In this paper, we present a real-time control strategy for a unmanned aerial vehicle operating under external disturbances, utilizing a Tube-based Model Predictive Control (Tube-MPC) framework integrated with a zonotopic representation. The approach ensures robust control by enforcing a bounded deviation between the real system trajectory and the nominal trajectory. The UAV’s control problem is formulated as a linear MPC for nominal control, and a state feedback controller for disturbance rejection, using zonotopes to compute robust invariant sets. The proposed method is tested on a UAV equipped with a Pixhawk 4 flight controller and Raspberry Pi 5, where the system’s robustness is validated in outdoor experiments under disturbances. The results demonstrate both the system’s performance and safety under real-world conditions.

Based on an available digital twin of a Vero cell culture process composed of a dynamic mechanistic model and a soft sensor, the impact of viral amplification is studied and optimized in the process development environment of Sanofi (Marcy l’Etoile, France). The soft sensor (an extended Kalman filter) uses Raman probe online measurements of biomass and some metabolites to estimate the infection titer and is combined with a model predictive controller that aims to optimize the infection titer while regulating the main substrate concentration level. The setup is validated in simulation, and a first experimental investigation is discussed.

A macroscopic model that considers the shift to lactate consumption is proposed to simulate the therapeutic proteins produced by mammalian cells using human embryonic kidney 293 cells (HEK293-6E) cultures. The macroscopic model contains four macroscopic reactions: (i) substrate consumption and lactate production, (ii) lactate consumption, (iii) biomass death, and (iv) viable biomass maintenance. The maximum likelihood principal component analysis has been used to extract the number of reactions from experimental data collected in four batch experiments. The model parameters and confidence intervals are obtained via nonlinear least-squares identification. The model successfully predicts the dynamic behavior of cell growth and death, substrate consumption (glucose and glutamine), lactate production and consumption, and protein production. In addition, the proposed model structure is also validated as a generalized representation of the mammalian cell cultures that do not present lactate consumption.