Özet:

Detecting periodic impulse signal (PIS) is the core of bearing fault diagnosis. Earlier fault detected, earlier maintenance actions can be implemented. On the other hand, remaining useful life (RUL) prediction provides important information when the maintenance should be conducted. However, good degradation features are the prerequisite for effective RUL prediction. Therefore, this paper mainly concerns earlier fault detection and degradation feature extraction for bearing. Maximum correlated kurtosis deconvolution (MCKD) can enhance PIS produced by bearing fault. Whereas, it only achieve good effect when bearing has severe fault. On the contrary, PIS produced by bearing weak fault is always masked by heavy noise and cannot be enhanced by MCKD. In order to resolve this problem, a revised empirical mode decomposition (EMD) algorithm is used to denoise bearing fault signal before MCKD processing. In revised EMD algorithm, a new recovering algorithm is used to resolve mode mixing problem existed in traditional EMD and it is superior to ensemble EMD. For degradation analysis, correlated kurtosis (CK) value is used as degradation indicator to reflect health condition of bearing. Except of theory analysis, simulated bearing fault data, injected bearing fault data, real bearing fault data and bearing degradation data are used to verify the proposed method. Simulated bearing fault data is used to explain the existed problems. Then, injected bearing fault data and real bearing fault data are used to demonstrate the effectiveness of proposed method for fault diagnosis. Finally, bearing degradation data is used to verify the degradation feature CK extracted based on proposed method. All these case studies show the effectiveness of proposed fault diagnosis and degradation tracking method.

Anahtar Kelimeler: