Uncertainty Analysis

Errors in experimentation fall into two categories: bias and random. Bias errors are related to a consistent difference between the signal generated by a probe and that sensed by the measurement system. Most modern measurement systems permit self-calibration using a known in-built signal generator and bias errors can be easily minimized. Random errors arise from a host of uncontrollable factors that simultaneously affect the experiment. These include room temperature changes, supply voltage fluctuations, air currents, building vibration and roughness of nominally smooth surfaces. Random errors are a part of any experiment and these render the experimental data as fundamentally irreproducible. However, in any good experiment, measured data will have a stable mean value about which readings obtained in definite runs of the experiment are distributed. Such a distribution is called scatter; considerable effort must be expended to control and reduce scatter. This requires simultaneous improvement in the quality of the measurement systems and the test cell in which the experiment is performed. Experiments dedicated to determining scatter must be carried out to estimate its magnitude. Such experiments involve performing runs on: different days at identical room temperatures; on different apparatus that look alike; with measurement systems having identical specifications and with mild mis-orientation of the probe with respect to the flow. Data obtained from such experiments would result in a distribution of values about the mean.