Data driven tool wear and remaining useful tool life estimation

Accurately predicting the Remaining Useful Life (RUL) of cutting tools is critical for enhancing productivity, reducing unexpected downtime, and minimizing manufacturing costs. As tool wear directly affects product quality, process stability and overall efficiency, reliable prediction methods are key to achieving smart and sustainable manufacturing. With recent advances in machine learning, data-driven approaches have proven highly effective in forecasting key machining responses such as tool wear, RUL and surface quality. This study leverages a Bayesian Neural Network (BNN) to predict the RUL of milling tools. Unlike conventional machine learning models, BNNs combine artificial neural networks (ANNs) with Bayesian regularization, enhancing model robustness and reliability. This is especially valuable in manufacturing environments where data can be limited, noisy or variable across operating conditions. The proposed model is trained on an open-access dataset and rigorously validated through in-house milling experiments conducted on Inconel 718 and SS 304 under varying machining parameters. The separation of training and testing data sources prevents overfitting to a specific dataset and enables the model to learn generalized patterns, ensuring its estimation, adaptability and effectiveness across diverse machines, tools and real-world manufacturing scenarios. The predictive model achieved flank wear prediction accuracies between 83% and 91% and RUL prediction accuracies between 80% and 93%, highlighting its reliability in estimating the progressive wear of the cutting tool. In addition to prediction accuracy, the Bayesian framework provides a natural quantification of model uncertainty, which is essential for decision-making in manufacturing environments, even though uncertainty metrics are not explicitly reported in this study. By forecasting the tool's remaining useful life through flank wear behavior, the model enables proactive tool replacement and maintenance scheduling, thereby avoiding failures and reducing idle time. These findings highlight the BNN model's effectiveness in both predictability and transferability, making it a promising solution for real-world tool life estimation in manufacturing applications.