This paper proposes a hybrid digital twin framework that couples a real-time physics-based digital twin model with a data-driven diagnostic layer implemented through cloud-based data acquisition and analysis. This framework generates synthetic datasets across multiple speed levels and fault severities for bearing fault detection and classification in industrial spindle systems, where real fault recordings are costly, risky, and difficult to reproduce. Once the system is validated, a two-stage classifier is trained and used for online fault detection and fault-type identification, whereas the deep-sequence model provides offline verification. To improve robustness, training data are enhanced with multi-domain feature enrichment and targeted data augmentation techniques that simulate measurement noise and small operating variations. The resulting models achieved strong performance under previously unseen operating conditions within the validated digital twin envelope. Overall, the proposed approach reduces the dependence on real fault experiments by enabling the risk-reduced development and evaluation of data-driven bearing fault diagnosis.

This paper proposes an innovative model-free deep reinforcement learning-based controller (RL-C) for a grid-connected 5-level packed-U-cell (PUC5) multilevel inverter (MLI). The controller is designed to deliver a high-quality grid current while maintaining the PUC5 floating capacitor voltage at its reference level. In addition, the proposed controller supports both active and reactive power exchanges, adapts to variations in voltage and current references, and remains robust under grid voltage variations. The RL agent learns optimal switching actions through direct interaction with the PUC5 system, eliminating the need for data collection or reliance on existing control models. An Actor-Critic architecture is adopted, and the Proximal Policy Optimization (PPO) algorithm is applied for training (offline) using MATLAB/Simulink, where the RL-C is evaluated under diverse PUC5 configurations and operating conditions in the testing phase. The trained agent has been implemented on an Opal-RT real-time system and validated experimentally using a laboratory-made PUC5 prototype. The performance of the proposed RL-C approach is compared to both traditional approaches including finite control set model predictive control, sliding mode control, and PI control, and other state-of-the-art RL algorithms, demonstrating superior generalization and training efficiency. Moreover, a sensitivity analysis quantifying the impact of reward design, state space, network size, and key hyperparameters on convergence and performance is carried out.

This paper assesses a Hybrid Energy Storage System (HESS) at The Green Village (TGV) of Delft University of Technology (TU Delft), designed and developed as a combination of a lithium-ion battery and hydrogen storage systems to provide a residential energy supply. This paper will evaluate the combination of producing solar-powered green hydrogen through electrolysis, as well as daily and seasonal combinations of battery and hydrogen storage, and electricity generation through a fuel cell. Through an analysis of various sensor data that contained power and hydrogen flow, and control signals, this case study reports on the overall efficiency of the HESS, encourages user energy balancing strategies, and assesses its capability to store sustainable energy over long periods. Based on the same dataset, preliminary machine learning models have been developed and evaluated to predict hydrogen production from weather and PV inverter measurements, supporting future EMS optimization. Therefore, this case study indicates improvements that led to key observations.

This study presents a data-driven offline digital twin model of an operational residential hydrogen hub equipped with more than 100 sensors. The model enables analysis and scaling of hydrogen-based hybrid energy hubs from residential to larger systems. The hub integrates photovoltaic generation, battery storage, hydrogen production via electrolysis, compressed hydrogen storage, and fuel cell electricity generation. Using year-long field data, the model reproduces the current configuration (5.34 kWp PV, 15 kWh battery, ~ 45 kg H₂) and quantifies annual performance: 5,102 kWh PV generation, hydrogen production and consumption efficiencies of 48.0% and 39.6%, 25.3 kg H₂ produced versus 49.3 kg consumed, and net grid exchange of +114 kWh. Multi-scenario sizing shows that an optimized configuration (8.46 kWp PV, 30 kWh battery, 70 kg H₂) reduces grid import to ~ 30 kWh yr^-1 while exporting ~ 380 kWh, with the H₂ buffer ending the year near its initial state under a rule-based energy management strategy. The results demonstrate the capability of a sensor-validated framework for designing integrated PV-battery-hydrogen energy hubs.

This paper presents a simulation-based case study of a hybrid energy hub located at The Green Village (TGV), a living lab for sustainable innovations in Delft, The Netherlands. The energy hub integrates photovoltaic (PV) generation, battery storage, hydrogen production, seasonal storage, and usage to provide a fully electrified one-person residence, serving as a realistic testbed for the integration of renewable energy. A model of the hub is developed in Simulink/Matlab using historical operational data to simulate system behaviour under various edge case scenarios and system configurations. The model enables the evaluation of system-level interactions, operational strategies, and the impact of design choices on energy efficiency, self-sufficiency, and hydrogen integration. The simulation results show the sensitivity of the system performance to component sizing and EMS settings. This study provides valuable insights into the control and optimisation of The Green Village’s energy hub and its integrated energy systems, contributing to the practical deployment of resilient and sustainable energy hubs in the built environment.

The widespread use of modular multilevel converters (MMCs) in the evolution of complex power grids presents new challenges for grid stability. MMCs have highly nonlinear impedance characteristics due to their complex internal dynamics and intricate control architectures. Due to practical constraints, physics-based models cannot accurately compute these impedances, and the use of closed-box measurement techniques is time-consuming, resulting in a limited amount of data available for impedance characterization. Thus, using current methods to estimate impedances over a wide range of operating points can be unreliable. This paper presents a transfer learning-based framework for MMC impedance characterization using system-level parameters as operating point variables. The proposed approach predicts both AC and DC side impedances simultaneously by extrapolating impedances derived using state-space modeling approaches to real-time electromagnetic transient (EMT) simulations. Finally, the method is evaluated on a practical converter from the CIGRE B4 DC grid test system for various types of controllers and scenarios involving unknown parameters.

Hybrid AC/voltage source converter-based multi-terminal DC (VSC-MTDC) power grids play a crucial role in enabling long-distance power transmission and flexible interconnection between AC grids. To fully leverage the functional advantages of such systems, it is essential that they operate in or close to optimal power flow (OPF) conditions. To address this, ACDC-OpFlow is developed as an open-source and cross-language framework for solving AC/DC OPF problems. Its core innovation lies in a unified modeling structure that supports MATLAB, Python, Julia, and C++, with Gurobi used as a consistent solver backend. This framework is beginner-friendly and allows users to work in their preferred programming languages. Both text-based and graph-topology results are provided to help users understand the system-wide power flow distribution and operational status. This work presents the design concept of ACDC-OpFlow, showcases representative example results, and discusses the performance differences observed in multiple programming language implementations.

Electric vehicles (EVs) offer a promising solution for mitigating greenhouse gas emissions and minimizing the transportation sector's dependency on non-renewable energy sources. However, efficient energy management poses a significant challenge for their broader adoption, particularly optimizing battery usage, maximizing driving range, and improving overall vehicle performance. This paper presents the state-of-the-art Artificial Intelligence (AI) techniques used in electric vehicle energy management systems (EV-EMS), discussing a variety of deep learning algorithms of AI methodologies, such as, neural networks, and fuzzy logic. Additionally, This paper discusses the role of auxiliary techniques like transfer learning, which enhances model adaptability and reduces training time in AI-driven EMS applications. Through a systematic analysis of each method, this review identifies key trends, highlights the challenges and limitations of each technique, and offers perspectives on potential solutions and future research directions. The paper aims to support researchers, industry professionals, and policymakers in developing advanced, sustainable, and adaptable EV-EMS solutions that maximize battery life, improve vehicle performance, and facilitate real-time adaptive control. Finally, this review highlights the importance of AI-driven strategies in making EV technology more efficient, reliable, and scalable.

This paper presents a reinforcement learning controller (RLC) for a single-phase full-bridge rectifier as an interface for a battery energy storage system (BESS). A novel solution is presented that combines the traditional proportional-integral (PI) regulator with an RL-based control strategy using a proximal policy optimization (PPO) agent. In a high-fidelity Simulink-based digital twin setup, the agent learns to perform optimal switching actions for a single-phase full-bridge rectifier to achieve accurate current tracking and improved power quality. Simulation results show stable DC voltage regulation at 200V, tracking response under 0.1s, and harmonic compliance with THD equal to 2.38%. The hybrid control strategy guarantees robust dynamic performance and adaptability in the context of renewable energy and storage systems’ varying source and load conditions. The findings demonstrate the potential of coupling AI-driven control with digital twins to empower the autonomy and resilience of future smart energy systems.

This paper proposes a novel sensorless phase-shift modulation-based voltage balancing technique for a 5-level Packed U-Cell (PUC5) inverter. Two phase-shifted triangular carriers are used to modulate the reference signal and generate the appropriate gate pulses. The switching pulse generation is specifically designed to charge and discharge the capacitor at the speed of switching frequency, resulting in a fast voltage balancing of the auxiliary capacitor. Compared with the reported level-shifted modulation method, the proposed technique simplifies comparators and logic gates while keeping the benefit of fast sensorless voltage balancing and, consequently, the capacitor size reduction. In other words, it modifies the reference signal to achieve the fast voltage balancing of the auxiliary capacitor in PUC5. Simulation results are shown to investigate the effectiveness of the proposed technique.

In this paper, a Reinforcement Learning controller (RLC) is designed and implemented on a 5-level Packed U-Cell (PUC5) grid-connected inverter to control the injected current flowing into the electric network.The RL agent is trained using a Proportional-Integral (PI) reward function to optimize its control strategy. Moreover, the voltage balancing of the auxiliary capacitor in PUC5 is separated from the RL controller and integrated into the switching algorithm to reduce the training burden. This modification reduces the observation inputs required for RL training, significantly shorten the training time. Simulation studies conducted in Matlab/Simulink evaluate the performance of the proposed RL controller, demonstrating robust dynamic response and accurate tracking of reference signals across different operational conditions.