Abstract
The system we developed enables to detect knocking by sound signals even under a circumstance, where mechanical noises that continuously changing with driving conditions exist. The effective feature values are calculated with the spectrum analysis from sound data captured near the TDC with a microphone. Threshold values for detecting knocking are decided by online update anomaly detection techniques. Thereby realizing knocking detects system, which is resistant to change trend of background noises. The system is effective for detection and objective rating of knocking. It is useful for the monitoring of knocking on automatic operation.
