Evaluation (Validation and Verification) - Medical Device Design: Innovation from Concept to Market - page 220

o/p
i/p
Figure 9.6
Sensitivity.
The main aim of all calibration activities is to obtain a graph of input versus output. This
graph reveals a plethora of information.
9.3.3.1 Sensitivity
The usual protocol to follow is to vary the input to a device and measure the output. These
values are then plotted on a calibration graph with input as the abscissa (horizontal) and the
output on the ordinate (vertical).
The sensitivity is the gradient of the best-fit line through the points. This graph is easily
obtained using a spreadsheet (such as Excel); but be careful to use the x-y scatter graph.
Obtaining a best-fit line is also easy and the gradient is easily obtained. Using a spreadsheet
or a data analysis package is by far the best method as the statistics are all done for you.
You are able to produce a nonlinear system, but in this case sensitivity will also be nonlinear.
Sensitivity will then be an equation in the form of a polynomial, moving average, or other suitable
function. These are, of course, far more difficult to deal with and, in general, are to be avoided.
9.3.3.2 Range
Most real devices produce a saturation curve. This has three distinct regions. At low levels of
input the physical errors in the system (internal friction, etc.) make measurements unreliable and
as a consequence the output is nonlinear (output is not proportional to input). At larger inputs the
measurements, again, become unreliable as the device's limits of operation have been exceeded.
Once again, the output is not proportional to input. In between there is, normally, a region where
the device behaves itself and the output is proportional to input: the device is behaving linearly.
The region of the input where this is true is called the range of the device ( Figure 9.7 ).
Once again, spreadsheets and data analysis packages come to the rescue; range is determined
with relative ease.
9.3.3.3 Repeatability
Repeatability is a measure of, well, whether the outputs are repeatable. To put this another
way, for the same input do you always get the same output? Conducting the input-output
experiment repeatedly and plotting all the points on a single graph will allow you to obtain
this measure ( Figure 9.8 ).