We
assrune the following two. ( 1. ) The lifetime under constant voltage
s.tress T is subjected to the Weibuli distribution. ( 2 ) There exists
the relation of the inverse powver Iaw t =1t~L"-" betlveen
voltage stress L' and lifetime l. In addition, we set the censored
time 7' of cutting off the test. Then. ~ve can minimize the variance
of estimator 7~ under the ser\'ice \'oltage T･~. 'I'hat is-, we can
optimize the V-! test, \vhere 7:* is the time ~vhen x(o./o) test pieces
are broken. The optimized test has the followillg wa},'･ \Ve s_et
the 2 voltage stress level T~ and V!. \vhere l･', is. set as high
as- possible. T･1 and Nt (the numbcr of test pieces at level Vf) are
set to lllillimize the variElnce (,f 7~. ~1'he calculation of minilnizati()n
is s o c(]nlpiex that we can not kno\v T"t and Nl easily \\ ith(,ut
the aid of numerical analysis_ s_uch as Romberg integration and N
e\vton･ Raphson method. 'I'hus, \ve propose the quasi-optilnum tes;.t
method. This_ n]ethod is \'ery eas. y al]cl the efriciency is. not
s_o Vol. 106. No 9,,10, Sep.,,'Oct , 1986
reduced. The way is the following, \,Ve set the voltage V, corresponding
to time 7' on the provisional V-t line and N,/N=0.75, O.70 (x=50,
5 respectively), where, N is the number of total test pieces. ~Ve
can expect to save a geat deal of test pieces by using optimum or
quasi･optimum test method compar-ed with the conventional one, where
the voltage levels are 3 -4 and NilN (the rate of the test pieces
at stress level i, i=1...m) is 1/m. For instance, when x=50, we sometimes
can cut it off about a half. |
|
|
|
|
|
|
|