Meeting: June 11, 2026

In positioning systems based on ironless permanent-magnet linear synchronous motors (PMLSMs) supported by rolling-element bearing guides, friction constitutes the dominant disturbance. Achieving sub-micrometer positioning accuracy, therefore, requires explicit consideration of friction effects. In model-based friction estimation and compensation strategies, the LuGre model [1] provides a favorable compromise between physical fidelity and parameter parsimony; however, its identification remains challenging due to nonlinear dynamics and strong parameter coupling. To address these challenges within a nonlinear optimization framework, this work proposes a comprehensive step-by-step LuGre identification methodology. The approach integrates a refinement stage based on nonlinear optimization, following the alternating linear and nonlinear regression procedure of [2], with an experiment-based parameter initialization derived from the classical two-step approach [3, 4]. A friction-oriented experimental protocol is proposed to support the identification procedure. Experimental validation is conducted on a precision linear stage from MKS Instruments. The identified LuGre model is then implemented in a feedforward friction-compensation scheme, providing a practical assessment of the proposed identification strategy for high-precision positioning systems.

Slides – Video

Type 1 Diabetes Mellitus (T1DM) is a chronic autoimmune disease characterized by the destruction of pancreatic β cells, i.e. the cells responsible for producing the endogenous insulin needed for maintaining Blood Glucose (BG) within the range of 70–180 mg/dL. As a result, patients with T1DM require lifelong exogenous insulin therapy. Nowadays, the use of Automated Insulin Delivery (AID) systems is well established as the most promising therapeutic approach for BG regulation. Due to meal intake, physical activities, or physiological variations (e.g. stress), T1DM patients are, however, still subject to the risk of hypoglycaemia (BG 180 mg/dL) [1]. In this context, the introduction of BG prediction into control policy has received increasing attention [2], see e.g. the use of BG prediction models coupled with Model Predictive Control (MPC) setups in [3] or the ARX models in [4]. Such predictive capabilities additionally enable the anticipation of glycaemic events, allowing corrective action before glucose levels breach safety thresholds [5].
In this work, three Black-Box models for multi-step glucose prediction are identified using real patient data extracted from a cohort of 29 adult patients with T1DM [6]. Linear ARX, ARMAX (with moving-average noise) and Box–Jenkins (BJ) (with independent process and noise dynamics) models are investigated. Prediction performance is evaluated in terms of Root Mean Square Error (RMSE) and FIT for horizons of 5, 15, 30, and 60 minutes. All models areidentified on a patient-specific basis to account for inter-patient variability.

Slides – Video

Switching control strategies designed using the hybrid system framework are promising in the field of power electronics, by providing stability guarantees and robustness to parameter variations. However, they rely on solving Linear Matrix Inequalities (LMIs), which resolution does not scale well with the dimension of the considered system. To address this challenge, this work proposes a data-driven methodology that combines system identification and model order reduction for hybrid systems. The objective is to identify reduced-order models that preserve the essential states of the converter, in order to perform control design.
The effectiveness of the method is demonstrated through an application to a DC–DC buck mode power converter circuit with two legs from the OwnTech Foundation.

Slides – Video

This work focuses on the estimation of population dynamics models based on the Yellowhammer benchmark, which relies on observation data of the Yellowhammer in France, as well as climatic and habitat variables. The objective is to construct a partial differential equation model with variable parameters, capable of representing the effect of habitat quality on the spatio-temporal evolution of the population.
In order to account for landscape heterogeneity, a local approach is adopted, in which the model structure and the estimated parameters are adapted to each study area. A global sensitivity analysis of the estimation criterion then makes it possible to identify the most influential parameters, reduce the model, and improve its identifiability.
The selected sub-model is subsequently calibrated using the Levenberg–Marquardt algorithm. This approach leads to reduced models that are numerically more robust while preserving the essential mechanisms of the dynamics under study.

Slides – Video

The quality of an estimated nonlinear model highly depends on the data quality that was used for the system identification. By using a Gaussian Process (GP)-based optimal input design approach, a so-called space-filling dataset can be generated in the feature space of the system model. However, the resulting input design can be costly to apply to the real system. Therefore we propose a space-filling input design strategy that can minimize the experimentation cost in terms of a user defined measure, while still guaranteeing a prescribed level of space-fillingness. Through a Monte Carlo simulation study, we demonstrate that the proposed method can appropriately shape the excitation signal to significantly reduce the experimental cost compared to a classical optimality criteria while the identified model performance remains adequate.

Slides – Video

Parameter identification for hyperbolic PDE systems from partial observations is a major inverse problem in distributed parameter systems.
We propose a Physics-Informed Neural Network (PINN) approach in which the unknown physical parameters are co-optimized with the network weights through a loss functional including the PDE residuals.
Applied to a macroscopic crowd dynamics model, the method estimates the relaxation rate α and the curvature parameter c of the speed–density law, with relative errors below 1.5 % and without spatial discretization.

Slides – Video

Human activities, particularly the extensive use of synthetic fertilizers, have significantly disrupted the nitrogen cycle, causing major environmental impacts such as eutrophication and biodiversity loss. To support more sustainable agriculture, it is crucial to better understand soil nitrogen processes like nitrification. This study presents a preliminary approach to identifying key parameters of a dynamical nitrification model using experimental concentration data obtained under controlled laboratory conditions.

Slides – Video

Active Magnetic Bearings (AMB) are widely used in high-speed rotating machines due to their ability to levitate rotors without mechanical contact. The considered AMB-supported rotor system is a multivariable system that is open-loop unstable, which makes system identification particularly challenging and requires identification to be performed in closed loop. This article aims at developing data-driven identification methods to track the evolving dynamics of AMB systems in real time, with the objective of detecting performance degradation and anticipating potential failures.
At the current stage of the research, the system is assumed to be time-invariant and the objective is to obtain an initial characterization of its dynamic behavior, in particular the resonance and anti-resonance information. For this purpose, a nonparametric identification is carried out in the frequency domain. Multisine excitation signals are applied to the system and only steady-state responses are retained. Several estimators were evaluated, and the Errors-In-Variables (EIV) estimator was selected to estimate the frequency response Gk. This step provides an accurate nonparametric representation of the system dynamics. Based on this result, future work can be extended to the identification of parametric models.

Slides – Video

Despite continuous technological advancements, air pollution remains a significant health risk for the human population. In 2020, 96% of the European urban population was exposed to levels of fine particulate matter exceeding the health-based guideline level set by the World Health Organization. Due to these reasons, atmospheric models that can forecast air pollution phenomena and analyze them in a proper way are now essential. Atmospheric phenomena are very complex processes and the numerical methods currently in use are quite expensive in terms of computation time. Therefore, seeking a mathematical model capable of replicating and explaining the behavior of the phenomenon is essential. To accurately reproduce the dynamic behavior of the process, simulators embed nonlinear and highly complex dynamics. This leads to prohibitive computational cost when it comes to simulations, predictions, and estimations. A practical solution is the construction of reduced order models, which are easier to handle, interpretable, and can provide faster results. The reduced order models can be particularly useful for decision makers in emergency situations where a defined area (e.g., urban zones, industrial sites or airports) is exposed to a high concentration of pollutants. Indeed, existing tools for atmospheric modeling, such as Large-Eddy Simulations (LES), require substantial computational effort, whereas in emergency scenarios, it is essential to reduce the prediction time. This work addresses the above issues by providing a methodology that constructs nonlinear reduced order models in a quick and scalable manner.

Slides – Video