Fleet-based approach for tool wear estimation using sequential importance sampling with resampling

In this paper, we study the use of historical data from a fleet (i.e., a group) of equipment to improve condition monitoring and health assessment for each individual equipment (within the fleet) for maintenance purposes. In particular, we propose a fleet-based approach to estimate the tool wear of milling machines at arbitrary operating conditions (OCs) using the Monte Carlo sequential importance sampling with resampling (SIR) algorithm. Unlike traditional data-driven estimation methods which usually rely on historical data available at the same OC, the proposed method can be used when there is a lack (or insufficiency) of historical data at the target OC by leveraging on the availability of historical data at others. First, we introduce a similarity measure based on the well-known extended Taylor\´s tool life equation to identify other OCs (where historical data are available) that are most "similar" to the target one. We then combine historical data from these OCs to construct a suitable model for the SIR to estimate the tool wear at the OC of interest. Our experimental results on a simulated dataset show that the proposed fleet-based approach with SIR can significantly improve tool wear estimation accuracy.