generating an accurate model. In these cases, the inertial properties
have to be experimentally identified, and the required uncertainty bounds have to be determined for each application [3].
Several methods have been proposed for the measurement of the inertial properties of rigid bodies [4–6]. Determining
the mass has traditionally been done by measuring the weight and by estimating the gravity with standardized low levels of
uncertainty [7]. In order to identify the center of gravity, static and dynamic methods have been studied. Typically, static
methods result in low levels of uncertainty when identifying the two initial horizontal coordinates, yet identifying the
remaining vertical coordinate can be a complex task because the body has to be repositioned. The inertia tensor
identification is commonly made by generating motion on the body and by measuring the motion, forces and moments
acting on it. Depending on the experimental setup and procedure, the model and assumptions vary; some methods use the
general equation of motion while other assume small angular rotations or velocities. The most common methods require a
set composed by different experiments, each one involving rotational motion of the body relative to one fixed axis [8]. In
this case, the need for repositioning the body usually increases both time consumption and uncertainty levels. There are also
structural considerations, related to the high angular speed required to reduce the uncertainty [5].

n
Corresponding author. Tel.: þ57 1 3394949x1833.
E-mail addresses: pa-barre@uniandes.edu.co (J.P. Barreto), lui-muno@uniandes.edu.co (L.E. Muñoz).

http://dx.doi.org/10.1016/j.ymssp.2015.04.029
0888-3270/& 2015 Elsevier Ltd. All rights reserved.
208 J.P. Barreto, L.E. Muñoz / Mechanical Systems and Signal Processing 68-69 (2016) 207–216

Many modern methods have been developed with the aim of improving the measurement of the inertia tensor, by using
complex setups that allow rotation in different axes without repositioning the body. They involve mainly two different
techniques for the generation of motion: the action of gravity and the use of actuators. Experiments in which motion is
generated by gravity [9,10] have the advantages of a low cost experimental setup and applicability on large bodies; however,
their operation usually requires a high level of expertise and the repeatability of the tests is influenced by the initial
conditions. Experiments that suggest the use of actuators [11–13] improve the control on the rotational trajectory of the
body by using platforms that allow simultaneous rotations in different axes. Most of these systems perform a sequence of
rotations in mutually perpendicular axes for identifying the tensor. The general equation of motion can be exploited to
improve the identification by means of a complex trajectory. The definition of a proper trajectory and its implementation on
actuated platforms can lead to a reduction of the tensor's uncertainty.
This paper presents a two parts strategy for reducing the uncertainty when identifying the tensor by defining a complex
3D trajectory. The first part consists in keeping the center of gravity as static as possible during motion, which reduces the
propagation of the uncertainty associated to the location of the center of gravity. The second part of the strategy is making
the object rotate around a large number of uniformly distributed axes, describing a sphere of angular acceleration and
angular velocity. This particular trajectory has been shown to reduce the identification error when compared to commonly
used motions. In this paper, the development of the identification equations and the quantification of the uncertainty by
means of Monte Carlo simulations are presented, as well as the experimental validation of the method.

2. Nomenclature and definitions

2.1. Basic symbols

½a1 a2 a3  Components of the earth-fixed reference frame with origin in O


½b1 b2 b3  Components of the body-fixed reference frame with origin in Q
CG Center of gravity of the body
FB Force vector acting on the body (reference frame B)
MB Moment vector acting on the body (reference frame B)
ω ; ω
A B A _ B Angular velocity and angular acceleration vectors of B relative to A
 
rCG=p B Position vector from P to CG in reference frame B
B Rotation matrix from reference frame A to reference frame B
RA
ga Gravity acceleration vector in reference frame A
M Mass of the body
ICG Inertia tensor relative to the center of gravity in the body-fixed reference frame
In Column vector containing the six components of the tensor
OMB Center of the moving base of the platform
Q Origin of reference frame B (center of the load cell)

2.2. Mathematical definitions

The following definitions are useful for the formulation of the model and the development of the identification
equations:

2   3
ω_ 1 ðω
_ 2  ω1 ω3 Þ ðω_ 3 þ ω1 ω2 Þ  ω2 ω3 ω22  ω23 ω2 ω3 Matrix (3  6) of angular velocities and accelerations for
 2 
4 ω1 ω3
ΩR ¼ 6 ðω
_ 1 þ ω2 ω3 Þ ω3  ω21 ω_ 2 ðω_ 3  ω1 ω2 Þ  ω1 ω3 7
5 determining the moment acting on the body.
 ω1 ω2 ω21  ω2 2 ω_ 1  ω2 ω3 ω1 ω2 ω_ 2 þ ω1 ω3 ω_ 3
2 3
0 F3  F2 Matrix (3  3) of forces for determining the moment acting on the
6 F 7
Π¼4 3 0 F 1 5 body.
F2 F1 0

2.3. Inertia tensor

The inertia tensor is defined as a 3  3 symmetrical matrix that describes the distribution of the body mass relative to a
given point in a specific coordinate system. In this paper, the inertia tensor of a body relative to its center of gravity is
J.P. Barreto, L.E. Muñoz / Mechanical Systems and Signal Processing 68-69 (2016) 207–216 209

expressed, in the body-fixed system, as:


2 3
I11 I12 I13
6I I I 7
I CG ¼ 4 12 22 23 5 ð1Þ
I13 I23 I33

where the general term I ij is defined as:


!
R X
3
I ij ¼ m δij ζ k  ζ i ζ j dm
2

k¼1
(
1; if i ¼ j
δij ¼ ð2Þ
0; otherwise

where ζ r is the position vector from the center of gravity to a differential element of mass ðdmÞ in the direction br for r ¼1,
2, 3. The inertia tensor components can be rearranged as a vector (I n ) when the equations of motion are written in matrix
form. This structure will be useful for the identification.
h iT
I n ¼ I 11 I 12 I 13 I 22 I 23 I 33 ð3Þ

2.4. Uncertainty

In this paper the meaning of uncertainty presented in [14] is adopted as a “parameter associated with the result of a
measurement that characterizes the dispersion of the values that could reasonably be attributed to the measurand”. Given
the nature of the chosen uncertainty definition, two different levels of analysis were implemented. First, Monte Carlo
simulations were used to quantify the uncertainty based on the recommendations by [15] due to the complexity of the
model and the difficulty to provide its partial derivatives. For the analysis of Monte Carlo simulations, 3σ (three times the
standard deviation) is used as the index of uncertainty, to characterize the dispersion of the simulated “measurements” of
the inertia tensor. Second, to analyze the experimental results, the absolute value of the difference between the average
measured and expected values is reported and called “deviation”.

3. Identification procedure

3.1. System description and general equations

The body under study is attached to a Stewart–Gough platform through a six-axis load cell. Therefore, the only forces
acting on the body are the weight of the body, the aerostatic and aerodynamic forces, and the forces and moments measured
by the load cell (acting at Q). An inertial measurement unit (IMU) is also attached to the platform in order to measure the
angular velocity of the system. A simplified diagram of the system is presented in Fig. 1.
For the formulation of the model, it was assumed that the body under study is rigid and that the aerostatic and
aerodynamic forces are negligible. Consequently, the forces acting on the body are caused only by gravity and the load cell.
Two reference frames are used for the model: an earth-fixed frame (A) and a body-fixed frame (B), as presented in Fig. 1. The
equations of motion are written using the components in the body-fixed reference frame. The force equation presented in

Body C.G

b3
b1
Load Cell Q
IMU

Stewart-Gough
Platform
a3
a1

Fig. 1. Diagram of the system.


210 J.P. Barreto, L.E. Muñoz / Mechanical Systems and Signal Processing 68-69 (2016) 207–216

Eq. (4) relates the force vector on the load cell to the gravitational acceleration, the mass of the body, its orientation, and the
acceleration of its center of gravity.
h 
F B ¼ m aCG=o B  B RA g a  ð4Þ

The moment equation (5), about the center of gravity, arises from the sum of the moment directly measured by the load
cell (M B ) and the resultant of the forces (F B ) acting on the load cell origin (Q).

M B  r CG=Q  F B ¼ I CG A ω
_ B þ A ωB  ðI CG A ωB Þ ð5Þ

When the cross-products are solved, Eq. (5) yields the following equation:
2 3
I 11
6 I 12 7
2  3  6 7 2 3
ω_ 1 ðω
_ 2  ω1 ω3 Þ ðω_ 3 þ ω1 ω2 Þ  ω2 ω3 ω22  ω3 2 ω2 ω3 6 6 I 13
7
7 0 F3  F2
 
M B ¼ 4 ω1 ω3 ðω
_ 1 þ ω2 ω3 Þ ω3 2  ω1 2
6 ω_ 2 _ 3  ω1 ω2 Þ  ω1 ω3 7
ðω 6 7 6 F F1 7
56 I 7þ4 3 0 5r CG=Q
6 22 7
 ω1 ω2 ω1  ω2
2 2
ω1  ω2 ω3 ω1 ω2
_ ω2 þ ω1 ω3
_ ω3
_ 6
6I
7
7 F2 F 1 0
4 23 5
I 33

ð6Þ

3.2. Identification of the inertia tensor

The method for identifying the inertia tensor consists of generating a 3D motion on the body, and simultaneously
measuring the kinematic and kinetic variables in Eq. (6). The forces and moments are measured with a load cell, the angular
velocity is measured using an IMU, and the angular acceleration is computed as a numerical derivative of the angular
velocity. The center of gravity should also be determined before proceeding to identify the tensor. In order to determine
the components of the tensor, Eq. (7) is obtained by expressing Eq. (6) in matrix form by replacing the definitions of ΩR , I n
and Π:
M B ¼ ΩR I n þ Πr CG=Q ð7Þ

When motion is imparted on the body and data is recorded, every instant generates a system of equations as in (7),
where the six components of the tensor are unknown and only three equations are available. Consequently, at least two
different instants should be recorded. Since the experiment is dynamic, several points are available, and therefore,
redundant information is used to obtain a more precise estimation of the unknowns. When a set of n different instants is
stored, the resulting extended system is:
2 3 2 3 2 3
M B1 ΩR1 Π1 rCG=Q
6M 7 6Ω 7 6 Π2 rCG=Q 7
6 B2 7 6 R2 7 n 6 7
6 7¼6 7I þ 6
6
7
7 ð8Þ
6 ⋮ 7 6 ⋮ 7 6 ⋮ 7
4 5 4 5 4 5
M Bn ΩRn Πn rCG=Q

where M B (3  1) contains the components of the moment acting on the body through the load cell, ΩR1 (3  6) is composed
by the angular velocity and angular acceleration terms for that instant, and Π i (3  3) contains the components of the force
acting on the body through the load cell. The system of equations in (8) is written in a compact form as:

M ΩR In þ Π rd
b ¼d CG=Q ð9Þ

The components of the tensor that best fit the measured data are found by means of the least squares method [16], as
expressed in the following equation:
   1 h i
dT Ω
In ¼ Ω d ΩdT M b  Π d
r ð10Þ
R R R CG=Q

where
2 3 2 3 2 3
M B1 ΩR1 Π 1 rCG=Q
6M 7 6Ω 7 6Π r 7
6 B2 7d 6 R2 7 6 2 CG=Q 7
b ¼6 7 ΩR ¼ 6 7 Π rd 6 7
M 6 ⋮ 7 6 7 CG=Q ¼6 7
4 5 4 ⋮ 5 6 ⋮ 7
4 5
M Bn ΩRn Π n rCG=Q
J.P. Barreto, L.E. Muñoz / Mechanical Systems and Signal Processing 68-69 (2016) 207–216 211

4. Proposed trajectory for uncertainty reduction

Once the experimental setup and the model have been defined, it is now possible to define the trajectory for the
identification. Defining the trajectory requires two elements: the position of a point, and the orientation of the body as a
function of time. When Eq. (4) is replaced in Eq. (5), it is seen that the moments measured on the load cell are caused by
three different sources as shown in Eq. (11). One of them is the rotational motion of the body's inertia (first two terms of Eq.
(11)), another is the acceleration of the center of gravity (third term of Eq. (11)), and the last one is the weight of the object
and its orientation (last term of Eq. (11)). This breakdown leads us to propose a strategy for the reduction of the uncertainty
in the inertia tensor based on two parts: the first one intends to reduce the propagation of the uncertainty in the center of
gravity, and the second one seeks to improve the quality of the information gathered during the identification of the tensor.
   
M B ¼ I CG A ω
_ B þ A ωB  I CG A ωB þ r CG=Q  m aCG=o  r CG=Q  mB RA g a
B
ð11Þ

4.1. Translation of the center of gravity

The first part of the strategy is oriented to reduce the effect of the uncertainty in the location of the center of gravity
(r CG=Q ). This uncertainty increases the uncertainty associated to the inertia tensor because it is present in the last two terms
of Eq. (11). When the acceleration of the center of gravity (aCG=O ) is zero, the load cell will not sense a moment due to the
 
inertial force (r CG=Q  m aCG=O B ) and the uncertainty in the tensor due to the error in the center of gravity location will be
reduced. The proposed approach to achieve this is to reduce the motion of the center of gravity as much as possible during
the test, using the three-translational degrees of freedom of the platform in order to rotate the body around this point.
Accordingly, a relation between the acceleration of the center of gravity and the acceleration at the center of the platform is
stated. When the acceleration of the center of gravity is set to zero, the acceleration at the center of the moving base
becomes
   
aOMB =O B ¼  A ω
_ B  r CG=OMB  A ωB  A ωB  r CG=OMB ð12Þ

Once the acceleration at the center of the platform (aOMB =O ) is determined, it is rotated to the earth-fixed frame and
integrated twice in order to calculate the position of the center of the moving base (point OMB ). When the orientation and
position of the platform are known, the required lengths of the actuators are calculated using inverse kinematic analysis
[17]. Even though the point will not stand completely still because of the uncertainty in the location of the CG, control errors
 
and assembly inaccuracies of the platform, the reduction of the term r CG=Q  m aCG=o B will be significant.
Another important aspect in the reduction of the uncertainty associated to the tensor is the decrease in the uncertainty
in the location of the center of gravity itself; thus, to locate it, a method that is also focused on the reduction of uncertainty is
used. This method was presented in [18] and it also makes use of the Stewart–Gough platform to move the object and a six-
axis load cell to measure the forces and moments. Moreover, because both methods use the same experimental setup, it is
possible to identify the center of gravity and the inertia tensor using the same platform and sensor, and without reorienting
the body. This leads to a significant reduction in the duration of the experiment, compared to other methods.

4.2. Body rotation

The second part of the strategy is to collect more information about the inertia ellipsoid during the motion of the body.
Since the inertia ellipsoid describes the moment of inertia of an object as a function of its axis of rotation and the aim is to
identify all the inertia moments of the body (described by its inertia tensor), it would be useful to make the body rotate
around as many axes as possible. This will gather more information and improve the identification, as studied in [19].
The implemented solution was to move the body along a trajectory that generates angular velocity and angular
acceleration spheres. This allows studying a large number of axes (uniformly distributed) while maintaining a constant
magnitude. The angular acceleration and velocity spheres are shown in Fig. 2. The  ideal scenario consists on setting spheres
with large diameters (in order to increase the value of I CG A ω
_ B þ A ωB  I CG A ωB ). However, there is a limit imposed by the
kinematics and dynamics of the platform. A good compromise between uncertainty reduction and performance of the
platform was found using an angular velocity sphere with a diameter of 0.85 rad/s and an angular acceleration sphere with a
diameter of 8 rad/s2.
The angular acceleration sphere is generated using continuous functions, and the components of the angular velocity are
calculated as the integral of the acceleration. Afterwards, these components are multiplied by a sigmoid function in order to
guarantee a smooth motion of the platform. Finally, the angular acceleration is recalculated as the derivative of the speed
[18].
212 J.P. Barreto, L.E. Muñoz / Mechanical Systems and Signal Processing 68-69 (2016) 207–216

Fig. 2. Angular velocity (left) and acceleration (right) spheres for the identification of the tensor.

Table 1
Sources of error in the measurements.

Standard deviation (σ x ) of the precision error distribution Standard deviation (σ x ) of the bias error distribution

F [N] ½5:0 5:0 10 ½6:9 4:2 14  10  1


M [Nm] ½3:8 4:6 4:6  10 1
½1:4 1:6 1:9  10  2
ω[rad/s] ½4:3 3:7 19  10  4 ½5:8 6:0 5:9  10  3
rCG [mm] ½1:5 1:3 1:6  10 1 0

5. Uncertainty quantification

5.1. Estimation of the uncertainty associated to the proposed trajectory

Estimating the uncertainty in the inertia tensor consists on predicting the distribution of the population of measure-
ments of the tensor (or the error associated to them) conducted by means of the proposed method. For that purpose, it is
necessary to be aware or have a reasonable estimation of the distribution of the inputs, and to apply a method for estimating
the distribution of the output, such as the law of propagation of uncertainty or Monte Carlo simulations. In this paper, the
goal is to estimate the error in the population of measurements of the inertia tensor of a given body with the proposed
method. The inputs that need to be considered are the location of the center of gravity of the body, the forces and moments
measured by the load cell, and the angular velocity and angular acceleration (the former measured by the IMU and the latter
calculated as its derivative). Given the complex nature of the identification model, Monte Carlo simulations were used to
estimate the uncertainty, as recommended by [15]. For this purpose, a body with the inertial properties presented in this
equation was employed for the simulations
2 3
381:3 0 0
  6 7
m ¼ 19:7 kg rCG=Q ¼ 0b1 0b2 53b3 mm I CG gm2 ¼ 4 0 18:5 0 5 ð13Þ
0 0 379:1

In the simulations, the bias and precision error in the inputs have a normal distribution with mean zero standard
deviation σ x (see Table 1). The distribution of the bias and precision error in the load cell and the IMU were estimated from
preliminary measurements and from their data sheet, and the precision error in the center of gravity location was estimated
in [18]. In addition, the error in the angular acceleration depends on the error in the angular speed. The values of the
standard deviation of these distributions are presented in Table 1. For each variable, the standard deviation is expressed as a
vector with three components, since there is a different value for each axis of measurement.
The distribution of the error in the identification of the inertia tensor was estimated by carrying out 10,000 repetitions of
the identification procedure, using the body presented in Eq. (13) and the inputs in Table 1. The results of the simulations are
presented in Fig. 3. This figure illustrates the effect of each variable or group of variables on the uncertainty (3σ ) in the
tensor.
According to these simulations, the most influent variables on the identification of the tensor are the forces and moments
measured by the load cell. The uncertainty caused by the load cell corresponds, in average, to 97% of the total uncertainty.
The second most important variable is the location of the center of gravity; however, its effect is significantly lower than the
one of the load cell (20% of the total uncertainty in average). The kinematic variables (obtained from the IMU) are the less
influent variables, with 6% of the total uncertainty in average. It is important to notice that the effect of the inputs is not
additive (i.e. they do not sum 100%). This result suggests that although the load cell is the most influent source of
J.P. Barreto, L.E. Muñoz / Mechanical Systems and Signal Processing 68-69 (2016) 207–216 213

Total Due to Load Cell Due to CG Due to IMU


3

Uncertainty [gm ]
2
2

0
I11 I12 I13 I22 I23 I33

Fig. 3. Effect of each group of parameters on the inertia tensor uncertainty.

Fig. 4. Angular velocities used in simple motion experiments.

6 6
Total Due to Load Cell Due to CG Due to IMU
5 5
Uncertainty [gm ]
2

4 4

3 3

2 2

1 1

0 0
T1 T2 T3 T4 T1 T2 T3 T4

Fig. 5. Effect of each group of parameters on the uncertainty for different trajectories (Left: moments of inertia, right: products of inertia).

uncertainty, the other variables cannot be neglected for the uncertainty quantification. In fact, it is interesting to analyze
how their effect changes as the trajectory is modified. The following section presents the benefits of using the proposed
trajectory for estimating the inertia tensor.

5.2. Comparison to other trajectories

The proposed trajectory allows the study of a large number of instantaneous axes as opposed to the motion relative only
to the axes of the load cell. In order to compare the resultant uncertainty in the identification of the tensor, a simple
trajectory was also calculated. This trajectory consists of simple harmonic motions (SHM) in roll, pitch and yaw. The body is
rotated around these axes with the same angular velocity and angular acceleration magnitudes of the proposed trajectory.
This trajectory is similar to those used in common identification methods. It studies a large number of points while the body
rotates around only 3 mutually orthogonal axes (Fig. 4).
In order to analyze the effect of the proposed trajectory, four different trajectories were simulated. T1 and T2 correspond
to the sphere of angular velocity with a static and a moving center of gravity respectively. In T2, the point that remains static
during motion is the center of the moving base of the platform. T3 and T4 are harmonic motions in the three axes of the load
cell (Roll–Pitch–Yaw). In T3 the motion occurs relative to the center of gravity while in T4 the center of the moving base
remains static. In order to estimate the uncertainty and the effect of each group of variables, 10,000 repetitions of the
experiment were simulated for each case. Fig. 5 shows the average uncertainty caused by each group of parameters.
The results presented in Fig. 5 show that for all the analyzed trajectories, the groups of parameters with higher effect on
the inertia tensor uncertainty are the measurements obtained from the load cell and the location of the center of gravity.
The propagation of the uncertainty associated to the variables measured by the load cell was reduced by 45% when using T1
instead of T3 or T2 instead of T4. This result is obtained by using the sphere of angular velocity and angular acceleration. On
214 J.P. Barreto, L.E. Muñoz / Mechanical Systems and Signal Processing 68-69 (2016) 207–216

Table 2
Estimated uncertainty in the inertia tensor.

Trajectory Name 3σ (Moments) (gm2) 3σ (Moments) (% of NV) Difference with T4 (%) 3σ (Products) (gm2) Difference with T4 (%)

Sphere acm ¼ 0 T1 2.82 5.13  43.5 1.86  46.4


Sphere acm a 0 T2 3.15 6.15  36.9 1.97  43.2
SHM acm ¼ 0 T3 4.78 10.3  4.21 3.41  1.73
SHM acm a 0 T4 4.99 10.9 – 3.47 –

3σ : average uncertainty in the three products or the three moments.


NV: nominal value.
SHM: simple harmonic motions.

Fig. 6. Experimental setup.

Table 3
Experimental results for the inertia tensor using the proposed trajectory (T1).

I 11 I 22 I 33 I 12 I 13 I 23

Expected value ½gm2  382.89 18.72 380.66 0 0 0


Average measured value ½gm2  387.24 18.30 380.99 1.19 1.20 -1.78
Standard deviation of the measurements ½gm2  0.22 0.01 0.48 1.79 0.15 0.21
Deviation ½gm2  4.54 0.42 0.33 1.19 1.20 1.78
Relative error 1.2% 2.2% 0.1% — — —

the other hand, the propagation of the uncertainty associated to the location of the CG was reduced by 50% when using T1
instead of T2 or T3 instead of T4.
Table 2 summarizes the results of the simulations, the total uncertainty as a percentage of the nominal value, and the
variation in relation to T4. These results show the effect of the proposed trajectory on the uncertainty in the tensor, as
intended in Sections 4.1 and 4.2. With the use of T1 instead of T4, there is a 44% reduction of the global uncertainty
J.P. Barreto, L.E. Muñoz / Mechanical Systems and Signal Processing 68-69 (2016) 207–216 215

Fig. 7. Angular speed (left) and acceleration (right) spheres generated during the experiment.

Table 4
Average results for the experimental deviation with different trajectories.

Average deviation

Trajectory Name Moments (gm2) Products (gm2)

Sphere acm ¼ 0 T1 1.7 1.39


Sphere acm a 0 T2 3.72 1.6
SHM acm ¼ 0 T3 1.73 1.76
SHM acm a 0 T4 4.19 2.05

corresponding to the moments of inertia and 46% corresponding to the products. This reduction is the result of combining a
static center of gravity with a sphere of angular velocity and acceleration.
It is important to realize that this reduction and the estimation of the uncertainties may vary between different bodies,
and the uncertainty in the products cannot be expressed as a percentage, since they have a nominal value of zero, and their
magnitude depends on the coordinate system.

6. Experimental Results

In order to validate the method, the inertia tensor of a test body was measured using the proposed trajectory (T1). With
the purpose of having an accurate prediction of the inertial properties, a test body was designed with a simple shape, out of
a single material and manufactured precisely. The body was manufactured out of stainless steel to avoid corrosion or the
application of any coating that could alter the expected parameters. The body is a cylinder with a 76 mm diameter and a
500 mm length. The experimental setup is presented in Fig. 6.
The experimental results, that validate the proposed methodology for finding the inertia tensor, are presented in Table 3.
In these results, the average measured value corresponds to the mean of five measurements and the “deviation” is the
absolute value of the difference between the average measured value and the expected value.
The measured angular velocity and acceleration are presented in Fig. 7. As expected, the angular velocity and acceleration
trace a sphere.
Table 4 presents the comparison between the identification results with the proposed trajectory and the simple
trajectories presented in Section 5.2. These results show an improvement in the identification that is obtained when the
center of gravity acceleration is decreased (comparing trajectories T1 with T2 or T3 with T4). Moreover, the improvement
obtained when using the proposed spheres can be seen by comparing T1 with T3 or T2 with T4.

7. Conclusions

An experimental method for the identification of the inertia tensor focused on the reduction in the uncertainty of the
estimate was intended. The proposed strategy is based on the use of a 3D complex trajectory, which maintains the center of
gravity fixed while it gathers information about the inertia ellipsoid from different instantaneous rotations.
The uncertainty in the inertia tensor is reduced when the proposed strategy is implemented for the identification.
According to Monte Carlo simulations, with the proposed strategy, in average, for a given body there is a 44% reduction on
216 J.P. Barreto, L.E. Muñoz / Mechanical Systems and Signal Processing 68-69 (2016) 207–216

the uncertainty in the identification of the moments of inertia and a 46% reduction on the uncertainty in the identification of
the products of inertia. This reduction is based on a comparison to an identification in which the body is rotated in simple
harmonic motions around three orthogonal axes.
The identification method was validated experimentally. It was found that the order of magnitude of the difference
between the average measured value and the expected value is the same of the obtained in the simulations. The typical
deviation on the principal moments of inertia is 1.7 gm2 and the typical deviation on the products of inertia is 1.39 gm2,
while for a traditional trajectory (simple harmonic motion with movement of the center of gravity) the deviation is 4.19 gm2
and 2.05 gm2 for the moments and the products, respectively.

References

[1] G. Mastinu, M. Gobbi,C. Miano, The influence of the body inertia tensor on the active safety and ride comfort of road vehicles, in: Proceedings of the
SAE-IBECD Conference, Paris, 2002.
[2] P. Eberhard, W. Schiehlen, J. Sierts, Sensitivity Analysis of Inertia Parameters in Multibody Dynamics Simulations, IFToMM World Congress, Besancon,
2007.
[3] L. Munoz, M. Gobbi, G. Mastinu, Uncertainty bounds of inertia properties required for vehicle dynamic analyses, in: Proceedings of the 19th
International Conference on Design Theory and Methodology; 1st International Conference on Micro- and Nanosystems; and 9th International
Conference on Advanced Vehicle Tire Technologies, Parts A and B, Vol. 3, 2007, pp. 1009–1018.
[4] C. Schedlinski, M. Link, A survey of current inertia parameter identification methods, Mech. Syst. Signal Process. 15 (1) (2001) 189–211.
[5] R. Boynton, K. Wiener, Mass properties measurement handbook, SAWE Paper No. 2444, in: Proceedings of the Annual Conference of the Society of
Allied Weight Engineers, Wichita, Kansas, 1998.
[6] G. Genta, C. Delprete, Some considerations on the experimental determination of moments of inertia, Meccanica (1994) 125–141.
[7] R. Boynton, Precise Measurement of Mass, in: Proceedings of the 60th Annual Conference of the Society of Allied Weight Engineers, Arlington, Texas,
2001.
[8] M. Bacaro, F. Cianetti, A. Alvino, Device for measuring the inertia properties of space payloads, Mech. Mach. Theory 74 (2014) 134–153.
[9] C. Doniselli, M. Gobbi, G. Mastinu, Measuring the inertia tensor of vehicles, Veh. Syst. Dyn. Supplement 37 (2002) 301–313.
[10] M. Gobbi, G. Mastinu, G. Previati, A method for measuring the inertia properties of rigid bodies, Mech. Syst. Signal Process. 25 (2011) 305–318.
[11] H. Hahn, M. Niegerball, Development of a measurement robot for identifying all inertia parameters of a rigid body in a single experiment, IEEE Trans.
Control Syst. Technol. 9 (2) (2001).
[12] R. Brancati, R. Russo, S. Savino, Method and equipment for inertia parameter identification, Mech. Syst. Signal Process. 24 (2010).
[13] V. Bogdanov, N. Veselov, V. Parshev, I. Panchenko, V. Parshev, V. Petronevich, E.K. Chumachenko, Test rig for measuring the object's mass, center of
gravity coordinates and inertia tensor, Autom. Remote Control 72 (2) (2011) 425–434.
[14] Joint Committee for Guides in Metrology, evaluation of measurement data – guide to the expression of uncertainty in measurement, JCGM 100:2008,
2008.
[15] Joint Committee for Guides in Metrology, evaluation of measurement data – supplement 1 to the “Guide to the expression of uncertainty in
measurement” – propagation of distributions using a Monte Carlo method, JCGM 101:2008, 2008.
[16] A. Bjorck, Numerical Methods for Least Squares Problems, SIAM, Philadelphia, 1996.
[17] J. Barreto, L. Munoz, Inertia parameter identification using a Stewart Platform, in: Proceedings of the International Symposium on Robotics – Robotik,
Munich, Germany, 2010.
[18] J. Barreto, L. Munoz, Uncertainty reduction on the identification of the inertial properties of rigid bodies, in: Proceedings of the American Society Of
Mechanical Engineers Conference on International Mechanical Engineering Congress and Exposition, ASME IMECE, Vancouver, Canada, 2010.
[19] L. Munoz, On the Measurement of the Inertia Properties of Rigid Bodies, Politecnico di Milano, Milano, Italy, 2008 (Ph. D. dissertation).

Você também pode gostar