Rolling bearings are critical components of marine shafting power transmission systems, and accurate prediction of their vibration signal trends is essential for predictive maintenance. To address the limited adaptability of conventional time-series forecasting models under varying operating conditions and their insufficient ability to capture strong noise and abrupt changes, this study proposes a vibration signal prediction method that integrates particle swarm optimization (PSO) with an improved Informer model. PSO is used to adaptively optimize key Informer hyperparameters for different operating conditions, while a rolling time-window mechanism is introduced to enhance the capture of abrupt signal variations. In addition, a mixture of sparse attention (MoSA) encoder with a collaborative dense-head/sparse-head structure is designed to balance global temporal dependency modeling and local fault feature extraction. Experimental results on the Case Western Reserve University (CWRU) bearing fault dataset show that the proposed model outperforms Long Short-Term Memory (LSTM), Transformer, Informer, iTransformer, and Flowformer in terms of Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root Mean Squared Erro (RMSE). The model achieves an MSE of 0.2015, which is 25.5% lower than that of the second-best iTransformer model. It also demonstrates robust performance under four different bearing operating states, confirming its adaptability to complex operating conditions. The proposed method provides a promising technical route for the predictive maintenance of rolling bearings in marine shafting systems.

This paper presents a genetic algorithm (GA) tuned Mamdani type fuzzy logic control (FLC) framework for trajectory tracking of a quadrotor unmanned aerial vehicle (UAV) using a nonlinear rigid body model. The proposed architecture adopts a cascaded structure in which an outer loop position controller generates attitude and thrust references $(\phi_{\mathrm{ref}},\theta_{\mathrm{ref}},T_{\mathrm{ref}})$, while an inner loop attitude controller generates body torques $(\tau_\phi,\tau_\theta,\tau_\psi)$. Both loops employ a shared Mamdani fuzzy inference system with normalized inputs (tracking error and error-rate) and a normalized control output. The GA automatically tunes scaling gains $(K_e,K_d,K_u)$ across all axes to minimize a robust objective that averages tracking error, control effort, and constraint violations over multiple scenarios with mass uncertainty and wind disturbances. Simulation results on a three dimensional figure eight trajectory indicate that GA tuning can reduce position and attitude errors while respecting actuator saturation and tilt safety limits, demonstrating a practical route to performance enhancement without requiring a high fidelity aerodynamic model. The methodology leverages the interpretability of fuzzy rules and the global search capabilities of evolutionary optimization within a UAV modeling framework consistent with established quadrotor dynamics literature.