Cabin noise fault detection using deviation energy method in cross time and frequency domain

This paper proposes a new method to detect annoying sounds in-car cabins. Based on the fact that annoying sounds are usually triggered off by harsh road conditions and hence occur occasionally and instantly. The signals for such sounds would have a broadband spectrum in the frequency domain. From this understanding, a detection scheme is developed to find the instants of the abnormal sounds in the lengthy signal. This will be is a primary step in diagnosing the source of the sound. In particular, a new method, referenced as deviation energy (DE) method is developed to highlight the small changes in the high frequency range (above 500Hz) in the abnormal sounds. In addition, the detection parameters including average spectrum and detection thresholds are adaptive to different data frames corresponding to different driving conditions. This makes the detection scheme more generic and reliable. Case study shows that this DE-based detection scheme produces more accurate and reliable results, benchmarked with commonly used statistical measures such as RMS, Kurtosis and Euclidean distance.