Hybrid modelling algorithm to improve engine predictive maintenance

Technical university ETH Zürich and WinGD (Winterthur Gas & Diesel Ltd) have announced the joint development of a new algorithm derived from a physical and data driven engine model which they claim can be used as a basis for improved predictive maintenance on marine two-stroke engines.

The system will be tested and validated further before it is then applied in future versions of the WinGD Integrated Digital Expert (WiDE) engine data analytics system, the partners said.

WiDE diagnostics are currently being piloted on WinGD engines in operation, and were made available for all new WinGD engines ordered as of the beginning of 2018. The platform uses machine learning and modelling based on performance benchmarking and sensor data to detect and predict failures.

The new system developed under the project with ETH Zürich builds on that technology by combining the data-driven engine modelling approach with physical modelling.

While data-driven models use condition monitoring or rules derived from experiments, physical modelling relies on complex simulations, which in this case will see a thermodynamic model applied that can deliver diagnostic data in milliseconds.

ETH Zürich is also collaborating with the US space agency NASA to validate the performance of the algorithm on turbofan engines, and notes that its use has demonstrated improved performance for the prediction of the remaining useful lifetime of equipment compared to pure data-driven approaches.  

“Tests with WinGD as well as with NASA’s data sets showed that we are more accurate on detecting failures than standard approaches,” said ETH Zürich Researcher Manuel Arias Chao. “Furthermore, we can differentiate between different types of failures.”

The joint research project has been funded by Innosuisse, the Swiss Innovation Agency.