A forecast-assisted approach to remaining useful life prediction: a predictive maintenance case study in hybrid Al/CFRP stack drilling

Predictive maintenance is a critical task in modern industries, particularly in high-value sectors such as aerospace, where the production of non-compliant parts can result in substantial financial losses. In aircraft manufacturing, the drilling of hybrid multi-material stacks is a key operation in the assembly process. Ensuring the optimal use of drilling tools is crucial for balancing the quality of the drilled holes with cost efficiency, with Remaining Useful Life (RUL) representing one of the key elements in achieving this balance. In this paper, a novel hybrid statistical-deep learning model is proposed to address the challenge of predictive maintenance in tool wear forecasting for multi-material stack drilling. Specifically, while prior studies on Tool Condition Monitoring (TCM) have focussed on estimating the current tool wear from sensor signals using deep learning, this work advances towards prediction by enabling real-time, multi–step-ahead forecasting of tool wear through a hybrid statistical–deep learning framework. This hybrid approach, explicitly designed to avoid data leakage, enables the prediction of the number of holes that can be drilled before tool replacement is required, facilitating timely planning for maintenance operations. Additionally, a new evaluation metric is discussed to assess the effectiveness of forecasts for the predictive maintenance task.