Cloud-Based Data-Driven Predictive Maintenance of eLOP using Digital Twin Technology

This thesis presents a comprehensive study on the development of a predictive maintenance model for electric lubrication oil pumps (eLOPs) that utilizes cloud technologies to provide real-time monitoring, diagnosis, and maintenance. The proposed solution involves a two-step architecture that integrates an unsupervised hybrid anomaly detection model with supervised Remaining Useful Lifetime (RUL) prediction models. The anomaly detection component leverages a two-layer recurrent neural network with a One Class Deep SVDD objective function and human expert-generated key parameters limits. The RUL prediction models are trained on continuous RUL measures obtained by transforming the binary indicators of the pumps’ health state. The integration of the predictive maintenance model with the cloud architecture ensures a continuous stream of incoming data to improve model predictions and enhance the effectiveness of the solution. The study also highlights the digital twin model as a key component and describes the main challenges encountered. The solution has the potential to greatly improve the reliability of eLOPs in vehicles, and ensure that the maintenance process is performed efficiently and effectively and opens up the potential for future works in the field.