This paper presents a digital-twin-based model predictive control framework for autonomous process control, demonstrated in a virtual experiment on thermal food processing in a convection oven. In combination with prior work, this approach enables simulation-centered food scientists to deploy physics-based simulation models in live process control environments. The digital twin is realized as a physics-based, data-driven reduced-order model (ROM) that provides faster-than-real-time predictions. The ROM is trained on trajectories from a high-fidelity multiphysics finite-element model of chicken fillets. A central contribution is a model predictive control scheme that overcomes the common fixed-initial-condition limitation of augmented neural ordinary differential equation ROMs: a dedicated sub-optimization step re-synchronizes the surrogate to the measured state of the food item at each control instant, allowing reliable live re-optimization without access to internal ROM states. The controller optimizes oven temperature setpoints to meet target food-quality metrics (core temperature, moisture content, texture) and autonomously accommodates changes to the planned end time during operation. Quantitatively, the ROM achieves relative time-series errors of 0.18–0.49%, and the control algorithm evaluates 501 trajectories of 1800 s real time in a total of 46.6 s on a single core of a processor, demonstrating on-device feasibility without cloud or edge resources. Receding-horizon model predictive control of the remaining setpoints mitigates model–reality mismatch, enforces user-defined food metrics, and sustains closed-loop performance under autonomous operation. • Software-agnostic pipeline to deploy digital-twin surrogates for online control. • Accurate, faster-than-real-time predictions enable autonomous operation. • Online sub-optimization re-synchronizes predictions to measured core temperature. • Level 5 autonomy case beyond the trained time window. • On-device digital twin operation possible without cloud or edge resources.

• Neural network-based MPC for heat recovery from PEM electrolysis with a heat pump. • Physics-based simulation model provides virtual training data. • Performance tested under volatile power input and compared with PI control. • MPC increases heat recovery efficiency and reduces temperature setpoint deviations. Hydrogen is expected to play a key role in future energy systems, with PEM electrolysis being particularly suitable for producing hydrogen from renewable sources. However, a significant amount of heat is released during operation. Utilizing this heat in heating networks or industrial processes could improve economic efficiency and help decarbonize the heating sector. Since PEM electrolyzers operate at relatively low temperatures, many applications need heat pumps to increase the temperature of the waste heat. Such a coupled system requires a controller that can handle the thermal management of the electrolysis stack with the heat pump. However, the rapid load changes of a PEM electrolyzer coupled with fluctuating renewable energy sources pose a challenge to the controller, as it must stabilize both the electrolyzer’s cooling cycle and the heat pump’s refrigeration cycle simultaneously. Therefore, this paper presents a model predictive controller (MPC) based on neural networks that efficiently controls waste heat recovery while meeting the temperature requirements of the electrolyzer and the heat sink. We investigated the performance of the control strategy in numerical case studies and compared it with conventional control using proportional-integral (PI) controllers. We could demonstrate that our method provides significant improvements in terms of minimizing temperature fluctuations and maximizing heat recovery efficiency. Under the influence of volatile power input, the MPC could increase the average efficiency of the heat pump by up to 7 %, reduce the use of auxiliary heating energy by up to 52 %, and reduce the average and maximum deviation from the temperature setpoints by up to 1.4 K and 10.4 K, respectively, compared to PI control.

Model predictive control with state estimation for heavy oil hydroprocessing in a slurry-phase reactor

This article presents an adaptive ridge regression (ARR)-based data-driven current prediction model aimed at enhancing robustness in permanent magnet synchronous motor (PMSM) drives. Although least squares (LS) methods have displayed certain potential for improving parameter robustness in model predictive control (MPC) applications, the influence of the inevitable data noises has not been fully considered in such a data-based method. Aiming to improve the robustness against data noises, an adaptive ridge regression (ARR) data-driven current prediction is proposed for permanent magnet synchronous machine (PMSM) drives in this article. The proposed method introduces an online evaluation of data noise severity through a variance inflation factor (VIF). A dynamic ridge coefficient adjustment mechanism according to the VIF is then applied. As a result, an adaptive penalization to the data noises is achieved, leading to an improved robustness. Experimental verification on a PMSM drive system shows improvements compared to conventional LS methods, especially under challenging low-speed operating conditions where noise sensitivity is most evident. The proposed ARR method preserves the parameter-independent benefits of data-driven approaches while achieving superior current prediction accuracy in noisy settings, thereby advancing the practical implementation of robust MPC strategies for PMSM drives.

Deep reinforcement learning has proven an effective method to solve many intricate tasks, yet it still struggles with data efficiency and generalization to novel scenarios, as required in settings such as robotics. Recent approaches to deal with this include (1) unsupervised pretraining of the agent in an environment without reward signals, and (2) training the agent using offline data coming from various possible sources. In this paper we propose to consider both of these approaches together, resulting in a setting where different types of data streams are available and fast online adaptation to new tasks is required. Towards this goal we consider the Unsupervised Reinforcement Learning Benchmark and show that unsupervised training's primary value lies in its use as a source of exploration trajectories, beyond its role in pretraining a policy. Following this observation we develop a method based on a world-model as a generative model of offline exploration data and model predictive control (MPC) planning. We show that this approach outperforms previous methods and demonstrates task adaptation which is 10 times faster than previously shown. We then propose a setup that includes access to both unsupervised exploratory data and offline expert demonstrations when testing the agents' online performance on adaptation to novel tasks in the environment.

Artificial intelligence in pharmaceutical manufacturing: Applications, case studies, and GxP implementation considerations.

Safety-critical control using high-dimensional sensory feedback from optical data (e.g., images, point clouds) poses significant challenges in domains like autonomous driving and robotic surgery. Control can rely on low-dimensional states estimated from high-dimensional data. However, the estimation errors often follow complex, unknown distributions that standard probabilistic models fail to capture, making formal safety guarantees challenging. In this work, we introduce a novel characterization of these general estimation errors using sub- Gaussian noise with bounded mean. We develop a new technique for uncertainty propagation of proposed noise characterization in linear systems, which combines robust set-based methods with the propagation of sub-Gaussian variance proxies. We further develop a Model Predictive Control (MPC) framework that provides closed-loop safety guarantees for linear systems under the proposed noise assumption. We apply this MPC approach in an ultrasound-image-guided robotic spinal surgery pipeline, which contains deep-learning-based semantic segmentation, image-based registration, high-level optimization-based planning, and low-level robotic control. To validate the pipeline, we developed a realistic simulation environment integrating real human anatomy, robot dynamics, efficient ultrasound simulation, as well as in-vivo data of breathing motion and drilling force. Evaluation results in simulation demonstrate the potential of our approach for solving complex image-guided robotic surgery task while ensuring safety.

Precise regulation of carbon dioxide (CO 2 ) concentrations in plant growth chambers is critical for ensuring reproducible and physiologically relevant research outcomes. CO 2 assimilation varies significantly with plant genotype, growth conditions, crop density, and phenological stage. However, estimation and control approaches heavily dependent on mechanistic crop models are at odds with the objectives of plant characterization units (PCU), where model availability for specific crops may be lacking. Moreover, in Bioregenerative Life Support Systems (BLSSs), such methods may struggle with multiple crops, intercropping, staggered harvesting and unknown growth stages. We propose a real-time, crop-agnostic method to estimate photosynthetic and respiration rates from CO 2 concentration data, without relying on crop-specific mechanistic assumptions. This improves robustness against the time-varying conditions typical of BLSSs, and supports operation with crops lacking validated physiological models. The resulting rate estimates support diagnostic algorithms, supervisory logic and CO 2 concentration controllers, and provide the modeling foundation for our second contribution: a hybrid Model Predictive Control (MPC) strategy for CO 2 regulation. The controller employs a mixed-integer formulation to handle the disjoint operating ranges of injection valves and incorporates explicit compensation for CO 2 measurement delays, ensuring accurate mass balances under real operating conditions. We demonstrate the effectiveness of the approach through in vivo experiments in a PCU realized under the ESA-MELiSSA framework. • Real-time observer for estimating photosynthesis and respiration rates. • Hybrid Model Predictive Control to ensure highly precise CO 2 regulation. • Robustness to intercropping, staggered harvesting and unknown growth stages. • Control formulation favoring integration in Bioregenerative Life Support Systems. • Validation through in vivo experiments in a Plant Characterization Unit.

Finite-Time Model Predictive Force Control for a Cable-Driven Elbow Rehabilitation Exoskeleton Aided by Series Elastic Actuator

Stabilization of strictly pre-dissipative nonlinear receding horizon control by terminal costs

A frequency-informed adaptive variable-step model predictive control framework with stability guarantees for cascade canals

Impact of structure on the performance of distributed model predictive control — Insights from an experimental case study

Self-loop physics-informed neural network for model predictive control of PEM electrolyzers

Comparative analysis of model predictive control and MPC-informed Rule-Based control for thermal storage operation in Ultra-Low temperature 4th generation district heating networks

A model predictive control approach for preventive maintenance optimization of an electronic assembly system

Dual-layer aperiodic event-triggered model predictive control for formation of autonomous marine vehicles with prescribed-time strategy

Implementation of a machine learning supported model predictive control for a 5th generation district heating and cooling energy hub

Neural ODE–driven safe model predictive control for functional electrical stimulation cycling

A novel model predictive control law for over-actuated space cable-driven manipulators

Experimental validation of parallel model predictive control on multiple low-resource IoT devices