SIX SIGMA QUALITY METRICS vs TAGUCHI LOSS FUNCTION Luis Arimany de Pablos, Ph.D.
outside the mean 2 a maximum 25% of the values outside the mean 3 a maximum 11.11% of the values outside the mean 4 a maximum 6.25% of the values outside the mean 5 a maximum 4% of the values outside the mean 6 a maximum 2.77% of the values FOR ANY DISTRIBUTION
outside the mean 2 there are 4.55% of the values outside the mean 3 there are 0.27% of the values outside the mean 4 there are 0.006% of the values outside the mean 5 there are 5.74·10 -5 % of the values outside the mean 6 there are 19.8·10 -8 % of the values FOR NORMAL DISTRIBUTION ( two tails )
One of Motorola´s most significant contributions was to change the discussion of quality, from quality levels measured in % (parts-per- hundred), to one, in parts per million, or, even, parts per billion
to the right of the mean + 2 there are 22,750 per million to the right of the mean +3 there are 1, per million to the right the mean + 4 there are per million to the right of the mean + 5 there are per million to the right of the mean + 6 there are per million FOR NORMAL DISTRIBUTION ( one tail )
DEFECTIVE PRODUCT OR SERVICE X USLX LSL If we set the Specification Limits at m 3 On average 0.27 % defectives 2.7 per thousand 2,700 per million 1,350 per million (one tail)
We should have a process with such a low dispersion that Specification Limits are at: m 6 defective per million per million in one tail per million
Process Capability Index, Cp (Potential Capability) Cp = ( USL-LSL)/6 USL-LSL = Specification interval 6 = Process Capability
Process Centred at Target Process CpLSLUSL Right hand ppm defective 11 22 33 44 55 66 158,655 22,750 1, m- 1 m+ 1 m-2 2 m+2 2 m-3 m+3 m-4 4 m+4 4 m-5 5 m+5 5 m-6 6 m+6 6
We should have a process with such a low dispersion that Specification Limits are at: m 6 defective per million per million in one tail per million
Working with 6 methodology you get 3.4 defectives per million How can this be, if the exact figure is ppm (or ppm if we consider only one tail)?
Even if a process is under control it is not infrequent to see that the process mean moves up (or down) to target mean plus (minus) 1.5 . If this is the case, the worst case, working with the 6 Philosophy will guarantee that we will not get more than 3.4 defectives per million products or services
Let us assume that the process mean is not at the mid-point of the specification interval, the target value m, but at m+1.5
Process Capability Index, Cpk Cpk = ( USL-mp)/3 USL = Upper Specification Limit mp = process mean 3 =Half Process Capability
Process Centred at m ProcessCpk USL Right hand ppm defective 11 22 33 44 55 66 691, ,536 66,807 6, m+ m+2 m+3 1.5 m+4 m+5 m+6 Z score
Process Centred at m Process Right hand ppm defective 11 22 33 44 55 66 691, ,536 66,807 6, Process Centred at m Cpk Right hand ppm defective Cp ,655 22,750 1,
QUALITY The Loss that a product or service produces to Society, in its production, transportation, consumption or use and disposal (Dr. Genichi Taguchi)
L=k(x i -m) 2 E(L)=k 2
Loss Function (Process Centred at Target) Six Sigma Metric Cp R H ppm defective 11 22 33 44 55 66 158,655 22,750 1, Loss Function 33 1.5 0.75 0.6 0.5 Standard Deviation 9k k 2 1k k k k 2
Loss Function (Process Centred at m+1.5 ) Six Sigma Metric Cpk R H ppm defective 11 22 33 44 55 66 691, ,536 66,807 6, Loss Function 33 1.5 0.75 0.6 0.5 Standard Deviation 29.25k k k k k k 2
Six Sigma Metric Cpk 11 22 33 44 55 66 Loss Function (Process Centred at m+1.5 ) 29.25k k k k k k 2 Loss Function (Process Centred at m) 9k k 2 1k k k k 2 Cp
Six Sigma Metric Cpk 11 22 33 44 55 66 R H ppm defective (Process Centred at m+1.5 ) R H ppm defective (Process Centred at m) Cp ,655 22,750 1, , ,536 66,807 6,
AVERAGE RUN LENGTH 3 Sigma process Probability to detect the change 0.5 Average Run Length 2
AVERAGE RUN LENGTH 4 Sigma process Probability to detect the change Average Run Length 6.42
AVERAGE RUN LENGTH 5 Sigma process Probability to detect the change Average Run Length 43.45
AVERAGE RUN LENGTH 6 Sigma process Probability to detect the change Average Run Length
Six Sigma Metric Standard Deviation 33 44 55 66 3 3 Probability of Defectives after the Shift Expected Number of samples to detect the Shift Average Run Length USL
Six Sigma Metric Standard Deviation 33 44 55 66 3 3 Probability of Defectives after the Shift Expected Number of samples to detect the Shift Average Run Length