‘Taguchi Orhogonal Aras, Page 1 Taguchi Orthogonal Arrays or oho M. Cin, Pe Sat Uninet stron 1 opener 0 Introduction There are options for creating Taguchi arrays for the design of experiments, depending on how many times you choose to test each level of each parameter. For example, consider an experiment with 3 parameters and 3 levels of each parameter (P= 3 and discussed in a previous learning module. We showed two Taguchi arrays for this case: © a 6-run array for testing each level of each parameter twice © a 9-run array for testing each level of each parameter three times The 9-run array is more desirable (if cost and time permit) because for each level of any one parameter, all three levels of the other parameters are tested. Of course, cither array here costs less to run than a full- factorial analysis, since the number of required runs for a full factorial analysis is N= L" ~ 3° = 27. We call a Taguchi array an orthogonal array (some authors call it a full orthogonal array) when for each level of a particular parameter, all L levels of each of the (P-1) other parameters are tested at least once. Sometimes, as P incteases, it is necessary to test all levels of all parameters more than once in order to meet the rules for Taguchi arrays, as discussed previously. In such cases, there should be no unnecessary repeats. For example, consider the P = 4, = 3 case on the next page. When parameter a is at level 1, parameter b is tested at levels 1, 2, and 3 (all levels). Similarly, parameters c and d are tested at levels 1, 2, and 3 (all L levels). The same thing holds when parameter a is at level 2 o level 3. The same thing holds for all of the parameters, Hence, we see that the definition of an orthogonal array holds for this case — for each level of a particular parameter, all levels of each of the (P-1) other parameters are tested at least once — only once in this particular case. The required number of runs is therefore 3 (J.~ 3 levels) x 3 (each level of each parameter tested 3 times) = 9 required runs for this orthogonal array Consider another example, the P = 5, L = 3 case on the next page. When parameter a is at level 1, parameter bis tested at level 1 twice, level 2 twice, and level 3 twice (twice for all 3 levels). Similarly, parameters c and d ate tested twice at levels 1, 2, and 3 (all Z levels). The same thing holds when parameter a is at level 2 or level 3. The same thing holds for ail of the parameters. Hence, our definition of an orthogonal array also holds for this case — for each level of a particular parameter, al [levels of each of the (P-1) other parameters are tested at least once — actually twice in this particular case. The required number of runs is therefore 3 (L 3 levels) x 6 (cach level of each parameter tested 6 times) = 18 required runs for this orthogonal array. Orthogonal arrays are the “best” and most common type of Taguchi array, and you are encouraged to use orthogonal arrays whenever time and cost permit, A table of Taguchi orthogonal arrays is provided below for values of P (number of parameters) ranging from 2 to 5, and L (number of levels) ranging from 2 to 5, 3),as z 3 7 3 Taguchi, P 7 Run# ab ¢ da abede Taguchi, [tape oaea||{ 1 | af) aft 1 [aa apajat Run# ab X||Run# a) 6 2 [alalafal 2 [alalafafa 2 | 1 {a 1 JEJE! 3 | aj 2} 2) a a 2 2 | 1) 2 eS M2 4 | aj 2} 2] 2 4 | 4] 2] 2] 2) 2 # | al 3 fala) s || a] aia s | 2) al 2] a 2 4 [22 4 [af ai ql 6 | a a| 2 2 6 | 2] al 2] 2 a 7 | 2) 2) a 7 | 2) 2) 1) 1 2 8 || 2i a2 8/2) 2] a] 2) 4‘Taguchi Orhogonal rays, Page 2 z 3 7 3 Taguchi, P Raa#] | 3] o| dl el x 1 fala an 4 1, 2) 2] 2 2k, (Taguchi, P=3,2=3/| [Taguchi P oo [Rundl s] b[e[ X]| Rand) «| & a [at 3 ea ale 1 11 1 1a if 2 2 2) 3) 3\ Xs 2 12 2 us 6 2 3} 3] 1) 1X, eae qa 7 [3 a) 2 a) 3) 3 rac: 2) 4 [24 B32 3/2) a = ola aT al al 9 | 3] 3] a] al 2ix S33 a 10 [4 1) 3] 3) 2m 4 a fa af af a) ian = Tal oe [aia 12 | 1 3 2] 2] ite B21 23, 1% a" | Sis a | ays 14 | 2 2] 3] a] airy 15 | 2) 3{ 1] 2] 3ix5 16 3| 1] 3) 2 3X! 7 [3a a) 3) tm i [3 3] 2 al 2m =3.L=4 Taguchi, P = 4,1 Taguchi, P = 5,1 Run# a) b| X | Run# a 6 c X | Run# a) b c dX | Run# a boc d 1 1 Xx, 1 1a) 1X 1 aja) a} ay x 1 yuu 2 1) 2|% z 1) (2) 2) X2) 1] 2) 2) 2) x2 2 a] 2] 2) 2 = 1} 3} 3) 3 1) 3) 3) X53) 1] 3) 3| 3) Xs 2 1} 3} 3| 3 4 1 4|% 4 104 4X) 164 4 4X 4 1644 4 = 2] 1| Xs s 2| 1) 2) xs 2| 1) 2) 3| 4s S 2) 1] 2) 3] 6 2| 2| Xe 6 2) 2) 1| Xe 2} 2,1 4|% 6 2} 2] 1) 4 ¥ 2| 3] x7] - 2) 3) 4) 2X) 2| 3) 4) 1) 47 7 2] 3) 4, 1 4 8 2) 4) Xs 8 2) 4) 3| Xe 2| 43 2) Xe 8 2| 4) 3| 2) 9 3| 1) Xo) 9 31, 3| Xo! 3) 1, 3) 4| Xo! 9 3} 1] 3| 4 10 3| 2X10 10 3) 2) 4 10 3| 2) 4) 3\Xw 10 3] 2] 4| 3 i 3| 3/Xu ey 3) 3, 1 i 3] 3) 1) 2\tn iW 3] 3] 1) 2 2 3| 4X n) 12 3.4 2 2 3] 4.2 1/%n 12 3] 4) 2) 1 13 4) 1X45) 13 414 13 4] 1) 4) 2% 13 4) 1) 4 2) 14 4) 2X! 14 4, 2] 3 14 4| 2) 3| 1/%u 14 4,23 1 15 4) 3X45) 1s 4, 3| 2 15 4| 3) 2) 4)%s 15 4| 3) 2) 4 16 4| 4X16 16 441 16 4] 4 1 3\% 16 4} 4/ 1) 3‘Taguchi Ochogonal rays, Page 3 3 4 5 Taguchi, P=3.£=5||| Taguchi, P=4,2 Taguchi. P = 5, Runé] a 6c) X\|(Run#| «| bc @ Run#] a] bc d| e| X a [ap ay afta fap ay apa a {a} a) a) al ae 2 | al a] 2|x||{ 2 [af 2] al a 2 | a{ 2] 2) 2] 2| x] 3 | a/ 3) 3/3/|| 3 | a] 3/3) 3 3 | a] 3| 3| 3|_3[ x] 4 [ala 4 | a[ al al alaxl}| 4 [al 4) al al af x s [al 4 5 s [al sl sls s [al s|s/ s| sla 6 | 2[ a] 2 6 | [3/2 6 | 2 a] 2) 3 7 | 2[ 23 7 | 2 23 7 | 2 2] 3| 4 3 | 23] 4 a [23] 4 8 | 2] 3/ als 9 | 2 4 5 9 | 2 4) 5 9 | 2 4[ sf io | 2| 5] 1 10 | 2| 5,1 Ee) n | 3{ a] 32 n | 3{ 4] 3 n [3 a] 3] 4 12 | 3 2] 4 12 | 3| 2| 4 12 | a{ 2| 4) 1 B 3 3 | 3 3| si 13 | 3] 3| 5 13 | a[ 3[ 5| 2] 4 | 3 a4 | 3) 4) 1Xul|] a4 | al 4) a 14 | 3] 4/1) 3) 1s 3 1s | 35) 21Xs||| 1s | 3] 52 1s | 3] 52 4) 16 | 16 | 4) 1) 4[Xu|[ 16 | 4[ a 4 16 | 4] 1] 4) 2| a4 17_| 4) 2) sil} a7 | al 2) 5 17 | al 2[ 5] 3) 4 1s | 4) 3,1 1 | 4{ 3/1 1s | 4 3/1] | 4 19 | 4 4 2 is | al 4[ 2 i | al al 2 5 20 4 20 | 4| 5/3] 20 | 4 5) 3 20 [4[ 5] 3| 1 als a_|s[ a 5 a [sa] 5 zi [sa] 5| al 2 | 3 z [si 2 1 nz [sl 21 zz [s| 2] al sl 23 | 5 B [5] 3 2 B A) 23 3 2| 1 [5 zm [s| al 3 m4 43 24 a] 3|_2| 2 [3 2 [| 5| 5] 4 25 | s| 5] 4 25 | s| 5] al 3| ‘+ Bottom line: Experimental test arrays are usually chosen based on a compromise between the cost of the experiments (cost includes the sime required to run the experiments) and required accuracy of the results. Below is a hierarchy of how you should choose a test array: Increasing | Most is nota big issue (in other words, you have enough time, and the runs are accuracy | inexpensive and don’t take too long), or ifthe accuracy of the results is critical, use a fll factorial array Taguchi Orthogonal Array: Ifthe cost (including time) of a full factorial array analysis is high, and the accuracy of the results is not so critical, use an orthogonal Taguchi array. ‘Taguchi Non-Orthogonal Array: If the cost is prohibitive (runs are extremely expensive or time consuming). and youcan | necreasing accept limited accuracy, use a non-orthogonal Taguchi array (but be sure to optimize it cost using the two rules given in the previous learning module for fractional factorial analysis)