Unmanned agricultural machinery is often ever-increasingly demanded for intelligent farming in recent years, due mainly to the labor shortage and the high operational costs. However, several challenges are still remained on the technical solutions, including the dependence on the regular-shaped fields, separation between path and velocity planning, and low control accuracy in complex environments. In this study, an integrated system of the trajectory planning and tracking control was developed to enhance the operational performance of the unmanned agricultural machinery under the field conditions. A cooperative framework was optimized to integrate the trajectory planning with the tracking control. In the trajectory planning, the Fields2Cover library was employed to generate the full-coverage paths for the irregular fields, the multiple objective functions, swath sorting, and turning connection. An adaptive velocity switching strategy was introduced using path curvature, in order to avoid the disconnection between path and velocity planning. S-curve velocity profiling was combined with the smooth velocity transitions. In tracking control, an improved Linear Quadratic Regulator (LQR) controller was designed using an Ackermann steering kinematic model. A curvature-adaptive weight matrix adjustment was dynamically optimized the controller's parameters, according to the real-time path curvature. Additionally, a target steering angle preview feedforward was implemented to reduce the overshoot for the tracking precision during curves. Simulation and real-world experiments demonstrated the effectiveness of the improved model. In the simulation ablation, the preview LQR was reduced the mean and maximum lateral errors by 30.0% and 33.6%, respectively, compared with the conventional LQR, indicating the independent contribution of the preview mechanism to the suppression of the curve overshoot. In the overall performance, the improved LQR was reduced the mean lateral error by 50.0%, 28.6%, 21.1%, and 40.0%, and the maximum lateral error by 49.6%, 24.1%, 4.3%, and 59.8%, respectively, compared with the conventional LQR, preview LQR, pure pursuit, and Stanley. In heading error, the improved LQR was achieved in the mean and maximum of 0.005 and 0.064 rad, respectively, the mean heading error was reduced by 28.6%, 16.7%, 16.7%, and 64.3%, and the maximum heading error by 31.9%, 28.1%, 29.7%, and 77.6%, respectively, compared with the four benchmarks. The velocity errors were also minimized at 0.014 m/s (mean) and 0.036 m/s (maximum). Real-vehicle tests on the uneven grassland were further validated the robustness. Quantitative analysis of the ablation group showed that the preview LQR was reduced the mean and maximum lateral errors by 26.1% and 23.3%, and the mean and maximum heading error by 13.3% and 3.9%, respectively, compared with the conventional LQR. The mechanism was effectively validated to enhance the heading stability. Furthermore, the improved LQR was reduced the mean lateral error by 48.9%, 30.9%, 13.0%, and 56.5%, and the maximum lateral error by 43.9%, 26.9%, 22.8%, and 72.7%, respectively, compared with the conventional LQR, preview LQR, pure pursuit, and Stanley. The mean heading error reached 0.018 rad, which was reduced by 40.0%, 30.8%, 18.2%, and 71.0%, respectively, compared with the four benchmarks. Although the maximum heading error showed a 15.5% increase, compared with the pure pursuit, whereas, it decreased by 9.4%, 5.7% and 49.8%, respectively, compared with the conventional LQR, preview LQR, and Stanley. In velocity tracking, the improved LQR exhibited the smallest fluctuations, indicating the superior stability under challenging field conditions. The path-velocity-control cooperative optimization was integrated to significantly enhance the tracking accuracy, operational smoothness, and field adaptability of the unmanned agricultural machinery. Synergistic planning and adaptive control were effectively realized to overcome the limitations of the conventional method. Therefore, this finding can provide a solid technical foundation to deploy the intelligent agricultural equipment in the irregular terrains. Future research can also explore the reinforcement learning for the parameter tuning, in order to further improve the system adaptability and deployment efficiency.

The remaining useful life prediction of rolling bearing holds significant importance in enhancing the operational reliability and reducing maintenance costs of the entire rotating machinery system. Deep learning techniques have shown promise in remaining useful life (RUL) prediction by leveraging their powerful representation learning capabilities. However, existing deep learning-based approaches still suffer from limitations such as reliance on hand-crafted features and lack of interpretability. Therefore, we propose an improved physics-informed neural networks (PINNs) based on deep separable convolutional network (DSCN) and attention mechanism for the RUL estimation of rolling bearings. Specifically, a deep separable convolutional network is introduced for feature extraction, which directly utilizes multisensor data as inputs and employs separable convolutional building blocks to automatically learn high-level representations. The features are then mapped to RUL using a self-attention mechanism-based physics-informed neural network. The hybrid prediction framework called DSCN-AttnPINN has demonstrated superior performance on the XJTU-SY dataset. The results of the experiments reveal that the DSCNAttnPINN can accurately predict RUL and outperforms certain current data-driven prognostics methods.