Agilent A Guard-Band Strategy for Managing False-Accept Risk - White Paper 5991-1267EN c20140530 [10] free download

File name A Guard-Band Strategy for Managing False-Accept Risk - White Paper 5991-1267EN c20140530 [10].pdf

Keysight Technologies A Guard-Band Strategy for Managing False-Accept Risk White Paper Abstract When performing a calibration, the risk of incorrectly declaring a device as in-tolerance (false- accept risk) is dependent upon several factors. Those factors include the specified tolerance limit, guard-band, the calibration process uncertainty and the a priori probability that the device is intolerance. A good estimate of the a priori probability may be difficult to obtain. Historical or device population information for estimating the a priori probability may not be readily available and may not represent the specific device under test. A common strategy for managing measurement decision risk is to choose a guard-band that results in the desired false-accept risk given the tolerance limit, the calibration process uncer- tainty and the a priori probability. This paper presents a guard-band strategy for managing false- accept risk with only limited knowledge of the a priori probability that a device is intolerance. Introduction When determining if measurement quantities are within specified tolerances, ANSI/NCSLI Z540.3-2006 specifies that the maximum level of false-accept risk be no more than 2%. False-accept risk is the probability that measuring an out-of- tolerance device will indicate an intolerance condition due to measurement error. False-reject risk is the probability that measuring an in-tolerance device will indicate an out-of-tolerance condition due to measurement error. False-accept and false-reject occurrences have financial consequences, and therefore, minimizing both is often a worthwhile objective. One strategy for managing false-accept risk is to apply a guard-band such that the acceptance limits are more stringent that the tolerance limits. A common practice (see [3]) is to set the guard-band to a value equal to the 95% expanded uncertainty of the calibration process. This level of guard-band guarantees the Z540.3 false-accept risk requirement and is attractive in that it only requires information that many calibration organizations routinely manage (that is, the tolerance limits and the 95% expanded uncertainty, which is set as the guard-band). However, when using a guard-band to reduce false-accept risk, a corresponding increase results in the false-reject risk. With the guard-band set to the 95% expanded uncertainty, the false-reject risk can be disproportionately high (see Figure 3). An alternative to applying a guard-band equal to the 95% expanded uncertainty is to determine the false-accept risk and set an appropriate guard-band, if necessary, that adjusts the false-accept risk to the desired level. To determine the level of false-accept (or false-reject risk) for a calibration measurement, the following information is necessary: