A FEW WORDS ABOUT TAGUCHI METHODS

Genuchi Taguchi started developing these quality improvement techniques in the 1950's. The three key concepts are :

• the continuous loss function
• off-line quality control
• on-line quality control.

The continuous loss function describes the effect of deviation from a target value as a function of the deviation. The effect is zero when there is no deviation, and increases continuously as the deviation increases. This is a more realistic measure of quality than the usual binary states of in-tolerance and out-of-tolerance. It implies that when a product is in-tolerance but not at the target value, there is still need for quality improvement.

Taguchi identifies three possible phases in product design - system design, parameter design and tolerance design. During the system design phase, overall conceptual design takes place. Theoretical engineering, scientific knowledge and practical experience are used to ensure that the product should function with the required behaviour.

During parameter design, target values of product and process parameters should be chosen either by analytical methods or by carefully planned experiments, so as to minimize variability.

Taguchi points out that although many companies neglect parameter design, and only carry out system design and tolerance design, it is almost impossible to produce a high-quality product without parameter design.

During tolerance design, the designer sets tolerances on parts and processes such that the product will meet requirements at minimum cost.

The Taguchi approach is particularly relevant to the parameter design phase in which the designer sets the value of design parameters (i.e. assigns specific values for product and process parameters) to get a stable reliable product. If the values are correct, the product will perform best when these values are attained.

Since any deviation from these target values reduces product quality, and since deviation is always likely to occur (e.g. due to temperature changes, aging of parts, manufacturing variability) the parameters should be set to optimize performance whilst minimizing deviation.

Generally, since there is no exact theoretical relationship between the design parameters and the fluctuations, the only way to find out the values of design parameters that minimize variability is to carry out experiments that show how the factors causing fluctuation affect performance. In view of the large number of parameters and the numerous combinations possible, it is generally impractical to investigate all possible combinations. Taguchi's experimental design techniques allow designers to experiment with a large number of variables with relatively few experiments.

Copyright 1998 by John Stark