Error Determination

For each analysis the error can be calculated. During the investigations carried out at the IAG it was recognized that this error is very large if one or more simulations did not converge. This is obvious since the deviation caused by the non-convergence is usually larger than the deviation caused by the small parameter variations of the Taguchi analysis. Thus the convergence of every simulation is crucial and absolutely necessary for a reliable quantification of the parameter influences and interactions. The error that occurred can be evaluated similar to the influences in the ANOVA analysis.

No. of column

1

2

3

4

5

6

7

1

3

2

5

4

7

6

2

1

6

7

4

5

3

7

6

5

4

4

1

2

3

5

3

2

6

1

7

Fig. 3 Lg-Interaction-Matrix

The degree of freedom of the error term is calculated as follows,

Error Determination

i=1 j=1

Подпись: (13)

If the degree of freedom is non-zero, the error variance can be determined.

The degree of freedom becomes zero if all columns of the Taguchi matrix are oc­cupied with parameters. In this case an error determination as shown above is not possible. Instead the so-called pooling-method must be used. This method takes the parameters with the smallest influences to determine an error term. Following equation describes this method exemplarily for two parameters C and D.

Error Determination

Error Determination
Подпись: (12)

(15)

In [2] it is mentioned that at least half of the degrees of freedom ftotal should be used for this method to determine the error of the investigation. In the present investiga­tion all runs except run 1 are performed with Taguchi matrices which are not fully packed with parameters. Therefore usually the first error determination method is used.