This work focuses on reducing the experimental need for creating a reliable tool life model for a data set of 46 tool data points with its resulting tool life for a single-tooth side milling application in medium carbon steel, C45 E.

Based on the data set, 615 180 unique tool life models are created using Colding’s equation.

This is achieved by creating models using different unique subsets of the complete data set where the cardinality is varied from 7 to 43.

The paper shows that the improvement from adding more data points to the modelling are neglectable after 34 data points are included in the modelling if a maximum absolute model error ≤ 9% is sought.

Furthermore, it is shown that the prediction error increases when extrapolating outside the range of equivalent chip thickness and cutting speed used for the modelling work compared to an interpolative error within the range.

By carefully planning the experimental set-up by maximising the cutting speed and feed range decreases the risk of creating a non-relevant model where the prediction error increases when located outside the modelling range.