Abstract
Robust parameter design (RPD) is an engineering methodology that focuses on reducing the variation of a process by appropriately selecting the setting of its control factors so as to make it less sensitive to noise variation. Then control factors are crucial in achieving robustness. If the control factors and noise factors in such a design are treated as the sub-plot (SP) factors and whole plot factors, fractional factorial split-plot (FFSP) design can be used. A minimum aberration of type SP (SP-MA) criterion is proposed to construct two-level regular FFSP designs, which is based on the FFSP-RPD effect hierarchy principle. We also derive the construction results of SP-MA \(2^{(n_1+n_2)-(k_1+k_2)}\) designs with \(k_1+k_2\le 4\). Some of the SP-MA FFSP-RPDs are better than the MA FFSP-RPDs in Table 5 of Bingham and Sitter (Technometrics 45(1):80–89, 2003). Finally, the SP-MA \(2^{(n_1+n_2)-(k_1+k_2)}\) designs will be searched and tabulated for 8-, 16-, 32- and 64-run designs.
Similar content being viewed by others
References
Ai M, Zhang R (2006) Minimum secondary aberration fractional factorial split-plot designs in terms of consulting designs. Sci China Ser A 49(4):494–512
Bingham D, Sitter RR (1999) Minimum-aberration two-level fractional factorial split-plot designs. Technometrics 41(1):62–70
Bingham D, Sitter RR (1999) Some theoretical results for fractional factorial split-plot designs. Ann Stat 27(4):1240–1255
Bingham D, Sitter RR (2003) Fractional factorial split-plot designs for robust parameter experiments. Technometrics 45(1):80–89
Box GEP, Hunter JS (1961) The \(2^{k-p}\) fractional factorial designs: part I. Technometrics 3(3):311–351 and 449–458
Chang MC (2022) A unified framework for minimum aberration. Stat Sin 32:251–269
Chang MC, Cheng CS (2018) A Bayesian approach to the selection of two-level multi-stratum factorial designs. Ann Stat 46(4):1779–1806
Chen J, Wu CFJ (1991) Some results on \(s^{n-k}\) fractional factorial designs with minimum aberration or optimal moments. Ann Stat 19(2):1028–1041
Cheng SW, Li W, Ye KQ (2004) Blocked nonregular two-level factorial designs. Technometrics 46(3):269–279
Dang HP, Zhao SL, Zhao QQ (2022) Constructing minimum aberration split-plot designs via complementary designs when the subplot factors are more important. Commun Stat Theory Methods. https://doi.org/10.1080/03610926.2022.2031221
Fisher R (1935) The design of experiments, 1st edn. Oliver and Boyd, Edinburgh
Fries A, Hunter WG (1980) Minimum aberration \(2^{k-p}\) designs. Technometrics 22(4):601–608
Han XX, Chen JB, Liu MQ et al (2020) Asymmetrical split-plot designs with clear effects. Metrika 83(7):779–798
Han XX, Liu MQ, Yang JF et al (2020) Mixed 2- and \(2^r\)-level fractional factorial split-plot designs with clear effects. J Stat Plan Inference 204:206–216
Joseph VR (2006) A Bayesian approach to the design and analysis of fractionated experiments. Technometrics 48(2):219–229
Joseph VR, Ai M, Wu CJ (2009) Bayesian-inspired minimum aberration two- and four-level designs. Biometrika 96(1):95–106
Kerr MK (2001) Bayesian optimal fractional factorials. Stat Sin 11(3):605–630
Li W, Zhou Q, Zhang RC (2015) Effective designs based on individual word length patterns. J Stat Plan Inference 163:43–47
Li W, Mee RW, Zhou Q (2019) Using individual factor information in fractional factorial designs. Technometrics 61(1):38–49
Lin CY (2014) Optimal blocked orthogonal arrays. J Stat Plan Inference 145:139–147
Mitchell TJ, Morris MD, Ylvisaker D (1995) Two-level fractional factorials and Bayesian prediction. Stat Sin 5(2):559–573
Mukerjee R, Fang KT (2002) Fractional factorial split-plot designs with minimum aberration and maximum estimation capacity. Stat Sin 12(3):885–903
Mukerjee R, Wu CFJ (2006) A modern theory of factorial design. Springer, New York
Taguchi G (1987) The system of experimental design. Kraus International Publications, New York
Tichon JG, Li W, Mcleod RG (2012) Generalized minimum aberration two-level split-plot designs. J Stat Plan Inference 142:1407–1414
Wang CC, Zhao QQ, Zhao SL (2019) Optimal fractional factorial split-plot designs when the whole plot factors are important. J Stat Plan Inference 199:1–13
Yan Z, Zhao S (2022) Mixed two- and four-level split-plot designs with combined minimum aberration. Metrika 85(5):537–555
Yang JF, Liu MQ, Zhang RC (2009) Some results on fractional factorial split-plot designs with multi-level factors. Commun Stat Theory Methods 38(20):3623–3633
Yates F (1937) Design and analysis of factorial experiments. Imperial Bureau of Soil Sciences, London
Zhao QQ, Zhao SL (2020) Constructing minimum aberration split-plot designs via complementary sets when the whole plot factors are important. J Stat Plan Inference 209:123–143
Zhao SL, Chen XF (2012) Mixed-level fractional factorial split-plot designs containing clear effects. Metrika 75(4):601–608
Zhao SL, Chen XF (2012) Mixed two- and four-level fractional factorial split-plot designs with clear effects. J Stat Plan Inference 142(7):1789–1793
Acknowledgements
The authors would like to thank the associate editor and anonymous reviewers for their helpful and constructive comments. This work was supported by the National Natural Science Foundation of China (Grant No. 12071096).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
On behalf of all authors, the corresponding author states that there is no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Below is the link to the electronic supplementary material.
Appendix: Proofs of lemmas and theorems
Appendix: Proofs of lemmas and theorems
Proof of Lemma 2.3
Suppose d is an SP-MA \(2^{(n_{1}+n_{2})-(k_{1}+k_{2})}\) design with \(n_{1}\) WP factors \(Q_{1},\ldots , Q_{n_{1}}\) and \(n_{2}\) SP factors \(R_{n_{1}+1},\ldots ,R_{n_{1}+n_{2}}\). Suppose the independent defining words of d are
where \(U_{i}\) (\(i=1,\ldots ,k_{1}\)) and \(V_{j}\) (\(j=1,\ldots ,k_{2}\)) denote the WP and SP type defining words, respectively. Without loss of generality, suppose \(R_{n_{1}+n_{2}}\) is not involved in any defining words of d. Suppose \(d'\) is determined by the following independent defining words
Clearly, \(d'\) is also a \(2^{(n_{1}+n_{2})-(k_{1}+k_{2})}\) design. We compare \(W_{1}(d)\) with \(W_{1}(d')\). For any SP type defining word of \(d'\) which does not involve the letter \(R_{n_{1}+n_{2}}\), it is also the SP type defining word of d. For the SP type defining word of \(d'\) which involves the letter \(R_{n_{1}+n_{2}}\), say \(V_{1}R_{n_{1}+n_{2}}\), it has one more factor \(R_{n_{1}+n_{2}}\) than the corresponding SP type defining word \(V_{1}\) of d. Hence \(d'\) has less aberration of SP type than d. This contradicts the assumption that d is an SP-MA \(2^{(n_{1}+n_{2})-(k_{1}+k_{2})}\) design. This completes the proof. \(\square \)
Proof of Theorem 3.1
The design \(d_{0}\) has only SP type defining words because the WP part of \(d_{0}\) is a full \(2^{n_1}\) design. Then \(W_1(d_0)=W(d)\). By Definition 2.2, if the design d is an MA \(2^{n-k_{2}}\) design, then the design \(d_{0}\) is an SP-MA design. \(\square \)
Proof of Theorem 3.2
The WP part of design \(d_{0}\) is a \(2^{n_{1}-k_{1}}\) design d. Note that the design \(d_{0}\) has only WP type defining words. Then \(A_{i,0}(d_0)=A_i(d)\), for \(i=1,\ldots ,n_1\). Thus the \(2^{(n_{1}+n_{2})-(k_{1}+0)}\) design \(d_{0}\) is an SP-MA design if and only if the design d is an MA \(2^{n_{1}-k_{1}}\) design. \(\square \)
Proof of Theorem 3.3
The SP wordlength pattern of the design \(d_{0}\) determined by (1) is
where \(A_{n_{1}+n_{2}-2,1}(d_{0})=1\), \(A_{n_{1}+n_{2}-1,1}(d_{0})=1\). To prove that \(d_0\) is an SP-MA design, we need only to consider the \(2^{(n_{1}+n_{2})-(1+1)}\) designs d with \(A_{i,1}(d)=0,i=1,\ldots , n-3 \) and \(A_{n-2,1}(d)\le 1\), where \(n=n_{1}+n_{2}\). By Corollary 2.5(1) and (2), for any such design, we have
Note that \(\sum _{i=1}^n A_{i,0}(d)=1\) and only designs with \(A_{1,0}(d)=A_{2,0}(d)=0\) are considered. Subtracting \(n-2\) times the Eq. (5) from the Eq. (6) yields
By Corollary 2.5(3), the Eq. (7) has the unique solution \(A_{n-2,1}(d)=1\), \(A_{n-1,1}(d)=1\), \(A_{3,0}(d)=1\). Hence the \(2^{(n_{1}+n_{2})-(1+1)}\) design \(d_{0}\) is an SP-MA design. \(\square \)
Proof of Theorem 3.5
The proof is divided into three parts according to \(r_{1}+r_{2}=0,1,2\) modulo 3.
(1) For \(r_{1}+r_{2}=0\) modulo 3, we first consider the case \(r_{1}+r_{2}=0\), that is \(r_{1}=0\) and \(r_{2}=0\). By (2) and Table 1, we get
The SP wordlength pattern of the design \(d_{0}\) is
where \(A_{2m+2l-1,1}(d_{0})=3\), \(A_{2m+2l,1}(d_{0})=3\). To prove that \(d_0\) is an SP-MA design, we need only to consider the \(2^{(n_{1}+n_{2})-(1+2)}\) designs d with \(A_{i,1}(d)=0, i=1,\ldots , 2\,m+2\,l-2\) and \(A_{2\,m+2\,l-1,1}(d)\le 3\). By Corollary 2.5(1) and (2), for any such design, we have
Note that \(\sum _{i=1}^n A_{i,0}(d)=1\) and only designs with \(A_{1,0}(d)=A_{2,0}(d)=0\) are considered. Subtracting \(2m+2l-1\) times the Eq. (8) from the Eq. (9) yields
where \(n=n_1+n_2\). By Corollary 2.5(3), the Eq. (10) has the unique solution \(A_{2m+2l-1,1}(d)=3\), \(A_{2m+2l,1}(d)=3\), \(A_{3,0}(d)=1\). Hence the \(2^{(n_{1}+n_{2})-(1+2)}\) design \(d_{0}\) determined by (2) is an SP-MA design.
Now, consider the case \(r_{1}+r_{2}=3\), that is \(r_{1}=1\) and \(r_{2}=2\), or \(r_{1}=2\) and \(r_{2}=1\). From (2) and Table 1, we can get the three independent defining words as
for \(r_{1}=1\) and \(r_{2}=2\), or
for \(r_{1}=2\) and \(r_{2}=1\). The SP wordlength pattern of the design \(d_{0}\) is
where \(A_{2m+2l+1,1}(d_{0})=3\), \(A_{2m+2l+2,1}(d_{0})=3\). The rest of the proof is similar to that for \(r_{1}+r_{2}=0\) and is omitted.
(2) For \(r_{1}+r_{2}=1\) modulo 3, that is \(r_{1}=0\), \(r_{2}=1\), or \(r_{1}=1\), \(r_{2}=0\), or \(r_{1}=2\), \(r_{2}=2\), the proofs are similar. We only prove the case of \(r_{1}=0\), \(r_{2}=1\). By (2) and Table 1, we get
The SP wordlength pattern of the design \(d_{0}\) is
where \(A_{2m+2l-1,1}(d_{0})=1\), \(A_{2m+2l,1}(d_{0})=3\), \(A_{2m+2l+1,1}(d_{0})=2\). To prove that \(d_0\) is an SP-MA design, we need only to consider the \(2^{(n_{1}+n_{2})-(1+2)}\) designs d with \(A_{i,1}(d)=0\), \(i=1, \ldots , 2\,m+2\,l-2 \) and \(A_{2\,m+2\,l-1,1}(d)\le 1\). By Corollary 2.5(1) and (2), for any such design, we have
where \(n=n_1+n_2\). Note that \(\sum _{i=1}^n A_{i,0}(d)=1\) and only designs with \(A_{1,0}(d)=A_{2,0}(d)=0\) are considered. Subtracting \(2m+2l-1\) times the Eq. (11) from the Eq. (12) yields
By Corollary 2.5(3), two solutions of the Eq. (13) meet our requirements: \(A_{2m+2l,1}(d)=6\), \(A_{4,0}(d)=1\) and \(A_{2\,m+2\,l-1,1}(d)=1\), \(A_{2m+2l,1}(d)=3\), \(A_{2m+2l+1,1}(d)=2\), \(A_{3,0}(d)=1\). Suppose \(d_1\) is a design with \(A_{2m+2l,1}(d_1)=6\), \(A_{4,0}(d_1)=1\). Let l be an SP factor in the SP defining words. By deleting all the words containing l, the remaining words define a \(2^{(n_1+n_2-1)-(1+1)}\) design \(d_2\) with \(A_{2m+2l,1}(d_2)=2\), \(A_{4,0}(d_2)=1\). From Corollary 2.5(2), we get \(2(2m+2l)+4=2(3m+3l)\), then \(m+l=2\), that is \(m=1\) and \(l=1\). Since \(A_{4,0}(d)=1\), we have \(r_1=1\) and \(r_2=0\). Hence \(n_1=4\), \(n_2=3\). Without loss of generality, the WP factors are denoted as \(w_1,w_2,w_3,w_4\) and the SP factors are denoted as \(s_5,s_6,s_7\). Three independent defining words of the design \(d_1\) is \( u=w_1w_2w_3w_4, v_{1}=w_3w_4s_5s_6, v_{2}=w_1w_3s_5s_7 \) and the wordlength pattern of the design \(d_1\) is
By Definition 2.2, when \(m+l=2\), the design \(d_1\) is optimal; when \(m+l>2\), the design \(d_0\) is optimal.
(3) For \(r_{1}+r_{2}=2\) modulo 3, that is \(r_{1}=0\), \(r_{2}=2\), or \(r_{1}=1\), \(r_{2}=1\), or \(r_{1}=2\), \(r_{2}=0\), the proofs are similar. We only prove the case of \(r_{1}=0\), \(r_{2}=2\). By (2) and Table 1, we get
The SP wordlength pattern of the design \(d_{0}\) is
where \(A_{2m+2l,1}(d_{0})=2\), \(A_{2m+2l+1,1}(d_{0})=3\), \(A_{2m+2l+2,1}(d_{0})=1\). To prove that \(d_0\) is an SP-MA design, we need only to consider the \(2^{(n_{1}+n_{2})-(1+2)}\) designs d that have \(A_{i,1}(d)=0\), \(i=1, \ldots , 2\,m+2\,l-1\) and \(A_{2\,m+2\,l,1}(d)\le 2\). By Corollary 2.5(1) and (2), for any such design, we have
where \(n=n_1+n_2\). Note that \(\sum _{i=1}^n A_{i,0}(d)=1\) and only designs with \(A_{1,0}(d)=A_{2,0}(d)=0\) are considered. Subtracting \(2m+2l\) times the Eq. (14) from the Eq. (15) yields
By Corollary 2.5(3), two solutions of (16) meet our requirements: \(A_{2m+2l,1}(d)=2\), \(A_{2m+2l+1,1}(d)=3\), \(A_{2m+2l+2,1}(d)=1\), \(A_{3,0}(d)=1\) and \(A_{2m+2l,1}(d)=2\), \(A_{2m+2l+1,1}(d)=4\), \(A_{4,0}(d)=1\). By Definition 2.2, the design d with \(A_{2m+2l,1}(d)=2\), \(A_{2m+2l+1,1}(d)=3\), \(A_{2m+2l+2,1}(d)=1\), \(A_{3,0}(d)=1\) is optimal. Hence the \(2^{(n_{1}+n_{2})-(1+2)}\) design \(d_{0}\) determined by (2) is an SP-MA design. This completes the proof of Theorem 3.5. \(\square \)
Proof of Theorem 3.7
The proof can be divided into seven parts according to \(r_{1}+r_{2}=0,1,\ldots ,6\) modulo 7. All proofs of the seven cases are similar. We only give the proof of \(r_{1}+r_{2}=0\) to save space.
When \(r_{1}+r_{2}=0\), that is \(r_{1}=0\) and \(r_{2}=0\), by (3), we get
The SP wordlength pattern of the design \(d_{0}\) is
where \(A_{4m+4l-1,1}(d_{0})=6\), \(A_{4m+4l,1}(d_{0})=7\), \(A_{4m+4l+3,1}(d_{0})=1\). To prove that \(d_0\) is an SP-MA design, we only need to consider the \(2^{(n_{1}+n_{2})-(1+3)}\) designs d that have \(A_{i,1}(d)=0, i=1,\ldots , 4\,m+4\,l-2\) and \(A_{4\,m+4\,l-1,1}(d)\le 6\). By Corollary 2.5(1) and (2), for any such design, we have
where \(n=n_1+n_2\). Note that \(\sum _{i=1}^n A_{i,0}(d)=1\) and only designs with \(A_{1,0}(d)=A_{2,0}(d)=0\) are considered. Subtracting \(4m+4l-1\) times the Eq. (17) from the Eq. (18) yields
From (19), we obtain the wordlength patterns of possible designs having less aberration of type SP than \(d_{0}\)
In the following, we show that no design has the wordlength patterns (20–22). Suppose \(d_1\) is a design with one of the possible wordlength patterns (20–22). Corollary 2.5(3) and the Eqs. (20–22) imply that \(d_1\) has 8 odd-length defining words. By adding a new letter \(n_{1}+n_{2}+1\) to each of the odd-length defining words, we have a \(2^{(n_{1}+n_{2}+1)-(1+3)}\) design \(d_2\). Clearly, every defining word of \(d_2\) has even length and \(d_2\) has the wordlength pattern as
Now we prove the impossibility of (23). Clearly, there exists a letter l appearing in one of the two longest words but not in the other. By deleting all the words containing l, the remaining words define a \(2^{(n_{1}'+n_{2}')-(1+2)}\) design \(d_{3}\) or a \(2^{(n_{1}'+n_{2}')-(0+3)}\) design \(d_{4}\) with \(n_{1}'+n_{2}'\le n_{1}+n_{2}\). From Corollary 2.5(1), the wordlength patterns of designs \(d_{3}\) and \(d_{4}\) must be
or
Then, we have
and
Both violates Corollary 2.5(2). Therefore such a design \(d_2\) does not exist. Hence the \(2^{(n_{1}+n_{2})-(1+3)}\) design \(d_{0}\) determined by (3) is an SP-MA design. \(\square \)
The proof of Lemma 3.9 is similar to that of Lemma 2.3 and is omitted.
Proof of Theorem 3.10
The SP wordlength pattern of the design \(d_{0}\) determined by (4) is
where \(A_{n_{1}+n_{2}-3,1}(d_{0})=2\), \(A_{n_{1}+n_{2}-2,1}(d_{0})=2\). To prove that \(d_0\) is an SP-MA design, we need only to consider the \(2^{(n_{1}+n_{2})-(2+1)}\) designs d that have \(A_{i,1}(d)=0,i=1, \ldots , n-4 \) and \(A_{n-3,1}(d)\le 2\), where \(n=n_{1}+n_{2}\). By Corollary 2.5(1) and (2), for any such design, we have
where \(n=n_1+n_2\). Note that \(\sum _{i=1}^n A_{i,0}(d)=3\) and only designs with \(A_{1,0}(d)=A_{2,0}(d)=0\) are considered. Subtracting \(n-3\) times the Eq. (24) from the Eq. (25) yields
By Corollary 2.5(3), three solutions of the Eq. (26) meet our requirements: \(A_{n-3,1}(d)=2\), \(A_{n-2,1}(d)=2\), \(A_{3,0}(d)=2\), \(A_{4,0}(d)=1\) and \(A_{n-3,1}(d)=2\), \(A_{n-2,1}(d)=1\), \(A_{n-1,1}(d)=1\), \(A_{3,0}(d)=3\), where n is odd number and \(A_{n-3,1}(d)=1\), \(A_{n-2,1}(d)=3\), \(A_{3,0}(d)=3\),where n is even number. Note that the WP part of a \(2^{(n_{1}+n_{2})-(2+1)}\) design d is a \(2^{n_{1}-2}\) design \(d^*\) which is determined by the WP type defining words of d. Thus, \(A_{3,0}(d)=3\) in the second and third solution above implies that \(A_{3}(d^*)=3\), i.e., all the three defining words of \(d^*\) have length 3. This contradicts Lemma 2.4(3). Hence the \(2^{(n_{1}+n_{2})-(2+1)}\) design \(d_{0}\) determined by (4) is an SP-MA design. This completes the proof of Theorem 3.10. \(\square \)
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Jiang, H., Zhang, C. Optimal two-level regular fractional factorial split-plot designs when the effects of subplot factors are more important. Metrika 87, 75–103 (2024). https://doi.org/10.1007/s00184-023-00907-3
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00184-023-00907-3