Taguchi Design of Experiments Online: Orthogonal Arrays and S/N Analysis
What Is the Taguchi Method?
The Taguchi Method, proposed by Genichi Taguchi, aims to find parameter combinations insensitive to noise so that performance stays stable under raw-material variation, environmental change and manufacturing error - robust design. It arranges experiments with orthogonal arrays and evaluates robustness with the signal-to-noise (S/N) ratio, making it a core parameter-design tool in quality engineering.
Inner and Outer Arrays
Taguchi design splits factors into control factors (set by the designer, such as material and dimension) and noise factors (hard to control, such as temperature, humidity and batch variation). Control factors go into the inner array (e.g., L9), noise factors into the outer array (noise level combinations), and the cross product forms the complete experiment: every inner-array combination is tested under multiple noise conditions so the S/N ratio of each parameter combination can be computed and the most robust level - response close to target and insensitive to noise - selected.
Choosing the S/N Ratio
The S/N ratio follows the quality characteristic type: larger-the-better (strength, life) uses S/N = -10*log(Sum(1/yi^2)/n); smaller-the-better (defects, wear) uses S/N = -10*log(Sum(yi^2)/n); nominal-the-best (dimensions) uses S/N = 10*log(ybar^2/s^2). A larger S/N means a stronger signal relative to noise, i.e., more robust; the tool computes the right formula automatically by characteristic type.
How to Use It
After sign-in, select an orthogonal array (L9 for 4 three-level factors, L16, L27, etc.), assign control factors to inner-array columns, set noise factors and the outer array, and the tool generates the full cross-product experiment plan. Run it and enter the responses under each noise condition, then the tool computes the mean and S/N per combination and outputs response tables and charts for choosing the level combination with the largest S/N as the robust optimum. Always run a confirmation experiment to verify the robustness gain is repeatable.