IFAC Journal of Systems and Control最新文献

Convolutional neural networks (CNNs) have enabled tremendous achievements in image classification, as the model can automatically extract image features and assign a proper classification. Nevertheless, the classification is lacking robustness to — for humans’ invisible perturbations on the input. To improve the robustness of the CNN model, it is necessary to understand the decision-making procedure of CNN models. By inspecting the learned feature space, we found that the classification regions are not always clearly separated by the CNN model. The overlap of classification regions increases the possibility to less perturbation induced input changes on classification results. Therefore, the clear separation of feature spaces of the CNN model should support decision robustness. In this paper, we propose to use a novel loss function called “conformity loss” to strengthen disjoint feature spaces during learning at different layers of the CNN, in order to improve the intra-class compactness and inter-class differences in trained representations. The same function was used as an evaluation metric to measure the feature space separation during the testing process. In conclusion, the conformity loss driven trained model has shown better feature space separation at comparable output performance.

The paper studies the problem of the spread of multi-competitive viruses across a (time-varying) population network and an infrastructure network. To this end, we devise a variant of the classic (networked) susceptible–infected-susceptible (SIS) model called the multi-competitive time-varying networked susceptible-infected-water-susceptible (SIWS) model. We establish a sufficient condition for exponentially fast eradication of a virus when a) the graph structure does not change over time; b) the graph structure possibly changes with time, interactions between individuals are symmetric, and all individuals have the same healing and infection rate; and c) the graph is directed and is slowly-varying, and not all individuals necessarily have the same healing and infection rates. We also show that the aforementioned conditions for eradication of a virus are robust to variations in the graph structure of the population network provided the variations are not too large. For the case of time-invariant graphs, we give a lower bound on the number of equilibria that our system possesses. Finally, we provide an in-depth set of simulations that not only illustrate the theoretical findings of this paper but also provide insights into the endemic behavior for the case of time-varying graphs.

Most of physical systems present time-varying delays in their inner dynamics. This causes instability, oscillation and even poor closed performance. Also, the present disturbance can cause instability. This article is addressing techniques to develop stability criteria for closed-loop and states estimation analysis of multiple time-varying delays systems. By selecting a suitable Lyapunov–Krasovskii functional (LKF), the derivative of double integration terms are upper bounded by using reciprocally convex matrix inequality. The closed-loop stability criteria are derived fulfilling $H_{\infty }$ performance index for multiple time-varying delays systems. Similar technique is also adopted to estimate unmeasured states fulfilling $H_{\infty }$ norm bound. The developed criteria are demonstrated to a numerical example. It is shown that H${}_{\infty }$ memory based controller has better performance on rejecting the introduction disturbance with having lower peak and shallow valley than other techniques.

This article discusses theoretical and practical aspects of real-time brain computer interface control methods based on Bayesian statistics. The theoretical aspects include how the data from the brain computer interface can be translated into a Gaussian mixture model that is used in the Bayesian statistics-based control methods. The practical aspects include how the control methods improve the performance of the brain computer interface. We use a reactive brain computer interface based on a visual oddball paradigm for the investigation and improvement of the performance of automatic control and feedback algorithms used in the system. By using automatic control for selection of the stimuli for the visual oddball experiment, the target stimulus is identified faster than if no automatic control is used. Finally, we introduce transfer learning using Gaussian mixture models, enabling a ready-to-use setup.

Mechanical Ventilation (MV) is an important therapy in the intensive care unit (ICU). Assisted spontaneous breathing (ASB) MV modes are a key and growing part of MV care, as they require less sedation and help avoid muscle atrophy. Equally, a lack of standardised approaches to MV care has led to the rise of model-based methods, which typically cannot estimate spontaneous breathing (SB) efforts, and are thus not able to be used for ASB MV. To address this issue, several models of SB effort have been created, though they require specialised added sensors and/or maneuvers. ►This research utilises a unique observational clinical dataset, which includes esophageal and gastric pressure measurements not typically taken in the ICU for N=6 patients. These measurements enable more direct model-based estimation of muscular pressure effort in SB MV patients. The data is analysed for all 6 patients for 3 models which include dynamics common to the current models. Models are assessed based on model fit. ►The results show all 3 models are unable to capture dynamics in 2 patients due to added muscular dynamics in their breathing, violating assumptions in the model dynamics or constraints. These results indicate the need for more flexible models and associated identification methods to better capture these dynamics.

A new methodology for fault tolerant control (FTC) is proposed to compensate actuator failures using Takagi–Sugeno systems. This makes possible to design the controller that represents actuator failures using a scaling factor by solving a family of linear matrix inequalities (LMIs). The resulting control system guarantees asymptotic stability, compensates the effect of actuator faults and ensures certain an $H_{\infty }$ performance level. This methodology is applied to the active suspension systems that motivated this research, where in the context of active suspension (AS) systems, the guaranteed performance correspond to ride comfort in the presence of road disturbances. Thus, a controller is developed for a quarter-car model with active suspension. The simulated results illustrate the effectiveness of the proposed approach.

This research delves into the domain of Human Activity Recognition (HAR) through sensor data analysis, offering a comprehensive exploration of three diverse datasets: UniMiB-SHAR, Motion Sense, and WISDM Actitracker. The UniMiB-SHAR dataset encompasses a diverse array of linear as well as non-linear and complex activities which involve the movement of more than one joint or muscle (for example Hitting Obstacles, jogging and falling with face down). This motion generates highly correlated sensor readings over a certain period of time. In this case, Convolution Neural Networks (CNNs) are effective in feature extraction as well as classification of HAR activities, but they may not fully grasp the combined features of spatial as well as temporal aspects in the HAR datasets and heavily rely on labelled data. Whereas, Graph convolution networks (GCN), with their capacity to model complex interactions through graph structure, complement CNN’s capabilities in classifying non-linear activities in the HAR dataset. By leveraging the Knowledge graph structure and acquiring the feature embeddings from the GCN model, in this study, a Noval ensemble CNN model is proposed for the classification of activities. The novel HAR pipeline is termed as Graph Engineered EnsemCNN HAR (GE-EnsemCNN-HAR) and its performance is evaluated on HAR datasets. Proposed model demonstrated a noteworthy accuracy of 93.5% on UniMiB-SHAR dataset, surpassing the Shallow CNN model with GNN with an improvement of 20.14%. The proposed model achieved a notable accuracy rate of 96.18% and 98% when evaluated on the Motion Sense and WISDM Actitracker dataset.

This paper proposes a novel approach to control and manage energy production based on grid requirements. The main challenge is the distance between the load and production area, making it difficult to quantify energy requirements in real time at the production area. To overcome this challenge, the proposed approach relies on an adaptive observer design that provides accurate and reliable estimates of multiple signals without expensive and unreliable sensors. In this study, a renewable energy source that consists of a wind turbine coupled to a permanent magnet synchronous generator is actuated with a Vienna power converter. However, it is crucial to emphasize that this approach can be implemented at any production site, regardless of its nature. The paper’s main contribution is the design of a novel adaptive high-gain observer that estimates the grid energy requirement based on the voltage value at the endpoint of the high-voltage direct current line. Moreover, the system load parameters are unknown and come non-linear in the system model. Simulations and analysis demonstrate the effectiveness of the proposed observer, showing convergence to the origin under the well-established condition of persistent excitation (PE).

Heat Integrated Distillation Columns (HIDiC) are highly energy-efficient technologies whose performance has been validated through robust optimization algorithms and practical tests. Despite these configurations are dynamically controllable technologies, the simultaneous relationship between dynamics and optimal energetic and economic performance under variable feed composition has not been analyzed. Thus, this paper tackles this gap in literature. Five binary mixtures and three feed composition were examined in this study. The optimization of these configurations was firstly achieved using a Boltzmann-based optimizer while the control properties were obtained through the closed-loop process analysis using the IAE criterion and rigorous simulations in Aspen Dynamics in a second stage. Results showed that the HIDiC configurations with the best dynamic behavior do not match with the HIDiC columns with the best energetic and economic performance. However, suboptimal HIDiC configurations experienced only slightly less energetic and economic benefits but better dynamic properties that the best HIDiC configurations. Particularly, the best suboptimal HIDiC columns to separate the mixtures with relative volatility ($\alpha$) lower than 1.4 were determined for a feed composition of 25 mol% for the light component. Nevertheless, the most adequate HIDiC columns to separate mixtures with $\alpha >1.4$ were obtained for equimolar feed composition and feed composition of 75% for the light component.