Machine learning for wear forecasting of naval assets for condition-based maintenance applications

Economic sustainability of running Naval Propulsion Plants is a key element to cope with, and maintenance costs represent a large slice of total operational expenses: last decades' approaches, based on a repairing-replacing methodology, are being trespassed by more effective approaches, relying on effective continuous monitoring of assets wear. In this framework, Condition-Based Maintenance (CBM) is becoming key thanks to the enhancing capabilities of monitoring the propulsion equipment by exploiting heterogeneous sensors: this enables diagnosis and prognosis of the propulsion system's components and of their potential future failures. The success of CBM is based on the capability of developing effective predictive models, for which purpose state-of-the-art Machine Learning (ML) methods must be developed. Nevertheless, testing the performance of ML models for CBM purposes is not straightforward, mostly due to the lack of publicly available datasets for benchmarking purposes: thus, we present in this work a new dataset, that will be freely distributed to the community working on ML models for CBM, generated from an accurate simulator of a naval vessel Gas Turbine propulsion plant. The latter is then used for benchmarking the effectiveness of two state-of-the-art ML techniques in the considered maritime domain.