Abstract: ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm and an enhanced bidirectional long short-term memory (BiLSTM) network is proposed. First，permutation entropy is introduced into the CEEMDAN algorithm to improve its noise reduction capability for gearbox vibration signals. Then，the logistic chaotic mapping and adaptive inertia weight Archimedes algorithm are employed to optimize the hyperparameters of the BiLSTM network，resulting in an improved BiLSTM network. This enhanced network is used to predict the denoised vibration signals. Finally，a health index is defined to analyze the prediction results. Simulation results show that the health index of a wind turbine gearbox is approximately 1 under normal operating conditions, about 0.7 under defective conditions，around 0.85 under wear conditions, and approximately 0.6 under crack conditions. These findings demonstrate that the proposed method effectively monitors the state of wind turbine gearboxes and facilitates the timely detection of potential faults.