Sensitivity Analysis-integrated Taguchi Method
Also known as: Taguchi-SA, SA-Taguchi, Taguchi method with sensitivity analysis, sensitivity-enhanced robust design
The sensitivity analysis-integrated Taguchi method augments the classical Taguchi robust design workflow with a systematic sensitivity analysis step that quantifies how much each control factor and noise factor contributes to response variability. By combining Taguchi orthogonal arrays with variance-based or ANOVA-based sensitivity indices, engineers can both optimize process settings and rank factors by their influence on output uncertainty, yielding more transparent and defensible engineering decisions.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use this workflow when you need both optimal robust settings (the classical Taguchi goal) AND a ranked understanding of which factors drive output uncertainty — particularly in manufacturing, materials processing, or product design where not all factors can be tightly controlled and resource allocation decisions must be justified. It is especially valuable when the number of factors is moderate (3–8), when noise factors are explicitly known, and when factor interaction screening is important. Do not use when the experimental budget allows only the minimum Taguchi runs with no replication — sensitivity estimation requires either replicated runs or a complementary simulation model. Also avoid when the response is categorical or ordinal; ANOVA-based sensitivity analysis assumes a continuous response.
Strengths & limitations
- Combines the resource efficiency of Taguchi orthogonal arrays with the interpretive power of sensitivity indices, yielding richer insight from the same experimental budget.
- Explicitly ranks control and noise factors by their contribution to output variance, enabling principled decisions about where to invest process control resources.
- Compatible with both physical experiments and computer simulation (metamodel-based sensitivity), extending applicability to costly or slow experiments.
- Transparent and auditable: the factor importance ranking is quantified, making the basis for engineering decisions explicit.
- Supports simultaneous optimization of mean and variance targets through the combined S/N and sensitivity framework.
- Requires more experimental runs or replications than a minimal Taguchi array alone if variance-based sensitivity indices are to be estimated reliably.
- Global (Sobol-type) sensitivity analysis presupposes independent control factors; correlated factors require specialized estimators that add complexity.
- The interpretation of sensitivity indices from small orthogonal arrays (e.g., L9) may be statistically uncertain; large arrays or simulation augmentation improve reliability.
- Factor interactions that Taguchi arrays deliberately confound can distort the sensitivity index estimates unless higher-resolution designs or follow-up experiments are used.
Frequently asked
Is variance-based sensitivity analysis the same as Taguchi's ANOVA percentage contribution?
No. Taguchi's ANOVA percentage contribution partitions the sum of squares from the orthogonal array model — it is a model-dependent, local measure that assumes the orthogonal array captures all variation. Sobol-type variance-based sensitivity indices (S_i) are global measures computed by averaging over the full input space and can capture non-linear and interaction effects. ANOVA contribution is easier to compute from a small array; Sobol indices are more rigorous but require more runs or a surrogate model.
Can I apply sensitivity analysis to an L9 array with only 9 runs?
ANOVA-based sensitivity (percentage contribution) can be computed from any complete orthogonal array. However, variance-based Sobol indices estimated from only 9 points are statistically unreliable. For rigorous global sensitivity analysis you typically need a surrogate model (e.g., response surface or kriging) fitted to the Taguchi runs, then Monte Carlo sampling through that surrogate to estimate sensitivity indices.
When should I prefer full factorial design over this workflow?
Use full factorial design when the number of factors is small (2–4) and you need to estimate all main effects and interactions without confounding. The Taguchi orthogonal array approach trades resolution for run economy; when run cost is low and interactions are critical, full factorial is preferable. Sensitivity analysis can be applied after either design type.
Does this method require specialized software?
Basic ANOVA-based sensitivity can be computed in any statistical package (R, Minitab, Excel add-ins). Global Sobol indices require packages such as the R sensitivity package or Python SALib. Taguchi orthogonal array generation is available in Minitab, JMP, and several R packages (e.g., DoE.base).
How does this differ from the Robust Taguchi method?
The robust Taguchi method focuses on selecting factor levels that minimize sensitivity to noise factors using S/N ratios — robustness is the goal. The sensitivity analysis-integrated variant additionally quantifies and reports the variance contribution of each factor as a numerical index, enabling factor importance ranking and guiding decisions about which factors to invest in controlling. It is a more analytically transparent extension of the robustness objective.
Sources
- Taguchi, G. (1987). System of Experimental Design: Engineering Methods to Optimize Quality and Minimize Costs (Vols. 1–2). UNIPUB/Kraus International Publications. ISBN: 978-0527916213
- Saltelli, A., Tarantola, S., Campolongo, F., & Ratto, M. (2004). Sensitivity Analysis in Practice: A Guide to Assessing Scientific Models. Wiley. ISBN: 978-0470870938
How to cite this page
ScholarGate. (2026, June 3). Sensitivity Analysis-integrated Taguchi Method. ScholarGate. https://scholargate.app/en/experimental-design/sensitivity-analysis-integrated-taguchi-method
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Design of experimentsExperimental design↔ compare
- Optimization-assisted Taguchi methodExperimental design↔ compare
- Response Surface MethodologyExperimental design↔ compare