Measurement error of a system of ideal elements • Consider the system shown below, consisting of 𝑛 elements in series.
• Suppose each element is ideal, i.e. perfectly
linear and not subject to environmental inputs. • If we also assume the intercept or bias is zero, i.e. 𝑎 = 0, then: 𝑂𝑖 = 𝐾𝑖 𝐼𝑖 for 𝑖 = 1, … , 𝑛, where 𝐾𝑖 is the linear sensitivity or slope. • It follows that 𝑂2 = 𝐾2 𝐼2 = 𝐾2 𝐾1 𝐼, 𝑂3 = 𝐾3 𝐼3 = 𝐾3 𝐾2 𝐾1 𝐼, and for the whole systems 𝑂 = 𝑂𝑛 = 𝐾1 𝐾2 𝐾3 … 𝐾𝑖 … 𝐾𝑛 𝐼 • If the measurement system is complete, then 𝐸 = 𝑂 − 𝐼, giving: 𝐸 = 𝐾1 𝐾2 𝐾3 … 𝐾𝑛 − 1 𝐼 • Thus, if 𝐾1 𝐾2 𝐾3 … 𝐾𝑛 = 1, then 𝐸 = 0 and the system is perfectly accurate. • The temperature measurement system shown appears to satisfy this condition.
• The indicator is a moving coil voltmeter with a
scale marked in degrees Celsius so that an input change of 1 V causes a change in deflection of 25 °C. • This system has 𝐾1 𝐾2 𝐾3 = 40 × 10−6 × 103 × 25 = 1 and thus appears to be perfectly accurate. • The system is not accurate, however, because none of the three elements present is ideal. – The thermocouple is non-linear, so that as the input temperature changes the sensitivity is no longer 40 μV °C−1. – Also changes in reference junction temperature cause the thermocouple e.m.f. to change. – The output voltage of the amplifier is also affected by changes in ambient temperature. – The sensitivity 𝐾3 of the indicator depends on the stiffness of the restoring spring in the moving coil assembly and this is affected by changes in environmental temperature and wear, causing 𝐾3 to deviate from 25 °C V−1. • Thus the condition 𝐾1 𝐾2 𝐾3 = 1 cannot be always satisfied and the system is in error. • In general the error of any measurement system depends on the non-ideal characteristics – e.g. non-linearity, environmental and statistical effects – of every element in the system. • Thus, in order to quantify this error as precisely as possible we need to use the general model for a single element The error probability density function of a system of non-ideal elements • We have seen that the probability density function of the output 𝑝(𝑂) of a single element can be represented by a normal distribution. • The mean value 𝑂ത of the distribution takes non- linearity and environmental effects into account. • The standard deviation 𝜎0 takes statistical variations in inputs 𝐼, 𝐼𝑀 and 𝐼𝐼 with time, and statistical variations in parameters 𝐾, 𝑎, etc., amongst a batch of similar elements into consideration. • These equations apply to each element in a measurement system of 𝑛 elements and can be used to calculate the system error probability density function 𝑝(𝐸). • Equations [3.6] show how to calculate the mean value of the output of each element in turn, starting with 𝑂1 for the first and finishing with 𝑂𝑛 = 𝑂ത for the 𝑛th. • The mean value 𝐸ത of the system error is simply the difference between the mean value of system output and mean value of system input (eqn [3.7]). • Since the probability densities of the outputs of the individual elements are normal, then the probability density function of the system output 𝑂 and system error 𝐸 is also normal (eqn [3.10]). • Equations [3.8] show how to calculate the standard deviation of the output of each element in turn, starting with 𝜎𝑂 1 for the first, and finishing with 𝜎𝑂𝑛 for the 𝑛th. • We note that the standard deviation of the system input is zero and that the standard deviation of the error is equal to that of the system output (eqn [3.9]).