With the advancement of modern industrial technology, complex mechanical systems have found extensive applications across various industries. Gears, integral components of these systems, play a crucial role in determining the stability and safety of the entire system. Wear and aging of system components during prolonged operations might lead to performance degradation or system failures. Historically, numerous methods for vibration signal analysis and gear wear detection have been proposed. However, these methods often exhibit limitations when applied to intricate systems, such as reliance on empirical rules and suboptimal handling of nonlinear vibration signals. In light of these challenges, the vibration genesis mechanism in complex mechanical systems has been deeply investigated. A "Gear Health Factor" has been introduced, and a wear prediction model for gears, incorporating Bidirectional Long Short-Term Memory (Bi-LSTM) networks and attention mechanisms, has been developed. This research offers fresh perspectives and methods for the health management of complex mechanical systems and holds significant practical implications.