Cyclostationary Vibration Analysis and Advanced Signal Pre-Processing for Bearing Fault Detection in Electric Drive Units

Rolling bearings are critical components for the reliable operation of rotating machinery. As raceway damage is one of the most common failure modes, continuous condition monitoring is essential for its early detection. This enables a predictive maintenance strategy, ensuring cost-effective and optimized repairs. To enhance the model's fidelity to real-world conditions, a complete automotive electric drive unit, including its various excitations is analyzed using a multibody simulation. Since the cyclostationary nature of raceway faults prevents unambiguous detection using conventional frequency analysis, the envelope spectrum is the established diagnostic standard. However, its effectiveness is severely limited in integrated electric drives. Due to the high-energy deterministic background noise generated by the transmission, the weak stochastic fault signatures are frequently masked, preventing reliable detection even in the envelope domain. To overcome this limitation, this paper evaluates advanced signal preprocessing techniques designed to enhance the signal-to-noise ratio prior to demodulation. Among the investigated methods, cepstrum pre-whitening (CPW) has proven particularly effective in suppressing deterministic excitations, thereby removing superimposed gear mesh excitations from the signal. This significantly improves early fault detection capability. When combined with minimum entropy deconvolution (MED), the impulsiveness of the signal is further optimized, allowing the damage characteristics to be highlighted even more distinctly.

Keywords: Raceway damage, envelope spectrum, cyclostationary signal, cepstrum pre-whitening, minimum entropy deconvolution.