Académique Documents
Professionnel Documents
Culture Documents
High-Rise Elevators
Joel Fortgang, Vlad Patrangenaru, and William Singhose
0 * 0 Δ 0Δ
-Amax
taccel tcoast tstop Fig. 4. Convolution Process for Input Shaping
Percentage Vibration
2
20
1.5
Acceleration (m/s )
2
1 15
0.5 10
0
5
-0.5
-1 0
0.6 0.8 1 1.2 1.4
-1.5 Normalized Frequency (ωa/ωm)
-2 Fig. 5. Sensitivity Curve for Various Input Shapers
0 5 10 15 20 25 30
Time (sec) [9], [10], and coordinate measuring machines [11]–[13]. It
Fig. 3. Acceleration Response of Elevator Cab to Bang-Coast-Bang Input has also proven effective in conjunction with a variety of
feedback control schemes [14]–[16].
where mdr , bdr , and kdr are values per unit length of cable. A wide variety of input shapers have been developed for
The dynamics of the other cables are analogous to (2). diverse applications. One common input shaper is the Zero
The specific model used here to investigate the proposed Vibration (ZV) shaper [2], [17]. This shaper has the shortest
solution is based on a 135-story building with a cab mass of duration using only positive impulses that results in zero
3400 kg. The upper-right drive cable varies by 494 meters vibration for a perfectly modeled linear system. Input shaper
and its mass changes from 17 kg to 4310 kg resulting in duration is important because the convolution with the input
a 17,000% change in bur and kur between the top and shaper increases the rise time of the command, as was shown
bottom floors of the building. These changes are offset by in Figure 4. For a ZV shaper this increase in rise time is one
corresponding changes in the parameters of the other cables. half the period of vibration it is designed to attenuate.
The input to the system is the angle of the drive sheave, A ZV shaper will cancel all vibration if it is perfectly
Θin . Figure 2 shows the baseline bang-coast-bang accelera- designed to match the physical system. However, if there
tion command used here. It has an acceleration limit, Amax , is a modeling error, then some vibration will occur. Figure
and a velocity limit, Vmax . This command is the fastest 5 shows the residual vibration magnitude resulting from
for any given velocity and acceleration limits. However, this input shaping, as a function of frequency modeling error.
command profile can lead to large cab vibrations, as shown in The frequency axis is normalized by dividing the actual
Figure 3. The response shown is for a move of 15 floors. The frequency, ωa , by the modeled frequency, ωm . If robustness
resulting overshoot at the desired floor is 5 cm. Other move to modeling errors is needed, then robust shapers such as the
distances generate similar vibration problems. The input- Zero Vibration Derivative (ZVD) [2] or the Extra-Insensitive
shaping control scheme developed here seeks to slightly (EI) [18] input shapers can be used. The sensitivity curves
modify the bang-coast-bang command so that vibration is for these shapers are also shown in Figure 5. It is obvious
attenuated, but high-speed motion is still attained. that the robust shapers keep the vibration at a low level over
a much wider range of frequencies. The cost of the added
B. Input Shaping
robustness of ZVD and EI shapers is an increase in the shaper
Input shaping is a technique by which the reference duration to one period of the vibration.
commands are altered to reduce vibration. For a known set
of vibratory modes there exists a set of impulses, called an C. Vibration Absorbers
input shaper, that excites minimal vibration when applied to Vibration absorbers are auxiliary mechanical systems that
the system. This input shaper can be convolved with any can be affixed to a system to reduce vibration. They act by
baseline input to produce a command that will induce low modifying the system dynamics to introduce a new mode
vibration comparable to the input shaper. Figure 4 shows of vibration that improves the response [19]. One of the
how this process works [2]. earliest applications of this principle occurred inside the
Although input shaping has not been previously applied British warship the HMS Inflexible in 1883. A volume of
to elevators, it has been implemented on numerous systems water in the hull was used as an oscillating mass [20]. The
such as long-reach manipulators [3], cranes [4]–[8], satellites deck motion resulting from the near-sinusoidal forcing of the
Shaper1
Shaper2
Amax
2
Shaper4
0
0 20 40 60 80 100 120 Shaper3
Floor
-Amax
Fig. 6. Linearized Frequency Response of Elevator Cab
taccel+Δtaccel tcoast+Δtcoast tstop+Δtstop
ocean swells was reduced by the counteracting motion of the Fig. 7. Floor-Scheduled Input-Shaped Bang-Coast-Bang Input
water inside the hull.
The early theoretical work on vibration absorbers only 1.2 No Shaper
considered the ideal undamped case. With the advent of Single ZV Shaper
0.8
computational techniques, it is now possible to design an
Acceleration (m/s )
0.4
2
absorber for a variety of situations and criteria, including
damping and non-linearity in both the absorber and primary 0
system [21]. Implementation of multiple absorbers [22] and
-0.4
active absorbers [23] whose parameters are controlled by an
external source have also been developed [24]. -0.8
Vibration absorbers have been used in a wide range of -1.2
applications, including architecture; rotational machinery;
and consumer goods [20], [25]. Absorbers are usually de- -1.6
signed to cope with periodic excitations, rather than transient -2
effects. Although, absorbers have been designed for random 20 24 28 32
excitations [26], as well as for the “time optimal case” [27] Time (sec)
defined by energy dissipation for an impulse response. The
step-like trajectories and disturbances present in elevators Fig. 8. Shaped and Unshaped Responses for a Move from Floor 30 to 50
have not yet been rigorously explored. However, a properly
tuned vibration absorber can be used to improve the settling B. Command Profile Modification
time of a step-disturbed lightly-damped system by up to 90% Two input shaping methods are investigated here. The
[28]–[30]. first approach is traditional input shaping which uses a
II. I NPUT S HAPING FOR E LEVATORS single input shaper to filter the entire command. The second
approach uses four different input shapers to better cancel the
In order to reduce elevator cab vibration, a novel type
vibration and produces the overall command profile shown
of input shaping is employed here. Input shaping requires
in Figure 7. The system is accelerated up to Amax using
knowledge of the frequency characteristics of the elevator.
Shaper1 . Then, the system returns to zero acceleration using
Therefore, a necessary first step is to analyze the elevator
Shaper2 , thus beginning a coasting period at maximum
for position-dependent frequency information.
velocity. The deceleration follows the same pattern using
A. Frequency Analysis of Elevators Shaper3 and Shaper4 . These shapers are selected by a
Frequency data of the nonlinear elevator model was ob- scheduling algorithm based on the elevator’s position.
tained by linearizing the cable parameters about possible
operating positions. It is important to establish the relation- C. Single Input Shaper
ship between the height of the elevator and the associated Standard input shaping provides a quick and simple so-
dominant linearized frequency in order to properly design lution to neutralize most of the stopping vibrations in the
an input shaper for a given motion. This frequency data is elevator. By shaping the entire bang-coast-bang accelera-
shown in Figure 6. The color variation is the magnitude of tion profile with a single ZV shaper designed to suppress
the frequency response. This surface is used to determine the dominant frequency at the stopping floor, the residual
the dominant linear frequency at each floor of the building. vibrations can be greatly reduced. This is demonstrated
The major trend is that the dominant frequency increases as for an example 20-floor move in Figure 8. The maximum
the elevator rises, because a shorter cable length between acceleration magnitude is reduced from 2.0 sm2 to 1.1 sm2 .
the drive sheave and the cab leads to higher frequency Input shaping is not completely effective at canceling
oscillations. vibrations because of the change in the dominant frequency
1.2 1.2
0.8 0.8
Acceleration (m/s )
Acceleration (m/s )
2
2
0.4 0.4
0 0
-0.4 -0.4
-1.6 -1.6
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
Time (sec) Time (sec)
Fig. 9. ZV and ZVD Shaped Response for a 35 Floor Move Down Fig. 10. Single Shaper and Gain Scheduled Shaper Response for a 35
Floor Move Down
of the cab. This is especially true for long moves where the ZV shapers are used for the first three changes in the
frequency of the elevator at the stopping floor is significantly acceleration command, while a ZVD shaper is used to bring
different that at the starting floor. This problem can be the elevator to a stop. The inherent robustness of the ZVD
reduced by using an input shaper that is more robust to fre- shaper better cancels the nonlinear effects of the system,
quency variations, like a ZVD shaper. This incurs the expense thus leading to lower residual vibration. However, since the
of a small increase in the move duration. Figure 9 shows how transient vibrations are not as important, and fast motion is
the increased robustness of the ZVD shaper cancels more cab critical, ZV shapers are used for the first three acceleration
vibration. Specifically, the peak acceleration is reduced by changes. When using this scheduled approach, the results
21% by switching to the ZVD shaper and the displacement are the same if the final floor is known a priori or if the
magnitude is decreased by 56%. desired floor changes during the motion, as would occur if
One drawback of this standard, single-shaper approach is a passenger pushed the button for another floor while the
that since the shaping scheme relies on the landing floor, a elevator is in motion.
mid-move change in the desired floor will result in increased This scheduled input shaping process improves perfor-
vibration when the elevator stops earlier than expected. mance over the single-shaper scheme as shown in Figure
Therefore, using a scheduled input shaping scheme can be 10, where the vibrations during the first portion of the
beneficial because it can overcome this drawback. move have been greatly reduced. To determine the overall
effectiveness of the proposed input shaping scheme, one-
D. Scheduled Input Shaping hundred and twenty random moves, in both the up and
the down directions, were simulated with and without the
While single shapers are effective at minimizing land- gain-scheduled input shaping. The maximum overshoot of
ing vibrations of the cab and provide vibration reduction the cab position was computed for each case, and the
throughout the move, there is a better solution that com- resulting percent improvement in peak overshoot is shown
pensates for the nonlinearity of the system. This solution in the histogram of Figure 11. Percent improvement in peak
minimizes vibrations for the entire elevator trajectory. This overshoot, MP , is defined as:
is done by using separate input shapers for each acceleration MP unshaped − MP shaped
change, as was shown in Figure 7. Each shaper is designed %Imp(MP ) = (4)
MP unshaped
for the elevator dynamics at the time of the acceleration
change. The following technique is used to create the shaped where MP unshaped and MP shaped are the peak landing over-
acceleration profile: shoot for the corresponding unshaped and shaped elevator
moves. On average, a 97 percent reduction in peak overshoot
1) The instantaneous dominant natural frequency of the is achieved. The cost of this improvement is an average
elevator at each acceleration change of the bang-coast- increase in the move time of only 0.67 seconds. In reality,
bang acceleration profile of Figure 2 is found using elevator systems utilize less aggressive S-shaped acceleration
data such as that in Figure 6. profiles that take longer than the bang-coast-bang profile used
2) These frequencies are used to design four separate here. With the use of input shaping these S-shaped profile
input shapers to modify each acceleration change. can be made significantly more aggressive.
3) ∆taccel and ∆tstop are computed to satisfy the maxi-
mum velocity constraint. III. V IBRATION A BSORBERS FOR E LEVATORS
4) Finally, ∆tcoast is used to adjust the cruise time so the Input shaping is an effective way to reduce predictable
elevator will stop at the desired position. elevator vibrations; however it cannot account for vibrations
60
50
40
Counts
30
20
15
10
10
0
0 20 40 60 80 100
Percent Improvement in MP 5
Fig. 11. Peak Overshoot Percent Improvement Using Scheduled Shaping 0
120 120
0 No Absorber 100 100
Hartog Absorber 90 80 80
90
-1 Optimized Absorber 60 60
Operating 40 40 Design
-2 Floor 20 Floor
Position (mm)
20
-3 0 0
-4
Fig. 13. Settling Time for a Variety of Absorber and Elevator Locations
-5
30
-6
25 No Absorber
-7
With Absorber
-8
0 1 2 3 4 5 6 Settling Time (sec) 20
Time (sec)
15
Fig. 12. Disturbance Response of Elevator with Various Absorbers
10
caused by disturbance forces, such as a child jumping
inside the elevator. To deal with this problem without using
5
feedback control on the cab position, a vibration absorber
can be utilized. A properly selected vibration absorber can
0
significantly reduce the duration of the transient vibrations 20 40 60 80 100 120
induced by a step disturbance force, as shown in Figure Floor
12. The figure shows the step disturbance response of the Fig. 14. Settling Time Comparison With and Without Chosen Absorber
elevator and the elevator equipped with an absorber designed
using Hartog’s classic approach [19]. The absorber uses
the disturbance occurred, as well as the floor for which
a 10% absorber mass. The response with an optimized
the absorber was designed. From Figure 13 it is clear that
absorber is also shown. This absorber outperforms the Hartog
operating near floor 90 is the most problematic regardless of
absorber because it is designed using techniques specially
which absorber is used. This result indicates that the absorber
developed for step disturbances [30]–[32], whereas Hartog’s
optimized for the 90th floor would be the best choice because
absorber is designed for sinusoidal disturbances.
it addresses the worst problem. This absorber also reduces
A. Absorber Design the settling time of the cab while operating at other floors,
Vibration absorbers, like input shapers, can reduce vi- as can be seen in Figure 14. The solid line on Figure 14
brations occurring at specific frequencies. The challenge represents a cross section of Figure 13, while the dashed
here is selecting which frequency to use in the absorber line is the response without the absorber.
design because the absorber parameters are fixed, unlike the
IV. C OMBINED I NPUT S HAPING AND V IBRATION
scheduled input shaper parameters which can easily vary
A BSORBERS
with elevator height. Here the goal of finding an effective
absorber for an elevator was simplified by only considering Position-dependent input shaping control and thoughtfully
the use of step-disturbance-designed absorbers [32]. designed vibration absorbers are complementary solutions.
The design procedure began by creating an absorber to Input shaping is able to deal with predictable motion-induced
attenuate the dominant frequency at each floor in the build- vibrations, but it does not address disturbance forces. If a
ing. Absorber effectiveness was measured by the settling vibration absorber is added to reduce the effects of distur-
time of the cab in response to a disturbance. Each absorber bances, then there are no significant differences in the input-
was tested at each floor throughout the building. Figure 13 shaped commanded-motion response, as shown in Figure
shows the settling times as a function of the floor on which 15. Note, that the elevator equipped the absorber actually
1.2 [9] T. Tuttle and W. Seering, “Experimental verification of vibration
Shaped with Absorber reduction in flexible spacecraft using input shaping,” J. of Guidance,
Control, and Dynamics, vol. 20, no. 4, pp. 658–664, 1997, command
0.8 Shaped without Absorber
Acceleration (m/s )
Shaping.
2