An improved method of detecting engine misfire by sound quality metrics of radiated sound

Abstract

Existing engine misfire detection techniques require direct contact with hot vibrating engine component(s). Thus, they need costly sensors and regular maintenance. To overcome this limitation, a novel method is proposed to detect cylinder misfire using sound quality metrics of the radiated sound, measured either near the cylinder block or near the exhaust tailpipe. This method was tested on a four-stroke, four-cylinder spark-ignition engine over a wide range of load torques and speeds. Sound signals were measured near the cylinder block and near the tailpipe. Vibration signals were measured on the cylinder block. Sound quality metrics namely, loudness, roughness, and fluctuation strength of engine and tailpipe sound using a support vector machine classifier correctly predicted misfiring with 94% test accuracy each. Whereas engine vibration statistical features predicted at 82% test accuracy, engine/tailpipe sound pressure level predicted at 85% test accuracy. Thus, sound quality metrics of the radiated sound successfully detect engine misfiring, with higher accuracy than conventional time domain features of sound and vibration signals. The method was shown to be speed and torque independent. Unlike existing techniques, the proposed method makes no direct contact with any engine component, is computationally quick and reliable over wider range of speeds and torques, and can be easily applied to any acoustic sensor placed under-hood, or near tailpipe.