Anda di halaman 1dari 81

Sensors & Actuators

Velocity and acceleration

2014-2015 Sensors & Actuators - H.Sarmento


Outline

• Velocity sensors
• Gyroscopes
• Accelerometers

2014-2015 Sensors & Actuators - H.Sarmento 1


Velocity and acceleration
• Position, velocity, and acceleration are all related.
– Velocity is how fast an object is moving: the first derivative
of the position.
dy d
vy  
dt dt
– Acceleration is how fast an object's speed is changing: the
second derivative of the position.
2 d  d 2

ay 
dv y d y
 2 a  2
dt dt dt dt
• In a noisy environment, taking derivatives may result in
extremely high errors.

2014-2015 Sensors & Actuators - H.Sarmento 2


Velocity measurement

• Velocity is a vector that consists of a magnitude (speed) and a


direction.

• Many velocity or acceleration sensors contain components that


are sensitive to a displacement.

• However, velocity can also be measured by direct sensors.

2014-2015 Sensors & Actuators - H.Sarmento 3


Linear velocity sensors

• Linear velocity of solids:


– Linear velocity transducers (LVT)
– Doppler radar sensors.
• Linear velocity of fluids:
– Particle image
– Laser Doppler
– Thermal anemometer
– Pitot probes

2014-2015 Sensors & Actuators - H.Sarmento 4


LVT (1)
coil 1 coil 2
• LVT consists of:
N S
– A core (a permanent magnet).
– Two electrical coils.
VT

• The two coils are wrapped with opposite polarity. The south pole of
the magnet induces a voltage primarily in coil 2, and the north pole
primarily in coil 1.

• Moving a magnet through a coil of wire will induce a DC voltage


(emf) in the coil according to Faraday's Law:
d
emf   N
dt
2014-2015 Sensors & Actuators - H.Sarmento 5
LVT (2)

• The induced DC voltage is proportional to the magnet's


velocity and field strength.
coil 1 coil 2

VT  B l v
N S

VT
[Source: Trans-tek]

B - component of the flux density normal to the velocity


l - length of the conductor
v - velocity

2014-2015 Sensors & Actuators - H.Sarmento 6


LVT (3)

[Source: Trans-tek]

• Working range: detection of velocity along a distance limited by the


size of the sensor
• The DC voltage is relatively independent of position within some
limited range near the center.

2014-2015 Sensors & Actuators - H.Sarmento 7


Commercial LVTs

[Source: Trans-tek]

• The core slides inside a hollow cylindrical tube.

2014-2015 Sensors & Actuators - H.Sarmento 8


LVT applications
• Shock Absorber Testing Machine
LVTs used to rate damper performance, which leads to a better shock
absorber design.
• Injection Molding Machine
LVTs used to monitor the injection rate of molten plastic flow.
• Glass Pipette Pulling
LVTs used to control the pull rate of molten glass to form pipettes used in
most laboratories.
PC Board Drilling
LVTs used to control velocity in PC Board drilling applications.

2014-2015 Sensors & Actuators - H.Sarmento 9


Doppler radar velocity measurement
• When radio waves strike a moving object, the frequency of the
reflected radio waves is altered.
2v cos 
f D 

f D 
v
[Source: J. M. Cimbala] 2 cos 

 – wavelength of RF incident waves in a moving object.


v – velocity of the moving object at angle  relative to the radar unit.
fd – shift in frequency of the reflected wave relative to the transmitted wave.

2014-2015 Sensors & Actuators - H.Sarmento 10


Particle image velocimetry (1)

• Velocity measured following tiny particles that move with the


fluid.

PIV
(Particle Image Velocimetry)

[Source: J. M. Cimbala]

2014-2015 Sensors & Actuators - H.Sarmento 11


Particle image velocimetry (2)

• A double-pulse laser illuminates a region of flow under study,


and a digital camera records two images timed with the two
pulses of laser light.

• Illuminated particles appear as bright spots on the photographs

2014-2015 Sensors & Actuators - H.Sarmento 12


Particle image velocimetry (3)

• With a double pulse two bright spots appear on the


photograph.
• The distance (displacement) d between the two bright spots is
measured.
• Speed is determined by
d
v
t
t - time interval between laser pulses.
• The direction of the particle movement is determined by image
processing.

2014-2015 Sensors & Actuators - H.Sarmento 13


Laser Doppler velocimetry (1)
• Velocity is measured at a fixed point in the flow, for tiny
particles that move with the fluid.
LDV
(Laser Doppler Velocimetry)

[Source: J. M. Cimbala]

• The laser beam is split into two parallel laser beams of equal
intensity.

2014-2015 Sensors & Actuators - H.Sarmento 14


Laser Doppler velocimetry (2)

• Beams pass through a converging lens that focuses the


beams at a point in the flow.
2014-2015 Sensors & Actuators - H.Sarmento 15
Laser Doppler velocimetry (2)

• At the convergence point, the waves interfere, creating a


bright and dark fringe pattern due to constructive and
destructive interference.

2014-2015 Sensors & Actuators - H.Sarmento 16


Laser Doppler velocimetry (3)

• Bright and dark fringe pattern :

[Source: J. M. Cimbala]

• Tiny particles that pass through the measurement volume


scatter laser light.
• The scattered light intensity is bright, then dark, then
bright, etc. as particle moves through the fringe pattern.

2014-2015 Sensors & Actuators - H.Sarmento 17


Laser Doppler velocimetry (4)

• The scattered laser light is collected by a receiving lens


and photodetector.
• Fluctuations in light intensity are converted to a
fluctuating voltage signal.
2014-2015 Sensors & Actuators - H.Sarmento 18
Laser Doppler velocimetry (5)

• The spacing between fringe lines


λ  - wavelength of the laser light
d
2 sin  2   - angle between the two beams

• The speed of the moving particle is linearly


proportional to the frequency of the fluctuating light
intensity:

fd 
2 sin  2 
2014-2015 Sensors & Actuators - H.Sarmento 19
Laser Doppler velocimetry (6)

• The Bragg cell permits to shift slightly the beam frequency,


causing the fringe pattern to move. This permits to detect
direction.

2014-2015 Sensors & Actuators - H.Sarmento 20


Thermal anemometer (1)

• The rate of convective heat transfer from a hot object to the


surrounding fluid increases as the speed of the fluid flowing
around the object increases.

• Thermal anemometers consist of a heated (by electric current)


temperature sensor.

• The sensor tends to cool down as fluid velocity increases, but


electronic control maintains it to a constant temperature,
increasing the current.

2014-2015 Sensors & Actuators - H.Sarmento 21


Thermal anemometer (2)

• E is the voltage across the sensor.

E2
P  I 2 Rsensor 
Rsensor
[Source: J. M. Cimbala]

• Speed can measured by E


E  a  bv n
a, b, and n are constants (calibrated for a given sensor).

2014-2015 Sensors & Actuators - H.Sarmento 22


Pitot tube (1)

• Velocity measurement based on pressure measurement.


• Pitot-static probe is a tube with stagnation pressure tap and
several circumferential static pressure taps.

[Source: J. M. Cimbala]

2014-2015 Sensors & Actuators - H.Sarmento 23


Pitot tube (2)

2P1  P2 
v

 - fluid density

[Source: J. M. Cimbala]

2014-2015 Sensors & Actuators - H.Sarmento 24


Angular velocity measurement

• Angular velocity sensors to measure shaft speed are many


times called tachometers (tacho means speed in greek).
• Gyroscopes (or gyros) can also be used to measure angular
velocity.
• Types of gyroscopes:
– Rotary: based on the principle of the conservation of
angular momentum.
– Vibratory: based on Coriolis acceleration.
– Optical gyroscope: based on Sagnac effect.

2014-2015 Sensors & Actuators - H.Sarmento 25


Rotary gyroscopes (1)
• Rotary gyroscopes are navigational tools.
• They are used in the stabilization devices, where a stable
directional reference is required, such as satellites, smart
weapons and robotics.
• The basic principle involved is the principle of conservation of
angular momentum.

2014-2015 Sensors & Actuators - H.Sarmento 26


Rotary gyroscopes (2)
spin axis

platform

[Source: J. Fraden, 2010]

input axis output axis

• A rotor (heavy disk/wheel).


• A platform with an inner and an outer gimbal that is free to
rotate about two axes.

2014-2015 Sensors & Actuators - H.Sarmento 27


Rotary gyroscopes (3)
spin axis

platform

[Source: J. Fraden, 2010]

input axis output axis

• When the rotor freely rotates, it tends to preserve its axial


position

2014-2015 Sensors & Actuators - H.Sarmento 28


Rotary gyroscopes (4)
spin axis

platform

[Source: J. Fraden, 2010] output torque


applied torque
input axis output axis

• If a torque is applied to the frame around one axis (input), the


platform develops a torque around a perpendicular axis
(output).

2014-2015 Sensors & Actuators - H.Sarmento 29


Rotary gyroscopes (5)
spin axis

platform

applied torque precession


[Source: J. Fraden, 2010]

input axis output axis

• The spin axis turns around the output axis. This phenomenon
is called precession of a gyro.

2014-2015 Sensors & Actuators - H.Sarmento 30


Rotary gyroscopes (6)
spin axis

platform

applied torque [Source: J. Fraden, 2010] precession


input axis output axis

• The precession becomes a measure of the applied torque and


can be used as an output to, for example, correct the direction
of a device.

2014-2015 Sensors & Actuators - H.Sarmento 31


Rotary gyroscopes (7)
spin
axis

applied torque [Source: J. Fraden, 2010]


precession
input axis output axis

• Application of torque in the opposite direction reverses the


direction of precession.

2014-2015 Sensors & Actuators - H.Sarmento 32


Rotary gyroscopes (7)

• The relation between the applied torque and the angular


velocity of precession is:

T  I 

T is the applied torque


 the angular velocity of the spin axis
I moment of inertia of the rotating mass
 is the angular velocity of precession.

2014-2015 Sensors & Actuators - H.Sarmento 33


Mass Moment of Inertia

• Moment of Inertia of a rotating mass (I) is a measure of an


object's resistance to changes in a rotation direction
• Moment of Inertia of a rotating mass depends:
– on the mass of the object
– its shape
– and its relative point of rotation.
• For a circular disk is
2
Mr
T
M - mass of the disk
r - distance between axis and outside disk
2

2014-2015 Sensors & Actuators - H.Sarmento 34


Rotary gyroscopes (8)

• When the wheel (rotor) freely rotates, it tends to preserve its


axial position.
• If the gyro platform rotates around the input axis, the gyro will
develop a torque around a perpendicular (output) axis, thus
turning its spin axis around the output axis. This phenomenon
is called precession of a gyro.

2014-2015 Sensors & Actuators - H.Sarmento 35


Vibratory gyroscopes

• In MEMS gyroscopes the rotating mass is replaced with a


vibrating element: the mass moves linearly in simple harmonic
motion.
• Vibrating gyroscopes rely on the phenomenon of the Coriolis
acceleration.
• The resulting Coriolis acceleration can be detected and related
to the angular velocity.

2014-2015 Sensors & Actuators - H.Sarmento 36


Coriolis acceleration (1)

rotation

v
linear motion
ax
Coriolis acceleration

• The Coriolis acceleration appears, whenever a body moves


linearly in a frame of reference that is rotating about an axis
perpendicular to that of the linear motion.

2014-2015 Sensors & Actuators - H.Sarmento 37


Coriolis acceleration

rotation

aCoriolis  2 v
v
aCoriolis
aCoriolis 
2v
Coriolis acceleration

• If the sensor is rotated in the plane perpendicular to the linear


vibration, an acceleration is obtained, proportional to the
angular velocity .

2014-2015 Sensors & Actuators - H.Sarmento 38


MEMS vibratory gyroscope (1)

[Source: Mems Mechanical Sensors, 2004]

• MEMS gyroscopes rely on a mechanical structure that is


driven into resonance and excites a secondary oscillation, due
to the Coriolis force.

2014-2015 Sensors & Actuators - H.Sarmento 39


MEMS vibratory gyroscope (1)

[Source: Mems Mechanical Sensors, 2004]

• Inertial mass (gyro element).


• Two-gimbal (inner and outer gimbals) platform supported by
torsional flexures.
2014-2015 Sensors & Actuators - H.Sarmento 40
MEMS vibratory gyroscope (2)

• The external gimbal is driven into oscillatory motion with a


constant amplitude.
• This oscillatory motion is transferred to the inner gimbal,
setting up an oscillating in the gyro element.
2014-2015 Sensors & Actuators - H.Sarmento 41
MEMS vibratory gyroscope (2)

• In the presence of an angular rotational rate normal to the


plane of the device, the Coriolis force will cause the inner
gimbal to oscillate about the secondary axis.
• The frequency is equal to the drive frequency and with an
amplitude proportional to the inertial input rate.
2014-2015 Sensors & Actuators - H.Sarmento 42
MEMS vibratory gyroscope (2)

• Frequency in the secondary axis is sensed by two pairs of


electrodes making up a (differential capacitor.)

2014-2015 Sensors & Actuators - H.Sarmento 43


Optical Gyroscopes
• Used extensively for guidance and control.
• No moving components.
• Based on the Sagnac effect: based on propagation of light in
optical fibers.
• There are different types of optical gyros.
• Example: coil fiber gyroscope.

2014-2015 Sensors & Actuators - H.Sarmento 44


Sagnac effect (1)

• Two beams of light generated by a laser propagate within an


optical ring in opposite directions.

• The two beams are received by the receiver at the same time.

2014-2015 Sensors & Actuators - H.Sarmento 45


Sagnac effect (2)

• If the optical ring is rotating, the beam travelling against the


rotation experiences a slightly earlier than the other beam.

2014-2015 Sensors & Actuators - H.Sarmento 46


Sagnac effect (3)

2R  l1
t1  c
v v
n
2R  l2
t2 
v

• t1 time the beam travel in the direction of rotation before being


detected.
• t2 time the beam travel in the direction opposite to rotation
before being detected.
• n is the refraction index.

2014-2015 Sensors & Actuators - H.Sarmento 47


Sagnac effect (4)
2R  l1
t1 
v
l1  R t1

t1v  Rt1  2R

2R 2R 2R


t1   t2 
v  R  R   R 
v1   v1  
 v   v 

2014-2015 Sensors & Actuators - H.Sarmento 48


Sagnac effect (5)

4R 2 1 4R 2
t1  t 2  v  R t1  t 2 
 R 
2
v2 v2
1  
 v 

v2 v2
vt1  t 2 
 t v
 t 
4R 2
4R 2
4R 2

v
 l
4R 2

• Angular velocity determined measuring l difference.

2014-2015 Sensors & Actuators - H.Sarmento 49


Accelerometers (1)

• An accelerometer is a sensor that measures the linear


acceleration experienced by an object.

• Accelerometers employ a moving mass in one form or another.

• Modern accelerometers are often small implemented as


MEMS.

2014-2015 Sensors & Actuators - H.Sarmento 50


Working principle (1)
• Conceptually, an accelerometer behaves as a seismic mass (or proof
mass) connected to a spring of stiffness (k) attached to the housing. .

[Source: Fraden, 2010]

• When the accelerometer is subjected to acceleration a, a relative


displacement x of the seismic mass is produced by inertia.

2014-2015 Sensors & Actuators - H.Sarmento 51


Working principle (2)

• Considering the:
– acceleration of the accelerometer body a

d 2x
– horizontal acceleration of the mass dt 2

• The equation of the movement is given by equation


d 2x dx
M 2  b  kx  Ma
dt dt
Damper force Spring force

2014-2015 Sensors & Actuators - H.Sarmento 52


Working principle (3)
• Laplace transform

X s  M 02 k b
 0  20 
As 
,
k s 2  20 s  02 M M

b b
  
2 M 0 2 kM

2014-2015 Sensors & Actuators - H.Sarmento 53


Working principle (4)

• The accelerometer output signal may have an oscillating


shape.

<1

=1

>1

2014-2015 Sensors & Actuators - H.Sarmento 54


Working principle (5)

• Flat frequency response where the most accurate measurement


can be made.
• Care shall be taken not to use the accelerometer close to its
natural frequency.

2014-2015 Sensors & Actuators - H.Sarmento 55


Working principle (8)
k
X s  M  2
0 
 0
M
As  k s 2  20 s   02
b b
  
2 M 0 2 kM

• A large mass increases the sensitivity (M/k), but reduces the natural
frequency and the damping ratio.
• Stiffness (k) increases the natural frequency but reduces the
sensitivity and the damping ratio.
• MEMS accelerometers are very stiff and have small mass: they have
a large natural frequency, but small sensitivity and damping ratio.

2014-2015 Sensors & Actuators - H.Sarmento 56


Types of accelerometers (1)

• Acceleration sensors can be classified according to the


physical principle they use:
– a direct measurement of a force:
(force strain) F
a
m
– an indirect measurement, by means of displacement or
deformation of a sensing element.
(force displacement) kx
a
m

2014-2015 Sensors & Actuators - H.Sarmento 57


Types of accelerometers (2)

F  ma

F kx
a a
m m
Piezoresistive Capacitive
Thermal
F x
Other:
Piezoelectric Potentiometric,
Optical,
Inductive, hall effect
2014-2015 Sensors & Actuators - H.Sarmento 58
Piezoelectric accelerometers (1)

• In a piezoelectric accelerometer the seismic mass is directly


connected to the piezoelectric crystal.
• When the accelerometer is subjected to an acceleration, the
mass moves on the crystal and the compression of the crystal
produces a proportional electric signal.

2014-2015 Sensors & Actuators - H.Sarmento 59


Piezoelectric accelerometers (2)

• The mass is enclosed between two piezoelectric crystals.

F
Q V const  coef piezo F a
m

[Source: Halit Eren]

• The measured electric signal (Q) is equal to the force (F)


applied on a crystal, due to the acceleration of the mass,
multiplied for the piezoelectric coefficient.

2014-2015 Sensors & Actuators - H.Sarmento 60


Piezoelectric accelerometers (3)

• Single ended compression accelerometer: the piezoelectric


crystal is sandwiched between the case and the proof mass.

F
a
m

[Source: J. Fraden, 2010]

2014-2015 Sensors & Actuators - H.Sarmento 61


Piezoresistive accelerometer (1)

• In piezoresistive accelerometers the seismic mass is supported


by elastic elements, which incorporates semiconductor strain-
gauge as sensing elements.
• The measured strain is related to the magnitude and rate of
mass displacement and, subsequently, with the acceleration.

F R 1 F
 E  E a
A R G m

• Most piezoresistive accelerometers use two or four active


gauges arranged in a Wheatstone bridge

2014-2015 Sensors & Actuators - H.Sarmento 62


Piezoresistive accelerometer (2)

• Cantilever design: proof-mass suspended by a “spring”, which


in MEMS is usually a cantilever or beam.

[Source: A. Albarbar] [Source: Chang Liu]

2014-2015 Sensors & Actuators - H.Sarmento 63


Piezoresistive accelerometer (3)
• Bulk-micromachined silicon accelerometer: motion perpendicular to
the wafer plane.

[Source: Chang Liu]

[Source: R.P. vanKampe, 1995]

• When the accelerometer is subjected to an acceleration, the mass


moves up and down, causing the piezoresistances to change.

2014-2015 Sensors & Actuators - H.Sarmento 64


Capacitive accelerometer (1)

[Source: Nathan Ida]

2014-2015 Sensors & Actuators - H.Sarmento 65


Capacitive accelerometer (2)
• An internal mass, supported by four silicon springs, is sandwiched
between the upper cap and the base:
– The upper plate and the mass, at distance d1, define a capacitance
C1.
– The base and the mass, at distance d2, define a capacitance C2.
upper plate

d1 C1

d2
C2
base [Source: J. Fraden, 2010]

2014-2015 Sensors & Actuators - H.Sarmento 66


Capacitive accelerometer (3)
upper plate

d1  d C1 

d 2  d C2 
base

kx
x  C C
te
a
m

[Source: J. Fraden, 2010]


2014-2015 Sensors & Actuators - H.Sarmento 67
Capacitive accelerometer (4)
• When the accelerometer is subjected to an acceleration, the mass
moves and the distance d1 and d2 and capacitances C1 and C2
change..
upper plate

d1  d C1 

d 2  d C2 
base

A A
C1    C1  C C2    C 0  C
dx dx
2014-2015 Sensors & Actuators - H.Sarmento 68
Capacitive accelerometer (5)
A A
C1  C2  2C   
dx dx
x
C  A 2
d x 2

• For small displacements: x 2  d 2

d2
x  C
A

2014-2015 Sensors & Actuators - H.Sarmento 69


Capacitive accelerometer (6)

• ADXL78 (MEMS):
– single-axis accelerometer with signal conditioned voltage
outputs that are on a single monolithic IC.

[Source: Analog devices]

2014-2015 Sensors & Actuators - H.Sarmento 70


Thermal accelerometers (1)
sealed
chamber Heater

detector 1 detector 2
Substrate [Source: Yu ZHANG]

• Gas is the seismic mass.


• A resistive heating element at the centre heats gas molecules.
• When subjected to acceleration, the less dense (heat) gas molecules
move in the direction of acceleration and denser molecules (cool)
move in the opposite direction, creating a temperature difference.
• Acceleration lead to a forced convection of the gas within sealed
chamber, creating a temperature difference.
2014-2015 Sensors & Actuators - H.Sarmento 71
Thermal accelerometers (2)

sealed
chamber Heater

detector 1 detector 2
Substrate

• Temperature sensors (detector 1 and detector 2) measure the


temperature difference.
• The difference of temperature is proportional to acceleration.

2014-2015 Sensors & Actuators - H.Sarmento 72


Thermal accelerometers (3)

• Under zero acceleration, a temperature distribution across the


gas cavity is symmetrical about the heat source.

acceleration

[Source: MEMSIC]

2014-2015 Sensors & Actuators - H.Sarmento 73


Thermal accelerometers (4)

• MX2125 dual axis accelerometer: a chamber of gas with a


heating element in the center and four temperature sensors
around its edge.

[Source: MEMSIC]

2014-2015 Sensors & Actuators - H.Sarmento 74


Other types of accelerometers (1)

• Other accelerometers: inductive (LVDT), Hall effect, optical,


etc.

2014-2015 Sensors & Actuators - H.Sarmento 75


Other types of accelerometers (2)

• A rod connected and moving with the mass links to a coil. The
inductance of the coil is proportional to the position of the
mass
• An LVDT may be used

[Source: Nathan Ida]

2014-2015 Sensors & Actuators - H.Sarmento 76


Other types of accelerometers (3)

• Acceleration changes distance to hall element


• Hall element output is calibrated as acceleration
• The magnet may be on the hall element side (biased hall
element)

[Source: Nathan Ida]

2014-2015 Sensors & Actuators - H.Sarmento 77


Other types of accelerometers (4)

• Optical accelerometers use fiber optics.

2014-2015 Sensors & Actuators - H.Sarmento 78


Bibliography (1)
• Jacob Fraden, “Handbook of modern sensors: physics, designs, and applications”, Springer,
third edition, 2010
• John M. Cimbala, Linear Velocity Measurement, Penn State University. Available at:
https://www.mne.psu.edu/me345/Lectures/Velocity_measurement.pdf
• Charles P. Pinney, William E. Baker, Velocity Measurement, Measurement, Instrumentation,
and Sensors Handbook, CRC Press, 1999
• Chang Liu, Foundations of MEMS, Prentice Hall, 2012
• Stephen Beeby, Graham Ensell, Michael Kraft and Neil White, Mems Mechanical Sensors,
Artech House, 2004.
• Linear Velocity Transducer Technology, Trans-tek Inc. Available at:
http://www.transtekinc.com/assets/files/Catalog_PDFs_04C/LVTs/LVT_tech04c.pdf
• John M. Cimbala, Linear Velocity Measurement,Penn State University. Available at:
http://www.mne.psu.edu/me345/Lectures/Velocity_measurement.pdf
• Fabien Napolitano, Fiber-Optic Gyroscope key tchnological advanatges, IXSEA, 2010.
Available at: http://www.ixsea.com/pdf/fog-key-advantages.pdf
• Greg Chase, Sagnac Interferometer, Phys 517 Qunatum Mechanics 2. Available at:
http://einstein.drexel.edu/~bob/Term_Reports/Chase_2012.pdf

2014-2015 Sensors & Actuators - H.Sarmento 79


Bibliography (2)
• Steps to selecting the right accelerometer, Meggitt’s Endevco Corporation.
Available at: https://www.endevco.com/news/newsletters/2012_07/tp327.pdf
• National Instruments, Accelerometer Principles. Available at:
http://zone.ni.com/devzone/cda/ph/p/id/12
• Craig, Kevin C., Mechatronics in Design: That fictitious force, EDN
Europe, Issue 11, p17, November 2012. Available at:
http://www.edn.com/electronics-blogs/mechatronics-in-
design/4397081/That-fictitious-force

2014-2015 Sensors & Actuators - H.Sarmento 80

Anda mungkin juga menyukai