Downloaded from https://www.everyspec.com AMCP "S-198 3-. ELIMINATION OF BIAS dissipation and temperature on resistors (e.1,A group of resistors is to be placed Bias in a reliability test may result. in in an oven which maintains a constant tern- Wole conclusions. Bias may arise through porature. The resistors are to be split into the choice of sample, experimenter influences, three groups, each with a different power instrument errors, physical or laboratory dissipation. If there are three rAcks In the variations, or the experimental design Itself. oven, the simplest procedure Is to place each Dias can be reduced through experimental power dissipation group on a rack.
controls and randomization (Rot. 1). 3.4..1 xperfftntolConmisDuring the life test, hot spots may develop 3..2 E~a~mntl onrosin the oven, espechilly since different levels ,I of power dissipation generate different In part qualification tests, parts used for amounts of heat, Also, because hot air rises, control are selected ftom the same population the top rack cvould be womcwhst hotter than as the test sample and are subjected to the the bottom rack. Placement of all resistors same conditions with the exception of the of one group on the some rack might lead to conditions that are boing studied (Ref. 1). biased results.
To avoid Lhii possibility, the Controls permit the test engineer to determine oven location of each resistor can bo detst, if the test results are a function of ilie con. mined in a random manner, huch as through ditinns applied or of other variables that aire the usp of a random-nutmber table, Then theI biasing the results. Trho test engineer may be unknown or uncontrollable effects of toni- interested, for example, In the variation of' peratuit' variation within the oven would be resistance with timne when resistors are sub.
distributed within till power dissipation Jected to a specific set of ororating and on- gop ba ~~ivre noa netit viromenal cndiions Ifcontol esisorswhich good statiolical analysis can estimate * arc not used, the engineer cannot bit sure If and correct for). observed fluctuations In resistarnco are time fluotuations, or are cau'(ed by the load, or In reliability demonstration testts, the order aic due to a combination of time antd law~l. and combination of aipplied stresses and A, group of simiflar resistors that did not have environgnents can be determined randomnly any load imposed on them cotild be used tis if they ate nut otherwise specified, This en-- a control.
suyes that all such combinations aril equally Tue se o cotrcl is oredificultt' ikey on the test and that nu tiases are implement on roliability dnmonkstration tests itrdc. because only a~ limited num bet of equipments 3-4.3 EXPERIMCINTAL UNCERTAINTY or systems are tested. If only a few systems ure available, it makes more stenau to use Experimental uncertainty Is the random thein for demonstration testing rather than effect of factors over which the experim~enter to waste a system as a control. does not have complete control (Re(, I).
Two important sources of experimental un- 3-4.' Rand-omixotioii certainty are (I) the inherent variability In, manufactured parts, and (2) the raiiiom The test engineer can use randomization variatio~n In the physical condoct of expari- procedures to protect against the introduction monts. of bias. An example of a situation in which randomization is required is ii life test experi- Experimental uncertainty is associated J ment designed to test the effects or power with the concept of precision, thea repeat. ......... .4. . .. .. .. ...