Electric solar wind sail (E-sail) is an innovative propulsion system, which can generate low and continuous thrust without any propellant by reflecting ions from the solar wind. Because the propulsive acceleration of E-sail is low and continuous, its trajectory is usually characterized by long flight time. In order to explore more flight plans, rapid trajectory design is useful for preliminary mission analysis and design. This paper presents a method for rapid generation of minimum-time, three-dimensional trajectories for an E-sail using the finite Fourier series shapebased method. Unlike the more conventional electric thruster, the thrust vector of an E-sail is constrained. In order to consider the characteristics of the E-sail thrust, inequality constraints of thrust acceleration are investigated based on a recent thrust model. The numerical simulation results show that the proposed method is useful to quickly design the flight trajectory of the E-sail by considering the actual characteristics of the thrust vector

The privacy and security of industrial data are receiving increasing attention. Digital twins can establish realistic cyber information models for industrial equipment. However, it is challenging to transfer diagnostic knowledge from low-cost equipment, such as normal ball bearings, to high-cost equipment, such as high-speed aerospace roller bearings, during the diagnostic process. We have developed a paradigm for establishing digital twin diagnostic models across devices with source model and a few target samples. We consider the privacy and security of source data in the transfer learning process, as well as the problem of excessive domain drift. First, we use fault data from low-cost normal ball bearings to generate the corresponding fault diagnosis model. During the process of transferring diagnostic knowledge to aerospace roller bearings, we ensure that the source data remains inaccessible. To address the issue of excessive domain drift, we introduce a small number of labeled target samples to establish a supervised link between the model and the target domain. Information maximization is employed to enhance the extraction of knowledge from these labeled samples. Second, we compute similarity and dissimilarity matrices among target domain samples. Contrastive learning is utilized to strengthen the connections within similar target samples and to distinguish dissimilar ones. Subsequently, we leverage information maximization to make reasonable inferences within the target domain. However, since information maximization focuses only on the overall uniformity of the domain, it inherently weakens the support set’s guidance to the query set. To overcome this limitation, we propose a mixed uniformity loss function to further improve the reliability of uniformity predictions for unlabeled target samples. We validate our approach by transferring diagnostic knowledge from a deep groove ball bearing dataset to two specialized aerospace roller bearing datasets. Comparative experiments demonstrate that the proposed CDT method outperforms existing approaches.