Accurate glucose prediction plays an important role in glucose management and closed-loop insulin delivery for subjects with diabetes. Due to its powerful data mining capability, neural networks are used to grasp the glucose trends from continuous glucose monitoring (CGM) data. However, this approach requires a large number of individual data and plentiful computing resources, but has relatively poor interpretability. Given that the glucose metabolism mechanism model contains abundant physiological information, fusing the information with neural networks can reduce the demand for data and computing resources, and improve interpretability. In this study, a physics-informed glucose-insulin neural network (PIGNN) model is proposed, of which the structure and loss function are designed based on the glucose-insulin dynamic model. According to the experiments of 22 real subjects and 30 in-silico subjects with type 1 diabetes, the prediction accuracy of this method achieves 0.726 ± 0.126 mmol/L. Compared with models without physical information, the proposed PIGNN shows a significant improvement in glucose prediction for real subjects with a limited sample size (only 48 data samples), resulting in an accuracy improvement of 12.33%. In addition, it is proved that with limited data more physical information can improve the glucose prediction accuracy significantly.

For uncertainty quantification of complex models with high-dimensional, nonlinear, multi-component coupling like digital twins, traditional statistical sampling methods, such as random sampling and Latin hypercube sampling, require a large number of samples, which entails huge computational costs. Therefore, how to construct a small-size sample space has been a hot issue of interest for researchers. To this end, this paper proposes a sequential search-based Latin hypercube sampling scheme to generate efficient and accurate samples for uncertainty quantification. First, the sampling range of the samples is formed by carving the polymorphic uncertainty based on theoretical analysis. Then, the optimal Latin hypercube design is selected using the Latin hypercube sampling method combined with the “space filling” criterion. Finally, the sample selection function is established, and the next most informative sample is optimally selected to obtain the sequential test sample. Compared with the classical sampling method, the generated samples can retain more information on the basis of sparsity. A series of numerical experiments are conducted to demonstrate the superiority of the proposed sequential search-based Latin hypercube sampling scheme, which is a way to provide reliable uncertainty quantification results with small sample sizes.

Full lifecycle high fidelity digital twin is a complex model set contains multiple functions with high dimensions and multiple variables. Quantifying uncertainty for such complex models often encounters time-consuming challenges, as the number of calculated terms increases exponentially with the dimensionality of the input. This paper based on the multi-stage model and high time consumption problem of digital twins, proposed a sparse polynomial chaos expansions method to generate the digital twin dynamic-polymorphic uncertainty surrogate model, striving to strike a balance between the accuracy and time consumption of models used for digital twin uncertainty quantification. Firstly, an analysis and clarification were conducted on the dynamic-polymorphic uncertainty of the full lifetime running digital twins. Secondly, a sparse polynomial chaos expansions model response was developed based on partial least squares technology with the effectively quantified and selected basis polynomials which sorted by significant influence. In the end, the accuracy of the proxy model is evaluated by leave-one-out cross-validation. The effectiveness of this method was verified through examples, and the results showed that it achieved a balance between maintaining model accuracy and complexity.

As a key module of aviation inverter, Insulated Gate Bipolar Transistor (IGBT) plays a decisive role in its safety and reliability. Considering the complex operating conditions of aviation inverter and the fact that IGBT is one of the most vulnerable components for failure, this paper analyzes the failure mechanism and key characteristic parameters of IGBT in aviation inverter. Based on this, an IGBT life prediction method is proposed by combing Long Short-Term Memory (LSTM) network with physical analytical model. The relationship is established for IGBT between its state monitoring data and junction temperature, and the cumulative damage of IGBT is obtained from the physical model, so as to achieve the real-time life prediction of IGBT. Finally, the IGBT accelerated aging experimental dataset provided by the NASA PCoE Center is applied to validate the prediction model. The corresponding results show that the LSTM network combined with the cumulative damage model can effectively predict the lifespan of IGBT, thereby contributing to improving the reliability and reducing the daily maintenance cost of aviation inverters.

The traction system, serving as the power core of urban rail transit trains, plays a crucial role in ensuring the safe operation of the trains. Reliability assessment and lifespan prediction are investigated in order to tackle the difficulties brought about by the traction system’s intricate structure and numerous failure types. The physics of failure model for motor demagnetization and insulated gate bipolar transistor (IGBT) are constructed. The degradation processes for those performance indicators are described by the Wiener process fusing failure mechanism while considering unit-to-unit variability. The Copula function is used to describe the dependent relationship between performance indicators. As for off-line parameter estimation, the Bayesian Markov chain Monte Carlo method estimates unknown parameters. As for online remaining useful life prediction, the algorithm combining Bayesian and expectation-maximization is implemented to update unknown parameters. The proposed model and algorithm are validated by the degradation data of the traction system. The results indicate that the reliability model considering the dependent relationship between the motor and IGBT improves the accuracy of reliability assessment. The remaining useful life prediction accuracy is improved by the parameter updating approach that combines Bayesian and expectation-maximization.

It is a challenge to investigate the interrelationship between the geometric structure and performance of sensor networks due to the increasingly complex and diverse architecture of them. This paper presents two new formulations for the information space of sensor networks, including Lagrangian and energy–momentum tensor, which are expected to integrate sensor networks target tracking and performance evaluation from a unified perspective. The proposed method presents two geometric objects to represent the dynamic state and manifold structure of the information space of sensor networks. Based on that, the authors conduct the property analysis and target tracking of sensor networks. To the best of our knowledge, it is the first time to investigate and analyze the information energy–momentum tensor of sensor networks and evaluate the performance of sensor networks in the context of target tracking. Simulations and examples confirm the competitive performance of the proposed method.

Most existing formation control approaches for Unmanned Aerial Vehicle (UAV) swarm assume that global position and global coordinate frame are directly available for each agent. To extend the application domain, this paper proposes a distributed bearing-based formation control scheme, without any reliance on global position or global coordinate frame. The interactions among UAVs are described by a directed topology with two-leader structure. To address the issue of unavailable global coordinate frame, we first present a distributed orientation estimation law for each UAV to determine its orientation under the coordinate frame of the first leader. Based on the orientation estimation, we then design a bearing-based formation control law to globally asymptotically track target moving formations. Finally, simulation results are provided to validate the proposed method, which show that the translation, scale and orientation of the formation can be flexibly controlled via two leaders.