This work enhances the performance of a steady-state open-loop optimizer for the automatic operation of solar photo-Fenton plants, an advanced oxidation technology used for the removal of organic microcontaminants from urban wastewater. The proposed strategy dynamically adjusts the reagent dosage in response to real-time variations in solar irradiance, improving process robustness without requiring additional measurements. Simulations under both clear- and cloudy-day conditions showed reductions in process constraint violations of up to 78 % for microcontaminant removal, with only minor increases in reagent consumption (< 15 %) and operating cost (< 2 %). The strategy effectively mitigates the impact of irradiance fluctuations, demonstrating its potential as a practical and cost-efficient solution for the robust optimization of large-scale solar photo-Fenton systems.
A ground-generation airborne wind energy system is considered. The device consists of a kite attached to an oscillating arm in the horizontal plane that generates electricity. The kite’s trajectory is constrained to follow a prescribed lemniscate pattern. An implicit modeling approach is adopted where the control force required to maintain the figure-eight trajectory is computed. This system is reduced to a two-dimensional second-order ordinary differential equation. Its limit cycles are studied numerically, with specific equilibria serving as starting points to obtain physically admissible periodic motions. These cycles depend on several design parameters including line length, arm inertia, and braking coefficient, that are tuned to maximize the average power generated by the arm’s oscillations. The system’s behavior is further analyzed across a typical range of wind speeds, demonstrating robustness to wind variations while maintaining positive line tension throughout the cycle.
This paper proposes a novel robust enhancement to the nullspace-based fault detection filter design method. Robustness against model uncertainties is achieved by minimizing the H∞ norm of the residual filter. The problem is formulated as a semidefinite program with linear matrix inequalities based on the Bounded Real Lemma, enabling a systematic and computationally efficient solution. The resulting robust filters reduce the effects of model uncertainties in the residual generation while preserving fault sensitivity. This ultimately increases the performance of the fault detection filters, as either lower detection thresholds can be implemented to enable faster detection times, or robustness against false alarms is increased when the original threshold values remain unchanged. The effectiveness of the approach is demonstrated first for a set of randomly generated linear systems. Subsequently, the method is applied to a realistic small fixed-wing unmanned aircraft model. Both analyses confirm the improved robustness of the fault detection filter designs.
Modern control systems require real-time monitoring to ensure safety and reliability, yet detecting changes in system behavior remains challenging in nonlinear, high-dimensional, and time-varying environments. We introduce an interpretable change-point detection framework based on truncated online Dynamic Mode Decomposition with control (toDMDc). The method combines optimal rank truncation with online system identification, enabling real-time adaptation to evolving dynamics while maintaining numerical stability. By comparing reconstruction errors between reference and test windows, the framework detects changes in system behavior. We demonstrate convergence and validate the approach on three case studies: synthetic step changes, nonlinear two-tank system with input delays, and industrial battery energy storage system. Results show that toDMDc-based detection achieves accurate change-point identification with interpretable statistics, bounded detection delays, and computational efficiency suitable for real-time deployment in safety-critical control applications.
Reducing anesthetic drug use while ensuring patient safety remains a key goal in perioperative care. This work introduces a digital twin framework for optimizing hypnotic drug delivery during general anesthesia. The twin combines pharmacokinetic–pharmacodynamic (PK/PD) models with patient-specific parameters identified from intraoperative data and is controlled by an EPSAC strategy. Two schemes are compared: a standard EPSAC and an AI-enhanced version using recursive least-squares learning to correct model bias. Both were tested on five VitalDB surgical cases, where the induction phase was replicated from clinical data and the maintenance phase simulated in a closed loop with five-second control updates. EPSAC reduced Propofol use by about 10–12% compared with standard Target-Controlled Infusion (TCI) while maintaining stable bispectral index (BIS) tracking, and the AI-augmented version achieved smoother infusion dynamics. These findings highlight the potential of digital-twin-based predictive control for safer and more efficient anesthesia management.
Incremental passivity facilitates the development of output regulators via decoupled designs of a passivation controller and an internal model. While this approach is effective for data-driven output regulation with noiseless data, it fails to handle the noisy case, as noisy data leads to a data-based system representation with uncertainties. This work addresses this issue by robustifying the data-driven incremental passivation design. We present a robust characterization of incremental passivity for a class of uncertain nonlinear systems and design a data-driven feedback controller that renders the closed-loop system incrementally passive. The proposed robust data-driven incremental passivation controller is then applied to data-driven output regulation via noisy data. Finally, a numerical example validates the proposed data-driven regulator.
This paper addresses the problem of resource-aware sensor selection for state estimation in linear dynamical systems, where the goal is to minimize the total number of sensor activations while maintaining a target estimation accuracy. As the problem is inherently combinatorial and NP-hard, optimal sensor utilization has often been approximated by leveraging sub/supermodularity within a greedy selection framework. However, the estimation metric is generally neither sub/supermodular. The main challenge lies in establishing approximation guarantees even when dealing with non-sub/supermodular objective functions, which can be applied to broader real-world applications. To address this challenge, we employ a curvature and supermodularity ratio-based greedy framework that quantifies how closely a non-supermodular objective approximates the supermodular case. Using this analysis, we derive a theoretical bound on the optimal number of sensor activations required to achieve a target accuracy, even under non-supermodular objectives, i.e. the mean squared error (MSE). Our theoretical results and numerical simulations across diverse linear systems demonstrate that the proposed bound reliably captures sensor usage while ensuring the desired estimation accuracy.
This paper addresses the challenge of intelligent evasion for high-speed flight vehicles under limited measurements (only partial observations such as line-of-sight angles). This restriction transforms the problem into a Partially Observable Markov Decision Process (POMDP), where conventional reinforcement learning policies struggle to generalize to unseen scenarios. This paper proposes a novel meta-learning solution based on Algorithm Distillation (AD). The resulting AD model exhibits in-context learning, enabling zero-shot generalization to new threat scenarios without gradient updates. Simulations show the AD method achieves substantially higher success rates across diverse test scenarios than both the original PPO and an expert-distillation baseline, narrowing the gap between static RL policies and the demands of real-world evasion tasks.
This paper addresses the problem of robust state estimation for nonlinear systems with non-additive noise. We build on ideas related to our previous work [1] by augmenting the sigma points with noise variables and embedding them into a minimax formulation. Specifically, the objective function is the variance of the state estimation error, the minimizer corresponds to the robust filter, and the maximizer represents the least favorable model within an ambiguity set about the nominal model. Simulation results demonstrate that the proposed filter achieves superior robustness and estimation accuracy compared to existing robust and standard sigma-point filters.
This paper studies the connectedness of graphons. It highlights that connectedness is related to some spectral property of the graphon-Laplacian operator, which is important for convergence of consensus and other diffusion-based dynamics on large-scale networks. Some equivalent characterizations of connectedness are given, and some subtleties in their definition are discussed through examples.
This paper addresses the problem of designing an H∞ state-feedback controller for sampled-data systems. We propose a new numerical technique to handle infinite-dimensional Differential Linear Matrix Inequalities defined over a compact real interval. For this aim, we use an extended robust version of Finsler’s Lemma to convert such infinite-dimensional conditions into a set of efficient, tractable Linear Matrix Inequalities. The proposed methodology provides a systematic way to eliminate the explicit dependence on the time variable, thereby avoiding the need for infinite pointwise evaluations.
We propose data-driven passivation strategies for switched linear systems with unknown system matrices. In particular, we consider two types of switching laws: controlled state-dependent switching and exogenous switching. Using offline open-loop input-state-output data, we derive linear state feedback controllers that render the switched system strictly passive via bilinear matrix inequalities (BMIs). Furthermore, passivation with noisy data, relaxation of the BMIs, and data-driven stabilization via passivity are discussed. Finally, numerical simulation results validate the performance of the proposed data-driven controllers.
The most conventional Multi-target tracking (MTT) techniques face challenging problems due to target deformation, occlusions, and clutters, thereby true tracks (target-track) becomes false track (clutter-track). In this paper, we have utilized a state-of-art sensor measurement method known as a motion capture system (mocap) sensor which is integrated with mocap Motive 2.2 sofware that is used to measure the position of targets such as unmanned aerial vehicles (UAVs). Although, mocap can track a UAV itself, however target occlusion, identification, and its trajectory behaviour remain challenging issues. Moreover, mocap measurement noise, UAV process noise, and unknown clutter measurements deteriorate the tracking performance. These technical issues can be mitigated by utilizing a Markov-Chain-Two (MC2) model for the UAV state dynamics, which effectively record the influence of the previous two state hypotheis, thereby the target behavior becomes more accurate. This work integrates linear multi-target tracking based on the integrated probabilistic data association (LMIPDA) by MC2 model and presents a MC2-based LMIPDA (LMIPDA-MC2) method to interface with the mocap system. The project prototype consists of a UAV, three robotics vehicles (RVs), mocap, computer, and a wireless network such as IEEE 802.11a WIFI protocol. The root-mean-squared position errors (RMSEs) of the proposed method is only 0.02 m, thereby achieving an improved estimation accuracy in occlusion and clutter as illustrated in the experimental results.
The notion of an L2+ induced norm, i.e., the L2 induced norm for non-negative input signals, has been proposed. Furthermore, the L2+ induced norm has been shown to be larger than or equal to $\frac{1}{{\sqrt 2 }}$ times the L2 induced norm for any LTI system. Also, a numerically effective method for computing the lower bound of the L2+ induced norm of given systems is proposed. This paper proposes a numerical method to compute a less conservative lower bound of the L2+ induced norm for given systems, which consists of optimizing the frequency and signal waveform parts. The proposed method also provides information about the signal waveform that achieves the lower bound of L2+ induced norm. A numerical example demonstrates the effectiveness of the proposed method.
Accurate state and parameter estimation is essential for monitoring and control, yet high-fidelity models are often unavailable or costly to obtain. As a result, many applications rely on data-based virtual sensors that learn mappings from time-windowed measurements or derived features to the quantities of interest. While effective, such approaches are prone to changes in process parameters and operating conditions—such as wear, load variations, or disturbances—which alter the underlying data–state relationship and introduce concept shift. Ensuring robustness therefore requires both selecting features that are insensitive to such shifts and choosing a mapping that preserves discriminative information under varying conditions. We analyse this challenge in the concrete setting of electromagnetic solenoids, where accurate estimation of the armature position is crucial for control and diagnostics but typically requires costly position sensors. However, virtual-sensor performance often degrades under varying external loads and wear-induced changes, which disrupt the current–position relationship. To mitigate this effect, we propose a virtual sensor design that (i) employs physics-derived features capturing position-dependent electromagnetic effects and (ii) trains the mapping on a mixture of data from multiple operating contexts to reduce shift sensitivity. Experimental results on a solenoid test bench demonstrate that this combined strategy effectively improves robustness and maintains estimation accuracy under changing load conditions.
This paper studies the stabilization of switched nonlinear systems with both unstable subsystems under time-dependent switching. Motivated by the observation that stability in such systems requires both frequent switching and proper activation ratios, we introduce a new stability condition based on average activation time (AAT), which can be viewed as a relaxation of strict periodic switching. Technically, we employ vector-valued Lyapunov functions (VLFs) combined with hybrid timers to construct a Lyapunov framework whose overall magnitude decreases despite the instability of each subsystem. We further derive explicit and easily verifiable stability conditions, which are validated through a numerical example.
This paper presents a structure-preserving model reduction framework for negative imaginary (NI) systems based on a state-space parametrisation. We first establish a constructive representation showing that any NI system without poles at the origin can be parametrised by a skew-symmetric, a positive semi-definite, an arbitrary, and a symmetric matrix, thereby revealing the internal state-space structure underlying the NI property. Leveraging this parametrisation, a projection-based model reduction method is developed that guarantees both the preservation of the NI property and exact interpolation of the transfer function at prescribed frequency points. Numerical examples on flexible structures are presented to demonstrate the validity of the proposed parametrisation, the interpolation accuracy and structure-preserving capability of the proposed method.
