Wednesday, June 23, 2004

Taguchi Methods for System testing

Recently, came across a good technique called the 'Taguchi method for testing' that's apparenty very prevalent in American manufacturing industry for quite some time. The testing technique heavily uses the Orthogonality Theorem that talks of acceptable probabilities for 2 mutually exclusive events in pairs.

In my learning, two of my friends, Unni and Deepa helped me get clarified on some of the nuances of using Taguchi's methods, especially where multiple levels were concerned.

Basically, Taguchi describes a method to test the functionalities of a particular product given a set of non-random variables affecting its performance, in combination. Assuming there are 3 (A,B,C) factors and each factor can take 2 (1,2) values, then by rote, there are 2 (power) 3 possible combinations of these factors, i.e.
{(1,1,1), (1,1,2), (1,2,1), (1,2,2), (2,1,1), (2,1,2), (2,2,1), (2,2,2)}.

Now, Taguchi defines an array through which helps do the set of tests in just 4 experiments, with the following combinations:
{(1,1,1), (1,2,2), (2,1,2), (2,2,1)}.

As the number of parameters increases, this reduction in number of tests to be carried out increases, thus making the testing technique very beneficial. Am planning to try it out in some of the experiments I do in course of my work.