REAL TIME ESTIMATION OF MEASUREMENT UNCERTAINTY*

The quality of a measurement can be determined by evaluating its uncertainty statement against the user's requirements. If the measured value and its uncertainty are within the tolerances, it may be assumed fit for the user's purpose. The end use of a measurement should be the major factor in selecting measuring equipment, standards methods, environment, personnel, etc., required for the measurements and the techniques used to control and estimate the associated measurement uncertainty. The current era of re-engineering businesses and emphasizing continuous improvement will continue into the year 2000. Companies must have reliable measurements to manage processes and test them against specific requirements. It cannot be over stressed that decision making requires accurate uncertainty estimates. Customers and measurement producers must increase their knowledge of measurement accuracy and the process for determining and controlling estimates of the associated measurement uncertainty. Three different techniques are commonly used to estimate uncertainty and quantify the effects of variation on measurement accuracy: • Use of manufacturer's specifications or calibration report values or some multiple of them. • Complete experimental characterization of all known influence factors and appropriate statistical methods to evaluate and combine experimental random and systematic variations with other sources of variation; and • Programs of continuous measurement performance monitoring to maintain statistical control and provide real time estimates of variation. The paper will discuss the various methods being used by measurement organizations to ensure and/or control the quality of their measurements. The paper will deal primarily with the third method that provides \"real time\" estimates of measurement uncertainty. When replicate artifacts of the process routinely measured as check standards in a measurement control program, derived uncertainty estimates capture the true variation within the influence factors affecting the results of calibration or measurements. A noteworthy feature of the software tested by the author's organization is the inclusion of tolerance limits in its control charts that illustrate how current measurements compare with tolerance limits.