<p>Tool wear condition monitoring is critical for ensuring machining accuracy and maintaining production efficiency. However, collecting sufficient labelled data from diverse machine tools poses significant challenges in real-world manufacturing environments. This limitation makes the development of methods that are capable of generalizing from known machine tools to unseen tools crucial. A feature augmentation-based adversarial domain generalization method that enables the effective generalization of tool wear monitoring models to unknown machine tools without accessing the target-domain data during training is proposed in this paper. Initially, the proposed method extracts domain-invariant features through adversarial learning and covariance-based domain alignment, which ensures robustness against domain shifts among different machine tools. These features are then augmented using a feature interpolation strategy, which enhances the diversity of features while preserving domain invariance. Furthermore, a large margin loss is employed to improve the discriminability of the generated labels, which further boosts the generalization performance of the method. Extensive experiments conducted across four datasets demonstrate that the proposed method outperforms the existing approaches in terms of generalization ability, thus offering a practical and effective solution for monitoring tool wear conditions on unseen machine tools.</p>