Sensing and Sensors: Fundamental Concepts II

MediaRobotics Lab, January 2010

Sensitivity

Range

Precision

Accuracy

Resolution

Offset

Hysteresis

Response Time

Source: sensorwebs.jpl.nasa.gov
 A sensor is a device that receives a stimulus and responds with an electrical signal

A transducer is a converter from one type of energy into another one.

Loudspeaker: electric signal -> variable magnetic field -> acoustic wave / sound

Direct sensor: converts a stimulus into an electrical signal by using a physical

effect (photo effect, for example)

Complex sensor: needs in addition one or more transducers before the sensor
can generate an electrical signal.
Sensors do not 'function' alone; they are always part of a larger system -
Basic concepts

Sensitivity
the minimum input (of a physical parameter) that will create a detectable output

Range
the minimum and maximum values of a given parameter the sensor can measure.

Precision
the ability of the sensor to reproduce the same results in repeated tests of identical conditions

Accuracy
the maximum difference between the actual value and the value indicated by the sensor

Resolution
the smallest detectable incremental change of input that can be detected in the output signal

Offset
the output existing in the absence of input

Hysteresis
the effect of direction of the input on the output

Response Time
Time required for a sensor to change from a previous state to a new state (within a tolerance
band)
The world is not perfect -

sensors convey the stimulus with distortions.

The ideal output can be described, based on empirical data observed in the
sensor's behavior, mathematically by a transfer function

A transfer function describes the dependence between the electrical signal and
the stimulus.

examples:

a linear function S=abx

a logarithmic function S=ablns
ks
an exponential function S=ae
a power function k
S=abs
Least square fitting of data:

->if the differences between an assumed function and the real data is minimum,
that function describes the data (sufficiently well)

where R are the vertical offsets

R2 is minimum when

http://mathworld.wolfram.com/LeastSquaresFitting.html
Regression analysis: Checking the goodness of the fit

One standard measure of the goodness of fit of a regression analysis is the coefficient
of determination, R2, and it can be calculated as follows:

2
R2 =1/ N x i
x y i
y / x y

...w here N is the number of observations, x'bar' is the mean of x (likewise for y), and is
the standard deviation

x = x i
2
x /N 2
y = yi y / N
 for a linear fit (line)

R2 is then

and the second

derivatives (with respect
to each variable) are

that can be solved for a and b, which define the slope and y axis intersection of the line
(that is all you need) ...
In practice you can use a computer program to calculate the
coefficients of a polynomial easily: python and matlibplot
python and matlibplot: polyfit and polyval

from pylab import *

# assuming this data set

t = (0,5,10,15,20,25,30,35,40,50)
v = (1,7,14,15,16,22,25,27,28,30)

#show the data

plot(t, v, linewidth=2.0)

#assume an order
n=?

#find the coefficient of the polynomial

coeffs = polyfit(t,v,n)

#get the polynomial output

best = polyval(coeffs, t)

#plot the result

plot(t,v,'-go', linewidth = 1)

best = [ 1.45e-007,-1.93e-005, 9.17e-004,-1.80e-002,1.07e-001, 1.18e+000, 8.57e-001,]

Full scale input and full scale output
Accuracy

better: Inaccuracy
measured as the highest deviation of a value represented by the sensor from the ideal
of true values at its input, (with a specific level of uncertainty)

example:
a linear displacement-sensor should generate 1 mV per 1 mm displacement. However,
measurement shows that a displacement of s = 10.5 mm produced and output of S =
10.0V

1/b = 1mm/mV
sx = S/b = 10.5mm, sx s = 0.5 (0.5 mm in error)
inaccuracy = (0.5mm/10mm)* 100% = 5%

->often the deviation from the ideal transfer function is given as (+-) in terms of the
measured value or as percentage of the full span input or in terms of the output

-> often one performs accuracy ratings to find the real performance of a sensor. They
include combined effects of variations, hysteresis, dead band, calibration and
repeatability errors -> worst case scenario
Repeatability

Ability to recreate the same result under the same condition at different times. It is
often expressed as the maximum difference between output readings from two
calibrating cycles:
r = ( / FS)*100% (FS: full scale)

Dead Band

An insensitivity of a sensor in a specific range of input signals the sensor's blind spot.
 Output Impedance

A: interface for sensor with voltage output B: interface for sensor with current output

To minimize output distortions sensors need to be matched with their connecting circuitry.

For voltage generating sensors, a lower impedance (Zout) is preferable and the circuit should
have a high input impedance (Zin)

A current generating sensor should have an output impedance as high as possible and the
circuit's impedance should be as low as possible.
Example of first order sensor

Temperature sensor for which the energy storage is thermal capacity (stays hot for n sec)
Often such sensors are described by a frequency response: how fast they can respond to
a changed input

Example of a second order sensor

Accelerometer that incorporates a mass and a spring. A second oder response is typical for
a sensor that responds with a periodic signal. This is the systems resonant frequency.
Typically, the operating range of a sensor is selected below or above the resonant frequency.
For some special sensors , however, the resonant frequency IS the operating point (that is
where they will have the strongest response).

Damping is the suppression of oscillation in a sensor of order > 1.

When the sensor's response is maximally fast without overshoot, the response is called
critically damped.

If an overshoot occurs, the sensor is said to react in underdamped response.

If a sensor reacts slower than its max response, it is said to react in overdamped response.
Reliability

Reliability is the ability of a sensor to perform a required function under stated

conditions for a stated period of time. It is often expressed in statistical terms as a
probability that the device will function without failure for a number of uses.

Because of this, reliability of a class or batch of devices is evaluated by observing

the behavior of a large number of devices. Devices that have been fabricated with
the same technology, same materials, and same processes are then assumed to
behave in the same way (assumed reliability).
A sensor can have more that one dimension if the sensor's output is influenced by more
than one input stimulus

example: infrared sensor V = G*( Tb4 Ts4) 4th order parabola

where Ts = temp of sensor surface, Tb = temp of

an object of measurement and G a constant
Hysterisis

Hysteresis refers to systems which have memory;

that is, the effects of the current input to the system
are not felt at the same instant. Such a system may
exhibit path dependence.

In a deterministic system with no hysteresis, it is

possible to predict the system's output at an instant
in time, given only its input at that instant in time. In
a system with hysteresis, this is not possible; there
is no way to predict the output without knowing the
system's current state.

Hysteresis phenomena occur in magnetic and

ferro-magnetic materials, in which a lag occurs
between the application and the removal of a force
or field and its subsequent effect.
Saturation

Everything has a limit. A sensor is no exception. Where the response no longer

matches the transfer function, the sensor operates in saturated mode.
Resolution

Smallest increments of stimuli which can be sensed. Resolution is often

expressed as a percent of full scale (FS).

Example: an angular sensor has a FS of 270 deg, and its 0.5 deg resolution
would be specified as 0.181% FS.

When there are no measurable steps in the output of a sensor it is said to be

continuous.

Often the resolution is not constant over the whole input range.

In digital sensors, the resolution is specified as the number of bits used to

represent the output, as in '8-bit resolution', where the result is represented by 8
bit long data for FS.
Excitation

This is the electric signal often needed to make a sensor active for operation.
Excitation is often listed as a range of voltage and or current. In some sensors, the
frequency of excitation is also important (and must be specified).

Dynamic Characteristics

Sensors react with a certain delay to inputs. Sensors are, in this regard, said to be time
dependent and show a dynamic characteristic.

Some sensors need a warm-up period before they can operate reliably.

Including time dependency leads to a different kind mathematical description of sensor

behavior, one based on differential equations.

Zero order sensor Out(t) = a + b * i(t) a: static sensitivity

First oder sensor b1*(dOut(t)/dt) + b0*Out(t) = i(t)

Second order sensor b2*(d2Out(t)/dt2) + b1*(dOut(t)/dt) + b0*Out(t) = i(t)

FIT Rates (failures in time)

FIT Rate is a statistic for a component that measures how many failures the component
will have per one billion operating hours. The lower the FIT Rate for a component is, the
better the component is. FIT Rate is used to find Calculated MTBF.

MTBF (mean time between failures)

If a product contains n unique components, qi is the quantity of

the ith component, and ri is its FIT Rate, then Calculated MTBF of
the product is :

Effectively, this is the inverse of the sum of the FITs

Reliability is then often formulated as:

R = 1 - F(t)
 'location parameter'
'shape parameter'
'scale parameter'

1/ = failure rate
when = 1
on a logarithmic scale, the
Weibull distributions are linear
with fixed

assumption: failures of one type

tend to cluster along the same
area

plots from http://www.weibull.com