<p>Printed circuit boards (PCBs) are widely used across various industrial sectors. The ongoing trend toward technological integration and product miniaturization has increased the demand for precision micro-drilling of these circuits. Since PCBs typically contain a large number of holes, it is essential to consistently monitor tool conditions to ensure a faultless final product. Unlike conventional tools, which are more likely to wear gradually, micro-drills are prone to sudden breakage due to size-related effects such as micro-scale inhomogeneity of the work material and low stiffness resulting from their relatively large aspect ratio. Moreover, the complex three-dimensional geometry of micro-drills makes it difficult to establish an algebraic breakage model. These characteristics justify the use of vision sensors, which enable direct and accurate assessment of tool states for predicting tool breakage. To the best of the author’s knowledge, this is the first study to employ cumulative direct vision-based micro-tool deflection as the primary feature for AI-driven prediction of tool breakage during PCB micro-drilling. The results showed that tool deflection occurred frequently and that its instability progressively increased over the tool’s operational life. However, the increase in tool deflection did not follow any specific pattern, making deterministic prediction of tool breakage difficult. To address this issue, accumulated deflection data were converted into image-based datasets and used as inputs to artificial neural networks (ANNs) for tool-breakage prediction. The results showed that the proposed deflection-based model successfully classified tool states into normal and abnormal conditions, achieving an accuracy of up to 97%, significantly outperforming a model based on cutting-force data. This demonstrates the potential of the method for precise and reliable prediction of micro-drill failure.</p>