Vector diversity entropy as a nonlinear dynamic feature for continuous health monitoring of rolling bearing

Entropy-based methods have gained widespread adoption in fault diagnosis owing to their superior classification capabilities, particularly diversity entropy (DE) which demonstrates exceptional consistency, robust noise immunity, and computational efficiency. However, DE exhibits limitations in characterizing continuous degradation trends due to its poor monotonicity on the continuous health monitoring. To address this issue, this article proposed a novel method called vector DE (VDE) for the continuous health monitoring of rolling bearing. First, in the proposed VDE, a magnitude entropy is proposed to converts the probabilistic state information of the global amplitude information into entropy value, which could improve the monotonicity while retaining the original frequency sensitivity. Second, the characteristic properties and the root cause of the poor monotonicity have been explored: the original DE only calculates the angles of orbits in the phase space, ignoring that the global amplitude reduces amplitude sensitivity which leads to poor monotonicity. Lastly, the effectiveness of the proposed VDE has been verified in the continuous health monitoring of rolling bearing through the experimental run-to-failure bearing signals.