An adaptive RUL prediction approach for cutting tools incorporated with interpretability and uncertainty

Reliable prediction of tool Remaining Useful Life (RUL) is essential for ensuring machining quality and promoting sustainability, serving as a key component of high-performance machining. Traditional approaches face significant challenges in addressing uncertainties and ensuring interpretability during machining processes. To predict the RUL of tools in scenarios with small, unlabeled datasets and to quantify uncertainty in degradation models, this paper presents an adaptive RUL prediction method based on multi-source data fusion and Bayesian inference. Specifically, an adaptive Bayesian iterative updating model, grounded in the degradation process of cutting tools is proposed. This model operates without the need for extensive labeled samples or offline training, while incorporating both interpretability and uncertainty. Furthermore, Bayesian inference, combined with delayed rejection and adaptive Metropolis-Hasting strategies, is employed to update uncertainty parameters in the degeneration model. This enables the degradation model adaptively approximate the actual wear tread in real time through continuous observation. The proposed method is validated using experimental tool wear data, demonstrating reductions in average RMSE of 27.00 % and 11.60 % compared to other approaches. Notably, it does not rely on large labeled historical datasets and effectively mitigates the impact of uncertainties, offering a novel approach to RUL prediction in real industrial applications.