Valve clearance is critical for diesel engine reliability, where abnormal valve clearance faults (AVCF) severely compromise performance. Traditional vibration-based diagnosis methods predominantly rely on data-driven approaches, which suffer from dependency on extensive training datasets, inadequate real-time capabilities, and noise susceptibility. To overcome these constraints, this study proposes a real-time health monitoring method based on a phenomenological vibration model using a Dual-Rate Extended Kalman Filter (DREKF). Specifically, a phenomenological vibration model of the cylinder head is established characterize valve impact signatures mathematically. Building upon this foundation, a nonlinear state-space representation incorporating three state variables is developed, providing the theoretical framework for Kalman filter design. This systematic modeling approach enables the subsequent design of a Dual-Rate Extended Kalman Filter for optimal state estimation under uncertain noise conditions. The DREKF extracts estimated derivatives as discriminative features that simultaneously suppress noise and amplify impact signatures. Simulation results demonstrate that the proposed DREKF maintains high estimation accuracy comparable to the traditional extended Kalman filter, while effectively tracking time-varying parameters. Furthermore, experimental validation on a marine diesel engine demonstrates accurate vibration tracking across operating conditions and reliable AVCF detection with estimated derivatives exceeding thresholds by more than 40% for all fault scenarios. This model-based approach offers a robust real-time solution, overcoming data-driven limitations in engine health monitoring.
Shi et al. (Tue,) studied this question.