The on-line detection of engine misfire at low speed using multiple feature fusion with fuzzy pattern recognition

This paper proposes a technique for the online detection of incipient engine misfire based on multiple feature fusion and fuzzy pattern recognition. The technique requires the measurement of instantaneous angular velocity signals. By processing the engine dynamics model equation in the angular frequency domain, four dimensionless features for misfire detection are defined, along with fast feature-extracting algorithms. By directly analysing the waveforms of the angular velocity and the angular acceleration, six other dimensionless features are extracted. Via fuzzy pattern recognition, all the features are associated together as a fuzzy vector. This vector identifies whether the engine is healthy or faulty and then locates the position of a misfiring cylinder or cylinders if necessary. The experimental work conducted on a production engine operating at low speeds confirms that such a technique is able to work with the redundant and complementary information of all the features and that it leads to improved diagnostic reliability. It is fully expected that this technique will be simple to implement and will provide a useful practical tool for the online monitoring and realtime diagnosis of engine misfire in individual cylinders.