5。1。2。 ANOVA results for experimental

Table 4c shows the ANOVA result of the dxexp。 A ‘‘Model F value” of 63。22 with a ‘‘Model P value” of less than 0。0001 implies that the selected model is significant。 A ‘‘P value” for the model term ‘‘A” (the melt temperature) and ‘‘D” (the packing pressure) are less than 0。0001, which are less than 0。05, indicating that the model term  ‘‘A”  and  ‘‘D”  are  significant。  There  are  a  interaction   term

‘‘AD” that has impacts to the dxexp。 Table 4d shows the ANOVA re-

sult of the dyexp。 A ‘‘Model F value” of 80。38 with a ‘‘Model P value” of less than 0。0001 implies that the selected model is significant。 A ‘‘P value” for the model term ‘‘A” (the melt temperature) and ‘‘D” (the packing pressure) are less than 0。0001, which are less than 0。05, indicating that the model term ‘‘A” and ‘‘D” are significant。 Additionally, a interaction terms ‘‘AD”, also have significant influ- ences  to  the dyexp。

5。2。 Regression models

Considering  the  most significant terms  from ANOVA result   of

dxsim, dysim, dxexp, and dyexp, regression models can be developed。

10756 C。-P。    Chen    et    al。  /    Expert    Systems  with    Applications    36    (2013)      10752–10759

SS 0:00275186

5。2。1。 Regression models for simulation

From Tables 4a and 4b, mathematic predicted models for the

dxsim, and the dysim  can be shown as    follows:

2 Model

SSTotal   ¼ 0:00280388 ¼

96:32% ð8Þ

dxsim ¼ 0:62857 — 1:295 × 10—3 × A — 1:27048 × 10—3 × C

4 —6

5。2。2。 Regression models for  experimental

Similarly, from Tables 4a and 4b, mathematic predicted  models

— 5:16667 × 10—

6

× D þ 5:28571 × 10

× A × C

for the

dxexp

and

dyexp

are shown as  follows:

— 9:42857 × 10—

× C × D ð5Þ 3

dy ¼ 1:34204 — 3:20867 × 10—3

× A — 1:74333 × 10—   × C

dxexp  ¼ 2:25257 — 6:2 × 10

4

× A — 0:03546 × D

— 0:018085 × D þ 4:96 × 10

× A × D ð6Þ

þ 1:05 × 10—

× A × D ð9Þ

3

Furthermore, by investigating the correlation coefficients, R2, which

dyexp  ¼ 3:50397 — 9:64875 × 10—