On-line diagnosis of a supercharger in a noisy environment

The article describes the engine on-line diagnosis system ODIN, a prototype that performs early `deviation´ recognition on the supercharger of a diesel engine. ODIN is undergoing real size testing on board a French Navy ship. Diagnosing on internal combustion engine is a complex task, for numerous reasons: noisy data, unavailability of a complete and faithful model, multiple faults, small number of experimental data. In ODIN, fault recognition is performed by partially matching the observed qualitative states of the supercharging process with the expected qualitative states corresponding to the various possible faults. This partial matching uses naive knowledge about the system. The partial matching has been implemented using an ATMS (assumption based truth maintenance system) module. This approach has shown accurate, robust to noise and able to detect multiple faults. It can be generalized to other diagnosis problems with noisy data