Issue 67

S. Chinchanikar et alii, Frattura ed Integrità Strutturale, 67 (2023) 176-191; DOI: 10.3221/IGF-ESIS.67.13

Feed ( f ) (mm/rev)

Expt. run

Cutting speed ( V ) (m/min)

Depth of cut ( d ) (mm)

Number of observations used for ANN

Observations can be referred from

Training

Testing

15 10

3 2 2 2 4 4 3 5 3 3 3 3 2 2 3

Fig. 3 Fig. 3 Fig. 3 Fig. 3 Fig. 3 Fig. 3 Fig. 3 Fig. 3 Fig. 4 Fig. 4 Fig. 4 Fig. 4 Fig. 4 Fig. 4

300 350 350 250 250 300 300 300 200 400 350 250 350 250 300

0.1

0.5 0.4 0.4 0.4 0.4 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.2 0.1

R1 R2 R3 R4 R5 R6 R7 R8 R9

0.08 0.12 0.08 0.12 0.05

9

10 10

6

13 14

0.1

0.15

9 7

0.1 0.1

R10 R11 R12 R13 R14 R15

11 14 16 12 14

0.08 0.12 0.12 0.08

0.1 Fig. 4 Table 4: Training and testing data set (flank wear observations) used for an ANN model.

The performance of the ANN model was evaluated by varying the number of hidden layers and neurons within them. Examining learning curve graphs is a typical practice for determining the convergence of neural network models. Typically, a graph of loss (or error) versus epoch is displayed. It is anticipated that as the number of training epochs rises, the accuracy will increase, and the loss will fall. After a while, they will, however, stabilize. After undergoing several training epochs, a neural network should eventually converge. Tab. 5 depicts the performance of ANN networks varying with the number of hidden layers and neurons within them. Mean squared error (MSE), computational time, and regression coefficient (R) values were used as performance criteria for the best model selection.

No. of hidden layers

No. of hidden neurons

Epoch Learning rate

Computational time (seconds)

Mean squared error (x10 -5 )

Regression coefficient (Overall)

1 1 1 3 3 3 5 5 5

10 30 50 10 30 50 10 30 50

1000 1000 1000 1000 1000 1000 1000 1000 1000

0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001

12 29 38

3.43 3.26 3.84 3.05 3.11 3.77 3.99 3.37 4.46

0.9975 0.9986 0.9979 0.9983 0.9983 0.9976 0.9967 0.9961 0.9951

70.2 123

135.6

91.2

139.2 142.2

Table 5: Performance of ANN networks.

From Tab. 5, the computational time (the time an ANN model took to converge) can be seen as increasing with the number of hidden layers and neurons within them. The lowest mean squared error and the highest regression coefficient (overall for the entire data set) can be obtained with one hidden layer and 30 neurons. The number of neurons can be seen as having a greater impact on ANN performance than hidden layers. It is found that the flank wear growth can be reliably predicted using an ANN model with one hidden layer and 10–30 neurons, considering a maximum number of epochs and the learning

186