Robust Engineering

A Methodology developed by Taguchi.

“Robustness is the state where the technology, product, or process performance is minimally sensitive to factors causing variability (either in the manufacturing or users environment) and aging at the lowest unit manufacturing cost.” Dr. Genichi Taguchi.
The key to achieving this is to maximise the signal to noise ratio. In other words to make the design sensitive to inputs that can be controlled (signals), but insensitive to uncontrolled inputs (noise). Sounds reasonable doesn’t it, but not always easy to achieve.

Consequently, he developed a strategy for quality engineering that aims to achieve this. The process has three stages:

  • System design

Initial high-level conceptual design

  • Parameter (measure) design

Once the concept is established, the nominal values of the various dimensions and design parameters should be chosen so as to minimize the effects on performance arising from variation in manufacture, environment and cumulative damage. Robust parameter designs consider controllable and uncontrollable noise variables; and to minimize sensitivity to the noise variables.

  • Tolerance design

When the nominal setting for the parameters has been established, the final step is to establish tolerances for these that provide an acceptable balance between manufacturability and product performance.

Design of Experiments

As an efficient way of establishing the sensitivity to variables (both signal and noise), Taguchi made use of orthogonal arrays. See more about Design of Experiments here.

Loss Function

To understand the effect of tolerances better, Taguchi came up with the idea of a ‘loss function’, which defines quality in a negative manner, “Quality is the loss imparted to society from the time the product is shipped”. The further away from the nominal (the ideal setting of a parameter), the greater the loss would be. Such losses are, of course, very small when an item is near to nominal. but grow as we diverge from nominal.

Taguchi specified three situations:

  1. Larger the better (for example, agricultural yield);
  2. Smaller the better (for example, carbon dioxide emissions); and
  3. On-target, minimum-variation (for example, a mating part in an assembly).

The first two cases are represented by simple monotonic loss functions. In the third case, Taguchi adopted a squared-error loss function.

Loss Function



Everyone should be designing and making robust products. Contact us if you want to discuss how you could be doing this in your company.