P O Z N A N UN I VE RS I T Y O F T E C HN O L O G Y ACA D E MI C J O URN A L S

No 77 Electrical Engineering 2014

Stefan BROCK*
Maciej GNIADEK*

ROBUST INPUT SHAPING FOR TWO-MASS SYSTEM

WITH VARIABLE PARAMETERS

Various mechanical systems are characterized with flexible joints. Exemplary this class
of systems are the conveyers motors, the industrial robots and the cranes. The mechanical
properties of those objects can be approximated by two-mass system. The paper presents
the time-optimal input shaping methods for two-mass system. The control signal has to
reach the set value of displacement in the shortest time, without of rising the mechanical
oscillations. It was established, that the summary moment of inertia is constant, but division
of mass is changing. For this case robustness for wide range of parameters deviations was
analyzed. The model included PMSM motor with limited dynamics of current regulation.
The research were conducted in Matlab/Simulink enviroment.

KEYWORDS: two mass system, input shaping, robustness

1. INTRODUCTION

Many mechanical systems in can be characterized with resonant characteristics.

This kind of systems is represented by the cranes [1], multi-mass systems [2][3], the
conveyers and other systems.
Usually one of the requirements for the control system is to drive the machine
without of vibrations. One of the control method used to complete this requirement is
input shaping. Very often the jerk is limited, what effects with oscillations limitation,
but leads to significant enlargement of reaction time. The paper shows another method,
based on input shaping leading to prevent the oscillations. Simultaneously the
dynamics parameters of driving system are maintained. The subject of the paper is the
analysis of the robustness of this types of control methods for wide range of parameters
deviations.

2. SIMULATION MODEL

The research was based on two-mass system with flexible joint. The system is
shown on Figure 1.

__________________________________________
* Poznan University of Technology.
210 Stefan Brock, Maciej Gniadek

Fig. 1. Two-mass system

To build the simulation model of system it is necessary to have the physical

equations describing it. The torque motor rotor Jm is equal to:
Tm  Te  Ts , (2.1)
where Te is the electromagnetic torque and TS is the torque transmitted through the
shaft. For the load side mass, analogically:
Tl  TS  TL , (2.2)
where Tl is the torque of second mass, TL is the total torque of load including
friction. The torque TS can be calculated from equation
 d d 
TS  kS  ( L  m )  d S   L  m  , (2.3)
 dt dt 
where θL and θm are the positions of both masses, and kS and ςS are the coefficients
of springiness and damping of the shaft respectively. The motor side and load side
position are determined from:
d 2 m Tm
 (2.4)
dt 2 Jm
and
d 2 L Tl
 . (2.5)
dt 2 JL

3. INPUT SHAPING

3. 1. Command generation

The research presented in following chapters are based on impulse commands for
vibration reduction. The base of command generation is to replace a single burst of
force with multiple lower bursts, that summary supplies same amount of energy, but in
smaller portions, applied in proper time. Each input signal can be calculated and
shaped as a convolution of determined impulses. The amplitudes and times are possible
 Robust input shaping for two-mass system with variable parameters 211

to calculate, if the basic characteristic values (like Td – period of damped vibration and
ωn – natural vibration pulsation) of the system are known. If the ratio R of load inertia
to motor inertia is defined R=JL/Jm then the natural vibration pulsation is determined
by
kS
 n  a 1  R  1 R (3.1)
JL
where ωa is the anti-resonance pulsation. The damping of the system is related to
internal damping of the shaft ds.
dS (1  R )
  (3.2)
2 n J L
According to equations presented in [1], the times and amplitudes of impulses can
be presented as:
 Ai   1 K 
 t   1  K 1 K  (3.3)
 i   0 0 ,5Td 
where K is
 
( )
1 2
K e (3.4)
If the model of two-mass system is known, the ωd and ωn are possible to
calculate [2]. Equations shown above will provide the time-optimal control for a
specific Td, but if any of parameters will be changed, the control will not be optimal
any more. This problem can be solved by modifications of the control algorithm. The
modifications are shown in chapters 3.2 and 3.3. Robustness of the solution is shown in
part 4.

3.2. Robust input shaping

The solution shown in part 3.1 is working really good if every part of process does
not contain any uncertainties or inaccuracies. If the resonant frequency would change,
one of the impulses will be set in not exact time or with not exact amplitude the whole
process of control without of vibrations will fail. Additional problems, like uncertain
parameters, will cause vibrations of whole system. To omit this problem it’s necessary
to use the robust version of the control algorithm. The first algorithm of robust control
with input shaping was developed by Singer, Signhose and Seering [4]. The shaper was
designed by requiring the partial derivative of the residual vibration to be equal to zero
at the modeling frequency.
Enforcing this constraint has the effect of keeping the vibration near zero when the
frequency starts to differ from the modeling frequency. The robust shaper is the
simplest way to solve problem of uncertain parameters of the system. If all of the
212 Stefan Brock, Maciej Gniadek

parameters are nearly constant during the system exploitation and research this solution
is usually sufficient to control the system with insignificant vibrations.
The robust shaper can be described with equations:
 1 2K K 
 Ai   2 

 t  ( 1  K )
2
( 1  K )2 (1 K )  (3.5)
 i  0 0 ,5Td Td
 
The shaper described with equation (3.5) will cause vibrations at level not bigger
than 5% for modeling frequency [1], but will provide robustness of the control.

3.3. Super-robust input shaping

The solution shown in part 2.2 works well if the parameters of the systems are not
changing significantly during the exploitation. To omit this limitation the super-robust
shaper has to be used.
The situation, when parameters of the system are changing in a wide range can be
found in various real systems. For example the moment of inertia of an empty and fully
loaded crane can be even hundreds time smaller. When the moment of inertia is
changing also the resonant frequency is varying. In situations like this the solutions
shown in parts 3.1 and 3.2 will be insufficient. The method of projecting the super-
robust shapers is called “multi-hump EI shapers”. The input is divided into many small
impulses convolution what effects with broaden the range of proper work of the control
system. The main problem of the method is to select the frequency, that will be treated
as the basis (modeling) frequency. The frequency of resonations will be changing with
the parameters. The modeling frequency can be selected for example as the middle of
the expected range. If the damping coefficient is very low, the super-robust shaper can
be described with equation:
 Ai  0.125 0..375 0 ,375 0.125 
t    0 0.5Td Td 1.5Td 
(3.6)
 i 
where Td is the period of modeling frequency.

4. SIMULATION TEST

The simulation tests were executed at the model described in section 2. To check
the robustness of shapers the resonant frequency of the system has to be modified. To
change the frequency the moments of inertia has to be modified. The sum of moments
of inertia is constant and equal to the moment of inertia of single mass system. The
moments of inertia are equal:
1
J M  JT (4.1)
R1
 Robust input shaping for two-mass system with variable parameters 213

R
J L  JT (4.2)
R1
where JL is the moment of inertia of load, JM is the moment of inertia of the machine
and JT is sum of moments of inertia of both masses. R is the coefficient of mass
division.
The test was based on supplying the torque to the system. The torque was selected
to obtain the time-optimal movement for the reference system with non-flexible joint
between the masses. The system was modeled as a single, rotating mass, characterized
by the moment of inertia JT. The control task was to rotate the mass by 40 rad, by speed
limitation to 100 rad/s. The movement can be divided into three parts: acceleration,
full-speed driving and deceleration. The reference and dual-mass characteristics were
compared.
Figure 2 shows the input torque characteristics for selected input shapers. The time
of movement was extended from 0.2 s. (for simple input shaper) up to 0.8 s (for the
robust shapers).

Fig. 2. Input torque for shapers

The shapers were calculated for the nominal situation, where the moments of inertia
of drive and load are equal (R = 1). Afterwards, the solutions robustness for mass
division changes was checked in range between R = 0.2 up to R = 5.
Figure 3 shows the characteristics for R = 1. The input shaper works properly, any
significant oscillations do not appear. At right column the final phase of movement was
shown. For the simple input shaper some small oscillations appear, for robust and super
robust shaper oscillations are equal to zero. To compare the results, figure 4 presents
the R=5 case. For every shaper oscillations appear, but for the robust and super-robust
shapers are less and less significant. For R = 0.2 same conclusions can be set.
As the criteria of robustness the maximal difference between position of the
machine and the load and the single mass was selected. The characteristics are shown
at figure 5.
214 Stefan Brock, Maciej Gniadek

The characteristics shown on figure 5 confirms the theoretical deliberations

described in part 3. The differences between all of strategies for the model frequency
are not significant. If the frequency is deviating, the robustness is becoming more
important. For both criteria the super-robust strategy is giving the best results. The
simplest strategy will be sufficient for clearly specified model without of uncertainties.
Applying the robust and super robust input shapers insignificantly enlarges the
movement time, but the oscillations amplitude reduction is pointing.

Fig. 3. Speed and position characteristics for various input shapers in case R = 1;
a) whole movement, b) final phase

Fig. 4. Speed and position characteristics for various input shapers in case R = 5;
a) whole movement, b) final phase
 Robust input shaping for two-mass system with variable parameters 215

5. CONCLUSION

The research has confirmed high robustness of presented methods of the input
shaping. Those methods can be characterized by a small computational complexity and
simple on-line realization. Approximated parameters of objects are sufficient to project
the shaper. Shaping of closed-loop regulation signals for selected objects will be the
subject of future research.

Fig. 5. Difference of the angular position between machine, load and single mass

REFERENCES

[1] W. Signhose and S. Warren, Command Generation for Dynamic Systems. 2011.
[2] K. Szabat, “Direct and indirect adaptive control of a two-mass drive system—a
comparison,” in Industrial Electronics, 2008. ISIE 2008. IEEE International
Symposium on, 2008, pp. 564–569.
[3] N. Avdiu and S. Buza, “Analysis of the mutli-mass drive system synamics with
induction motor,” Prague, Sep. 2011.
[4] W. Signhose, N. Singer, and W. Seering, “Preshaping command inputs to reduce
system vibration,” vol. 1212, pp. 76–82, 1990.