Image-based machine learning model for tool wear estimation in milling Inconel 718

Tool wear is a critical factor in machining processes, particularly when dealing with difficult-to-machine materials like Inconel 718, which are susceptible to rapid tool degradation due to their inherent toughness and low thermal conductivity. Traditional methods of predicting tool wear, which rely on empirical models or extensive sensor data, often suffer from limitations in accuracy and adaptability. To overcome these challenges, we propose a novel approach utilizing a ResNet50-based model that leverages high-resolution images of worn tools to directly predict wear levels. The study involved experiments using 35 different cutting tools, each coated with various materials, and subjected to varying combinations of cutting speeds and feeds. A comprehensive dataset of images from all the tools was collected and used to train the model. The ResNet50 algorithm, renowned for its deep residual learning capabilities, was employed to automatically extract relevant features from the images and establish a predictive relationship with actual tool wear measurements. The performance of the ResNet50 model was validated against other state-of-the-art deep learning approaches, demonstrating superior accuracy in predicting tool wear under different cutting conditions. The proposed model consistently provided tool wear predictions that closely matched experimental measurements and generalized effectively across different cutting tools and edges, showcasing both accuracy and robustness. This approach not only enhances predictive accuracy but also offers a scalable solution for real-time tool condition monitoring in industrial settings.