The structure of the drying section in papermaking process is complex and too compacted to install sensors. In order to monitor the parameters in dynamic and manage the process practically with virtual simulations instead of physical experiments, a digital twin-based process parameter visualization model is constructed in this study. Regarding to the possible missing data in the modeling framework, it is proposed to combine industrial data, and knowledge of mechanism with intelligent algorithms to fill in the missing parameters. Upon which, a digital twin-based data visualization model is established using CADSIM Plus simulation software. Both of the knowledge -based mechanism solution model and the random forest-based parametric prediction model perform well, and the predicted parameters can support the digital twin visualization model in CADSIM Plus. Visual modeling of surface condenser in the paper drying section was realized for example, and results show that the model is capable of monitoring the dynamic changes of parameters in real time, so as to support the optimization and decision making of papermaking process such as formation, drying, et al.
The steam turbine is a rotating device subject to axial and radial shaft shifts that can induce vibrations during operation. Tools such as monitoring systems and proximity probe sensors are essential to monitoring these vibrations. High vibrations affect the machine’s performance, increasing the risk of malfunctions and reducing its lifespan, and also pose risks to operational and maintenance personnel. The intensified vibrations in the bearing pedestals signify the underlying issues with the machine’s normal operation. Consequently, problems such as rotor imbalance, coupling misalignment, mechanical looseness, material failure, and bent shaft may be caused. In the current study, the latest field-proven automatic diagnostic of rotary equipment (ADRE 408) data acquisition system is installed by Bentley Nevada to investigate the root cause of high vibration. This advanced diagnostic system facilitates a comprehensive assessment, enabling us to effectively identify and address underlying problems. Hence, the current research includes a thorough diagnosis of the underlying problems to attenuate the risks of high vibrations in the steam turbine, coupled with strategic maintenance planning and corrective actions.
A new combat strategy that enables coordinated operations of gliding aircraft clusters for multi-target strikes imposes higher demands on the coordination, real-time responsiveness, and strike accuracy of gliding aircraft clusters. Due to the high speed and large inertia characteristics of gliding aircraft, traditional trajectory planning methods often face challenges such as long computation times and difficulty in responding to dynamic environments in real-time when dealing with large-scale gliding aircraft clusters. This paper proposes a distributed cooperative trajectory planning method for multi-target strikes by gliding aircraft clusters to address this issue. By introducing a multi-objective distributed real-time trajectory planning approach based on Multi-Agent Deep Deterministic Policy Gradients (MADDPG), the gliding aircraft execute distributed cooperative trajectory planning based on the trained model. Due to its robust real-time performance, the gliding aircraft do not need to recalculate trajectories for different initial positions of the cluster. Simulation results show that the average error between the gliding aircraft cluster and the target point is 2.1 km, with a minimum error of 0.06 km and a hit rate of 96.6%, verifying the significant advantages of this method in real-time planning capability and strike accuracy.
The diagnosis of paper breakage faults during the papermaking process is of great significance for improving product quality and maintaining stability in the production process. This paper develops a cross-condition transfer learning fault diagnosis model. This study proposes a fault diagnosis method based on transfer learning to address the issue of single-condition diagnostic models performing poorly when applied to different conditions..This method uses both parameter transfer and feature transfer to diagnose faults across different conditions. At the same time, in response to the issue of insufficient small sample operating data, we introduce federated learning technology to explore the impact of model compression rates on the diagnostic accuracy of the federated global model during the federated model training process. The results indicate that compared to single operating condition models, fault diagnosis performance based on transfer learning across different operating conditions has improved. The diagnostic model based on feature transfer performs even better, achieving accuracy rates of 98.31%, 94.64%, and 96.43% under different transfer tasks, allowing for accurate classification of the majority of samples. Additionally, the federated learning method provides an effective solution for fault diagnosis in small sample operating conditions, and an appropriate model compression rate can ensure diagnostic accuracy while protecting data privacy.
