Transformer-based Real-time Cylinder Failure Detection in a Gasoline Direct Injection Engine

Gasoline Direct Injection (GDI) engines are widely employed in modern vehicles due to their precise airfuel mixture control, which leads to higher power density.However, unexpected cylinder faults can hamper engine performance and pose significant risks if not detected promptly.This study introduces a transformerbased method for real-time cylinder failure detection, leveraging multivariate sensor data from practical GDI engines.We focus on signals collected from the ignition coil, camshaft and camshaft position sensors.By employing self-attention mechanisms, the proposed model accurately identifies cylinder anomalies.Experimental results on a bench-tested GDI engine reveal that this method can pinpoint anomalies within 0.06 seconds of the failure point, when trained with signals no longer than one second.These findings highlight the potential of Transformer architectures to advance real-time monitoring and ensure more reliable GDI engine operation. GDI . GDI . GDI . , , . GDI , 1 0.06 . GDI .