Table 3 Comparison of computational costs of different models.

From: Inverse solution of process parameters in gear grinding using hierarchical bayesian physics informed neural network (HBPINN)

 

Training set

size

Hidden layer

structure

Training

time (s)

R2

HBPINN

200

256-512-256

1.94

Vw*

0.9918

f *

0.9814

Vs*

0.9155

Average value

0.9629

BPINN

200

256-512-1024-512-256

9.33

Vw*

0.9893

f *

0.9986

Vs*

0.9008

Average value

0.9629

VI-BPINN

200

256-512-1024-512-256

8.21

Vw*

0.9841

f *

0.9968

Vs*

0.9008

Average value

0.9606

PINN

2000

256-512-1024-512-256

19

Vw*

0.9778

f *

0.9872

Vs*

0.9225

Average value

0.9625