Tool wear prediction for coated cutting inserts during milling of Ti-6Al-4V using a grey box model

Titanium alloys are widely used in aerospace, medical, and automotive industries due to their high strength and corrosion resistance, as well as low density. However, the machining of titanium presents significant challenges, including high mechanical loads, excessive heat generation, and, thus, increased tool wear. This wear leads to a poor surface quality, dimensional inaccuracies, and increased production costs due to an unplanned downtime and frequent tool replacements. Traditional monitoring methods for predicting tool wear, such as periodic inspections and simple threshold analyses, are often inefficient and time consuming. Artificial Intelligence has emerged as a powerful tool for identifying complex wear patterns, enabling accurate predictions of tool life and optimal replacement timing. However, Artificial Intelligence models are often used as black box models and therefore lack interpretability. In contrast, white box models, which are based only on physical principles, often fail to capture complex wear dynamics. Grey box models combine both approaches, integrating domain knowledge with machine learning algorithms to enhance prediction accuracy and reliability. A particular strength of the present study is that the dataset covers the complete wear curve, including the progressive wear phase, and not only up to the commonly used wear limit of 0.3 mm. This allows the model to be trained and evaluated across all stages of tool wear. In this article, a Long Short-Term Memory Neural Network with domain knowledge of the physical behavior of tool wear as a cubic function and a loss function that enforces monotonic tool wear behavior is presented. The performance of the model achieved a mean absolute error in the tool wear prediction of 0.0447 mm, which is good. However, it could not be compared to state-of-the-art approaches for predicting tool wear because the entire data curve is not usually determined. • Use of an exceptionally large milling tool wear data set with 18 different parameter combinations with varied tool diameters, feeds per tooth, and spindle speeds. • Industrially relevant application, as milling was carried out with coolant. • Tool wear prediction model considering tool wear domain knowledge. • A mean absolute error in the tool wear prediction of only 0.0447 mm.