*2.3. Comparative Methods*

The first test looked at the network architecture with the best accuracy. The 750 parts of the 'trainingdata' database were used. The input variables of these networks were the quantity and volume of each of the features groups and the material of the workpiece. The second one sought to study the influence of the amount of input data on ANN accuracy. For this purpose, the ANN network architecture with the lowest percentage error of Test 1 was used, differing from the amount of data used in training. Test 3 studied the influence of input variables on the accuracy of ANNs. Once again, and keeping the percentage error as a criterion, the best ANN from the previous tests was selected, and the input variables were studied. Workpiece material was always used as input. To understand the tables referring to Test 3, it is essential to define the following:


#### **3. Results**

In this section, the results obtained in each of the tests are presented and analyzed.

#### *3.1. Test 1—Variation of Network Architectures*

From the application of ANN of Test 1 to 'testdata', it was observed that the ANNs with the lowest ME were T1\_01 and T1\_05, the ANN with the lowest MAE was T1\_03, and T1\_07 presented the lowest MSE. The lowest EMax recorded was 19.54 min, in T1\_06. Using the percentage error as a criterion, the ANN with the best performance was T1\_07, with a percentage error of 2.52% (please see Table 3).


#### **Table 3.** Regression and error of the ANNs from Test 1.

Analyzing Table 3, the following conclusions can be drawn:

