Mar-RUL: A remaining useful life prediction approach for fault prognostics of marine machinery

Abstract

Although the maritime industry has the potential to lead smart maintenance methodologies, current maintenance routines within the sector focus on either reactive or preventive maintenance approaches. These approaches are increasingly conservative and often prompted by an increase of large costs or unnecessary maintenance actions. Attempts are being made, however, to optimise these costs and actions through the application of novel practices, such as the employment of prognostic-based maintenance, which is an approach that aims to minimise risk while maximising the lifespan of a system. In this respect, Mar-RUL is introduced to support Operations & Maintenance (O&M) decision making and address some of the main challenges that the sector is currently experiencing, including the lack of fault data analysis and the formalisation of deep learning technologies for the implementation of the Remaining Useful Life (RUL) prediction. In response, a degradation data simulation module is developed in tandem with an ensemble model comprised of three distinct deep learning architectures: Markov-Convolutional Neural Network (CNN), 1D-CNN, and Long Short-Term Memory (LSTM) neural network. To evaluate the performance of Mar-RUL, a case study on the turbocharger of a diesel generator of a tanker ship is presented. The case study results demonstrated that the application of time series imaging and ensemble methods can provide promising outcomes for the enhancement of RUL prediction.