Efficient Powertrain Control Using Signal Decomposition: Utilizing Signal Decomposition Constituents for Knock Detection in Internal Combustion Engines

Knocking combustions in an Internal Combustion Engine (ICE) are engine damaging combustions, and reliable detection of each knocking event is very critical. Engines usually rely on piezo-electric knock sensors to monitor structure-borne noise, which outputs a complex, continuous time series signal. Typically, knock combustions have an additional noise component along with the regular combustion signal, but differentiation of knocking vs non knocking signal (signal to noise ratio) based on visual inspection of this signal alone is challenging and requires computationally intense signal processing such as Fast Fourier Transforms (FFT) or Wavelet transforms followed by manual calibration [1]. In this paper, we propose an alternative to replace traditional knock detection with more reliable time-domain alternative signal decomposition technique. Here we decompose the raw sensor signal into seasonality, trend, and residual, and use the residual component as it is seen to retain abnormalities in the signal during knocking combustion. Further, based on the amplitude of the residual, we can easily classify the combustions as low, medium, high or very high knocking events thus providing a precise and reliable detection.