An Improved Weight Adaptive Gaussian Sum Algorithm Based on Sparse‐Grid Quadrature Filter for Non‐Gaussian Models

The problem of state estimation in nonlinear/non-Gaussian systems has generated significant interest in the literature. A new version of Gaussian sum estimation algorithm based on the high-degree cubature Kalman filter (HCKF) is proposed to further improve accuracy and stability of the traditional Gaussian sum filter. The proposed GS-HCKF approximates the predicted and posterior densities as a finite number of weighted sums of Gaussian densities. It is corroborated in the theoretical analysis and the simulation that the proposed GS-HCKF has integrated advantages with respect to computational accuracy and time complexity for nonlinear non-Gaussian filtering problems.

In the long-term service period, the bridge health monitoring system produced a huge amount of monitoring stress data; proper handling of these data is one of the main difficulties in the field of structural health monitoring, especially to predict the structural stress based on the monitored data. The objectives of this article are to present: (1) a nonlinear dynamic model, (2) a nonlinear mixed Gaussian particle filtering algorithm for predicting the monitored data based on the dynamic model, and (3) an approach combining nonlinear mixed Gaussian particle filtering algorithm with structural health monitoring data to predict the structural stress under uncertainty in real time. And an actual example is provided to illustrate the application and feasibility of the proposed models and methods.

Synapses play a central role in neural computation: the strengths of synaptic connections determine the function of a neural circuit. In conventional models of computation, synaptic strength is assumed to be a static quantity that changes only on the slow timescale of learning. In biological systems, however, synaptic strength undergoes dynamic modulation on rapid timescales through mechanisms such as short term facilitation and depression. Here we describe a general model of computation that exploits dynamic synapses, and use a backpropagation-like algorithm to adjust the synaptic parameters. We show that such gradient descent suffices to approximate a given quadratic filter by a rather small neural system with dynamic synapses. We also compare our network model to artificial neural networks designed for time series processing. Our numerical results are complemented by theoretical analyses which show that even with just a single hidden layer such networks can approximate a surprisingly large class of nonlinear filters: all filters that can be characterized by Volterra series. This result is robust with regard to various changes in the model for synaptic dynamics.

A combined experimental and theoretical investigation of collision-induced dissociation (CID) of the quasi-one-electron system ${\mathrm{Na}}_{2}^{+}$ He at 80 eV center-of-mass energy is presented. This study, complemented by additional measurements with an ${\mathrm{H}}_{2}$ target, provides a detailed analysis of the competition between the two basic CID mechanisms: via excitation of electronic states of the ${\mathrm{Na}}_{2}^{+}$ molecular ion and via momentum transfer to one of the ${\mathrm{Na}}^{+}$ core. The experimental method is based on a combined coincidence and time-of-flight technique, resulting in a complete measurement of the velocity vectors of the two fragments and giving a full determination of the collision parameters. The theoretical analysis is based on the so-called nonadiabatic quantum molecular dynamics, developed recently. This theory treats self-consistently and simultaneously classical atomic motion and quantum electronic transitions in dynamical processes of atomic many-body systems using time-dependent density functional theory. It allows one to simulate the experiment in microscopic detail and, thus, provides a deep insight into the excitation and dissociation mechanism. The combined theoretical and experimental analysis can later be extended to more complex systems like larger molecules or clusters.

Acousto-ultrasonic health monitoring of composite materials and geometrically complex systems, such as a full-scale air vehicle, is a very complex process. In full-scale systems, the acoustic wave travels through several interconnected and acoustically mismatched pieces such as sandwich core panels, stringer stiffened skins, lugs, and holes at various velocities while undoing multiple reflections, refractions, and mode changes before reaching a surface mounted transducer. Existing techniques for acoustic emission (AE) source location and severity analysis rely on the assumption of constant wave speed and uninterrupted propagation path between sources and sensors. In complex fullscale systems this assumption, however is not valid. The goal of this paper is to present a distributed data centric acoustic emission sensor network and data processing algorithm for in-situ SHM of a full-scale air vehicle. The AE sensor network uses several lightweight direct-wave and network-fence sensors as well as various acquisition control time parameters to filter noise and minimize effects of cascading, reflection, and signal distortion. Source location is performed using a time-offset method with the assumption of material quasi-isotropy. Results from case study involving a full-scale composite fuselage instrumented with multiple AE sensors have shown that AE based SHM in geometrically complex and full-scale systems, despite exceedingly challenging, can be performed with reasonable accuracy using networked sensors and a series of adaptive signal acquisition control parameters.

Parameter self-tuning of servo control systems is crucial for optimizing automation processes, especially in complex systems such as permanent magnet synchronous motors. In this paper, a nonlinear adaptive whale optimization algorithm (NAWOA) is proposed and applied to parameter self-tuning, which improves the traditional whale optimization algorithm (WOA) by nonlinearly adaptively adjusting two parameters during optimization to enhance fast convergence and global search capabilities. A servo control system with three parameters to be tuned is constructed using both simulation and physical methods. Simulation and experimental results show that the NAWOA outperforms the genetic algorithm, particle swarm optimization, and WOA in parameter self-tuning of the servo control system with lower error indicators and fast convergence speed. Although it still faces the challenge of initial condition dependency, the proposed NAWOA provides a powerful solution for real-time industrial applications.

liuyujiao2005@126.com Abstract Aiming at a nonlinear/non-Gaussian filter problem, the data processing of a single-yarn strength testing system, a filtering method based on the particle filter algorithm is put forward. This paper expounds the principle, and the working process and the procedures of particle filter. It discusses in detail the application of particle filter in the single-yarn strength testing, the modeling of testing system state equation, and the implementation of this algorithm in a project. It also indicates that particle filter algorithm used to process single-yarn strength test data achieves a good effect and satisfactory filtering precision by simulated tests. Finally, this paper predicts that particle filter will play an important role in general data processing and analyzing. The detection of the yarn quality is very important for cotton production. Yarn strength is the reflection of inner quality of yarn and is the necessary condition of yarn's processing performance and end-use. The unit of single-yarn strength is Newton (N) per centimeter (CN). By stretching testing with a single-yarn strength tester, the yarn's characteristic parameters will be obtained, and the physical index of breaking strength, breaking elongation, fracture strength, fracture time and strength tensile curves can be determined, so the quality of yarn will be ascertained (1) . This paper majors in the research of the parameter of the single-yarn strength of yarn. The strength of yarn will increase with increasing within a certain twist; however the strength of yarn will decrease when the twist exceeds the critical value. The relationship between yarn elongation and tension is often nonlinear. When the yarn quality is tested by single-yarn strength tester, the interference noise of test environment in the industrial field is non-Gaussian. So the processing of single-yarn strength testing data is a typical nonlinear and non-Gaussian filtering problem. The traditional digital filtering methods normally only filter out the caused by single reason and can't achieve satisfactory results for complex noise. In order to solve this problem, a large number of nonlinear recursive filtering algorithm such as extended Kalman filtering (EKF), the modified gain Kalman filter (MGEKF), U Kalman filter (UKF) algorithm come forth in recent years, and these filters are based on particular assumption to ensure the optimality. The actual data is usually very complex, and contains many factors such as non-Gaussian, nonlinear, high dimensional and noise; in this case, the Kalman filter generally can't get the analytical solution. This is a common problem and appears in many different kinds of fields (2-11) . Now, a new sequential Monte Carlo particle filter algorithm based on Bayesian principles has been paid more attention. The particle filter technology realize the recursive Bayesian filter with non- parametric Monte Carlo simulation method, which can fit to any state space model and the nonlinear systems which can't be expressed by traditional Kalman filter, and its accuracy approach the accuracy of optimal estimation (12) . This paper places emphasis on introducing the principle and method of the particle filter, and the application of the nonlinear data processing.

Conventionally modelling and simulation of complex nonlinear systems has been to construct a mathematical model and examine the system’s evolution or its control. This kind of approach can fail for many of the very large non-linear and complex systems being currently studied. With the invention of new advanced high-speed computers and the application of artificial intelligence paradigms new techniques have become available. Particularly neural networks and fuzzy logic for nonlinear modelling and genetic algorithms [Goldberg, D. (1989)] and evolutionary algorithms for optimisation methods have created new opportunities to solve complex systems [Bai, Y., Zhuang H. and Wang, D. (2006)]. This paper considers issues in design of multi-layer and hierarchical fuzzy logic systems. It proposes a decomposition technique for complex systems into hierarchical and multi-layered fuzzy logic sub-systems. The learning of fuzzy rules and internal parameters in a supervised manner is performed using genetic algorithms. The decomposition of complex nonlinear systems into hierarchical and multi-layered fuzzy logic sub-systems reduces greatly the number of fuzzy rules to be defined and improves the learning speed for such systems. In this paper a method for combining subsystems to create a hierarchical and multilayer fuzzy logic system is also described. Application areas considered are - the prediction of interest rate, unemployment rate predication and electricity usage prediction. Genetic Algorithms can be used as a tool for design and generation of fuzzy rules for a fuzzy logic system. This automatic design and generation of fuzzy rules, via genetic algorithms, can be categorised into two learning techniques namely, supervised and unsupervised. In supervised learning there are two distinct phases to the operation. In the first phase each individual is assessed based on the input signal that is propagated through the system producing output respond. The actual respond produced is then compared with a desired response, generating error signals that are then used as the fitness for the individual in the population of genetic algorithms. Supervised learning has successfully applied to solve some difficult problems. In this paper design and development of a genetic algorithm based supervised learning for fuzzy models with application to several problems is considered. A hybrid integrated architecture incorporating fuzzy logic and genetic algorithm can generate fuzzy rules that can be used in a fuzzy logic system for modelling, control and prediction

The heterogeneity of data between different clients limits the performance of federated learning training, and hyper-knowledge distillation has emerged as a potential solution to address heterogeneous data. However, existing hyper-knowledge distillation methods often sacrifice efficiency to improve model performance. To address these challenges, this paper proposes FedHDE, a novel federated learning framework based on hyper-knowledge distillation and efficient tensor optimization. FedHDE integrates a Dynamic Client Selection strategy within a teacher–assistant–student hierarchical structure to enhance global aggregation accuracy and reliability without relying on public datasets or server-side models. Furthermore, an efficient tensor optimization algorithm is introduced to improve numerical stability and reduce computation time. Theoretical analysis demonstrates that FedHDE achieves a convergence rate of $$O(1/\sqrt{T})$$ O ( 1 / T ) , ensuring efficient and stable learning. Experimental results on the CIFAR10 dataset show that FedHDE improves both local and global accuracy by 0.14%–4.32%, while reducing training time by 65.28% compared to state-of-the-art baselines.Code is public at https://anonymous.4open.science/r/FedHDE .

Digital Twins (DTs) and Virtual Reality (VR) have recently gained significant popularity. The term digital twin refers to a virtual representation of a real-world environment. DTs are known for providing intuitive, accurate, and real-time visualizations of complex physical systems. In this paper, we present a novel, cloud-based framework for creating digital twin environments using VR within a cloud environment. We demonstrate our framework with two real-world case studies inspired by clinical insights obtained from experts in post-stroke rehabilitation. In the case of rehabilitation, sessions within virtual scenarios mimicking patients’ everyday life may significantly improve their rehabilitation quality. Inspired by biofeedback, we used a digital twin in VR to allow patients to enhance hand stability using real-time data streaming. Our second case study demonstrated the utility of our approach for allowing patients to more comfortably navigate within their homes by generating a virtual twin of an apartment. Our work is designed to transcend the boundary between reality and virtuality by letting users interact directly with the real and virtual worlds simultaneously. To achieve real-time synchronization between the real and the virtual environments, we established an infrastructure blueprint that features efficient sensor data collection, streaming, processing, and storage in a secure cloud environment. Data may then be accessed from any site, directly from the cloud, to update the virtual environment. In the end, we show that our infrastructure is generalizable across clinically-relevant tasks. We hope to extend our work to other areas that would benefit from a digital twin-based VR space.

Computer systems are becoming increasingly complex, especially interactive software systems, namely software user interfaces. The scientific community relies on different methods to assess their safety. This article provides an updated literature survey on hazard analysis approaches used to improve the safety of complex systems. To support the survey, we conceptualise complex systems, highlighting the challenge in terms of assessing their safety. We provide a brief overview on the approaches historically available to tackle issues in those systems, along with their most common methods. Finally, the article focuses in one method of a non-traditional approach, which is described in more details, along with some of its extensions, which seeks to improve the hazard analysis in complex systems.

The invention of tissue-equivalent proportional counters simulating micrometre diameter volumes, intended to measure the linear energy transfer of a radiation field, resulted in a practical demonstration of the stochastic nature of energy deposition in small volumes. Besides contributing to a better understanding of the interactions between ionising radiation and biological systems, these detectors have had a significant impact on applied radiation dosimetry. The initial instruments were elegant but suitable only for laboratory experiments because of their sensitivity to environmental conditions and the complex support systems they required. However, their ability to separate the dose due to neutrons from that delivered by photons motivated detector design modifications that eventually resulted in robust detectors suitable for use as radiation survey instruments. Proportional counters simulating micrometre tissue volumes turned out to be the ideal detectors for monitoring the complex radiation environments, including on the space shuttle and International Space Station, and have served as the primary active dosimeters in space for nearly two decades. The need for more sophisticated measurements has led to further improvements in detector design, and the need for smaller and lighter dosimeters is motivating further developments in both detectors and data processing systems.

This paper introduces the study on adaptive beamforming algorithm for phased antennas, a modified orthogonal algorithm was put forward in this paper. The formulas were derived and steps of computation were given. The new algorithm had less computation than orthogonal algorithm with the similar performance. The theoretical analysis and computer simulation results showed that the algorithm was effective.

In the era of digital technologies, working with large volumes of data is undergoing significant changes due to new technologies. This article presents a theoretical and methodological description and analysis of the integration of technologies such as artificial intelligence, machine learning, cloud computing, and blockchain to address complex tasks in data processing and information management. The subject of the study is the development of an efficient data storage system to solve applied tasks in various scientific fields. Cloud computing provides seamless access to digital resources, supports remote services, and reduces dependence on physical infrastructure by hosting databases and management systems on virtual platforms. This fosters scalability and distribution of computational power for processing vast amounts of information and deploying complex artificial intelligence models. Artificial intelligence, particularly automated machine learning (AutoML) and generative models, automates labor-intensive stages of data preparation, significantly enhancing the efficiency and accuracy of models. Blockchain technology, in turn, addresses critical issues of security, data integrity, and digital rights management, creating a trusted and immutable environment for decentralized data processing systems. The joint application of these technologies lays the foundation for the creation of self-optimizing, scalable, and secure automated data processing systems, as demonstrated by examples in library science, healthcare, and linguistics. For instance, blockchain technology ensures the transparency, reliability, and security of library records. It plays an important role in managing intellectual property, digital rights management (DRM), and authenticating rare or archival documents. In linguistics, we note the ability of data structures to handle linguistic ambiguity and demonstrate complex relationships between language elements, using logical input mechanisms and principles of synergetics. Further work is expected to develop more advanced automation technologies for rapid deployment of trained models with fewer data and the potential for use in various fields.

Overview of Nonlinear Bayesian Filtering Algorithm