This chapter presents an approach to augmentation of experimental data used for training neural networks in titanium alloy turning problems. Given the high cost of conducting experiments, a method of artificial sample augmentation based on regression models built on a limited dataset was proposed. The mathematical regression models were used to generate additional data points, resulting in an expanded database of 2000 examples. This data was used to train neural networks whose task was to predict the surface roughness of titanium alloy VT6. During the study, two experiments were conducted with different data preprocessing techniques: standardization and normalization. The results showed that the neural networks trained on the augmented data achieved high prediction accuracy in terms of MAPE metric, namely 3.97%. This approach demonstrates the effectiveness of using data augmentation for model training under limited input data.