Page 181 - Data Science Algorithms in a Week

The Estimation of Cutting Forces in the Turning of Inconel 718 Assisted … 165

testing data which were not used for the training process shown in Table 2. In order to
evaluate the performance, the predicted cutting forces components were compared with
the experimental values. In order to evaluate the performance of developed training
methods of ANN, the predicted main values of cutting force, feed force and passive force
were compared with the experimental data and summarized in Table 5, Table 6 and Table
7, respectively. The mean absolute percentage errors for main cutting force, feed force
and passive force of BP-based ANN were 5.1%, 5.8% and 6.1%, respectively, which is
considered a good agreement between the simulated outputs and the experimental results.
However, the optimal results obtained using the GA-based and PSO-based ANN models
are even more accurate. The mean absolute percentage errors of GA-based ANN model
for main cutting force, feed force and passive force were 3.8%, 5.3% and 4.2%,
respectively. Finally, mean absolute percentage errors of PSO-based ANN model were
3.8%, 3.7% and 3.8% for main cutting force, feed force and passive force, respectively.
Hence, the learning of ANN using bio-inspired algorithms has demonstrated
improvement in training average error as compared to the backpropagation algorithm.

Figure 5. Results of parametric study for determination of optimal set of PSO parameters.