Tool Condition Monitoring Under Different Operating Conditions Using ML with Scarce Data: A Review

Tool condition monitoring (TCM) is essential for ensuring good quality products, machining reliability, efficiency, and sustainability. Machine learning (ML), including deep learning (DL), has been extensively used in the literature to address different TCM tasks such as tool state recognition, tool wear prediction, and remaining useful life prediction. Nevertheless, the adoption of existing methods in real-world manufacturing is still hurdled by different practical challenges. This paper focuses on two key challenges facing ML-based TCM: 1) Variability of operating conditions; 2) Data scarcity. The first challenge arises from the variations in data distributions (domain shift) caused under cross operating conditions, which leads a trained ML model to generalize poorly on data from operating conditions unseen during training. The second challenge stems from the impracticality of collecting sufficient data on real-world shop floors, especially when labeled data is needed. Addressing simultaneously both challenges inherently leads to problem and evaluation settings different from those concerning only one single challenge without having the constraints imposed by the other (e.g., addressing the domain shift assuming the availability of sufficient data, or addressing the data scarcity under single-operating conditions). This paper presents a review focused on TCM considering both different operating conditions and data scarcity scenarios. The works reviewed are based on adaptation- and/or generalization oriented solutions leveraging prior knowledge across various related-yet-distinct learning settings, namely transfer learning, domain adaptation, domain generalization, meta-learning, and hybrid settings. Future research opportunities are also presented. This review can serve as a guide for both researchers and practitioners, presenting state-of-the-art practices and concrete insights to tackle and advance the challenging industry application of TCM.