Ali Arsalan

In modern maritime operations, electric ship power systems (SPSs) play a crucial role as they integrate distributed energy resources to meet stringent environmental targets while ensuring reliability in dynamic and isolated marine environments. Energy management systems (EMS) are critical to optimizing fuel efficiency, reducing emissions, and maintaining power quality under variable load demands and the harsh conditions in which ships operate. This paper presents a comprehensive review of EMS methodologies for SPS, covering traditional methods such as evolutionary algorithms, model predictive control (MPC), fuzzy logic control, and more modern approaches such as machine learning (ML) and deep learning (DL). ML and DL for EMS offer predictive and adaptive capabilities for real-time optimization, but face limitations in data security and centralized computational demands. Federated learning (FL), which is a decentralized, privacy-preserving paradigm, is a viable solution to the disadvantages of the conventional centralized form of training in ML. FL enables collaborative model training across distributed systems without raw data sharing, addressing critical concerns in cybersecurity and communication overhead. However, FL does require individual devices or clients to have enough computational power to carry out distributed optimization locally. A detailed overview of FL, including the FL training process, categories, architectures and various applications of FL in ship and energy systems, is presented in this paper. To further establish the usefulness of conducting energy management (EM) in ship systems, the paper presents a case study on a notional four-zone DC SPS in which FL has been utilized to collaboratively train DL models across multiple generators without sharing sensitive local data. The results demonstrate better generator output power prediction and effective load management in comparison to the conventional centralized learning setup, indicating the potential of FL to enhance energy efficiency and reliability in ship systems. Finally, some challenges and possible research gaps in applying FL to ship systems are discussed.

Electric drive systems (EDSs) are vital for automotive and industrial applications but remain highly vulnerable to cyber and physical anomalies (CPAs), such as inverter open-circuit faults, sensor failures, and malicious cyberattacks. Ensuring reliable EDS operation requires the controller to receive accurate and uncompromised feedback and reference signals continuously. However, many existing data-driven detection and mitigation strategies rely on large training datasets, impose significant computational overhead, and often lose effectiveness under various abnormal operating conditions. To overcome these limitations, this paper introduces a trust evaluation framework that continuously assesses the reliability of all incoming signals to the EDS controller by combining behavioral analysis with historical reliability records. The proposed scheme offers a lightweight and model-independent approach, enabling reliable, adaptive decision-making by leveraging both current and historical signal behavior. To this end, this paper further integrates the resulting trust values into a torque-split optimization algorithm, enabling adaptive load optimization by dynamically reducing the torque contribution of motors operating under abnormal or low-trust conditions, thereby demonstrating clear applicability for automotive drive systems. The framework is validated in a real-time OPAL-RT environment across multiple CPA scenarios, demonstrating accurate anomaly detection and adaptive torque redistribution. Owing to its simplicity and versatility, the proposed method can be readily extended to other safety-critical drive applications.

Modern and autonomous hybrid electric vehicles (HEVs), as complex cyber-physical systems, represent a key innovation in the future of transportation. However, the increasing interconnectivity and reliance on digital components expose these vehicles to significant cybersecurity risks. To address these challenges, Zero Trust Architecture (ZTA) has emerged as a promising security framework. Operating on the principle of ‘never trust, always verify,’ ZTA offers a comprehensive approach to ensuring continuous trust verification in HEV systems. Despite its potential, the application of ZTA within cyber-physical vehicular systems remains underexplored, and its practical benefits and limitations are not yet fully understood by the engineering community. To bridge this gap, this article presents a detailed survey of ZTA tailored specifically to the needs of vehicular CPSs, highlighting existing technologies, security challenges, and the application of zero-trust principles in HEVs. Additionally, this work proposes a deep learning-based replay attack detection scheme for the battery management system (BMS) of HEVs. The approach leverages a deep learning model to estimate the battery's State of Charge (SoC), analyzing the Error of Estimation using the Inter-Quartile Range (IQR) technique. The detection system analyzes the Error of Estimation using the IQR technique, demonstrating a 74.25% containment ratio and detecting deviations up to 2.39 units during attack scenarios. The system maintains a balanced detection sensitivity with 25.75% detection density. While the proposed method demonstrates high effectiveness in detecting stealth replay attacks through simulation results, it faces certain limitations including computational overhead for real-time processing, dependence on high-quality training data, and potential vulnerability to adversarial attacks on the underlying deep learning model. These challenges highlight the need for careful consideration in practical implementations while opening avenues for future research.

Virtual integration of geographically dispersed laboratories through real-time co-simulation presents powerful capabilities for co-simulating massive complex systems but it is hindered by communication delays that compromise accuracy and stability. This challenge is particularly concerning for real-time power system co-simulation, where delays can induce synchronization loss and limit dynamic and transient studies. This study proposes an adaptive, multi-parameter model-free framework for predicting and compensating delays in co-simulated systems, addressing this critical issue. This framework leverages the improved damping impedance method interface algorithm and an adaptive, parameter-tuning predictor system that predicts and compensates for delays without requiring complex interface signal transformation, processing, decomposition, reconstruction, phase estimation, system models, and no human interference. The proposed approach is validated using a joint experiment between two laboratories at Clemson, SC, USA and Greenville, SC, USA, with the Damping Impedance Method as an Interface Algorithm between the two partitioned subsystems. The coupling errors, state tracking errors, and residual complex power are used as evaluation metrics for the proposed delay compensation. This approach enhances co-simulation accuracy and stability, facilitating reliable dynamic and transient analyses.

Power electronics building block (PEBB) concept involves integrating fundamental components into functional blocks that can be stacked, extending converter power ratings for all-electric ships (AESs). This modular approach reduces costs, size, weight, design complexity, and maintenance. PEBBs can be realized as modular multi-level converters (MMCs), which offer advantages like modularity, low switching losses, minimal voltage/current quantization, high reliability, and efficiency. However, effective switching control methods are crucial to balance capacitor voltages and suppress circulating currents. This paper proposes an improved nearest level control (NLC) method that employs smoothed trapezoidal reference signals instead of sinusoidal references, aiming to enhance capacitor voltage balancing, suppress circulating currents, and improve the output power quality of PEBBs in AESs. The proposed NLC method is analyzed in real-time for an N-level PEBB connected to an induction machine (IM) with variable speed and torque load. The real-time verification is conducted in the Typhoon HIL606 digital real-time simulator (DRTS). The results validate the feasibility and effectiveness of the proposed NLC method for a three-phase N-level PEBB concept for AESs.

In recent years, data-driven methods have shown promise in diagnosing various open circuit fault (OCF) modes in inverter drive applications. However, existing studies primarily evaluate the reliability of these methods based solely on classification accuracy, neglecting critical real-time factors, such as computational delays and data transfer latency associated with the data-driven approach and communication protocols, respectively, which can affect real-time operational reliability. This article addresses these gaps by proposing a universally applicable reliability score criterion that integrates classification accuracy with system timing profiles. In addition, a model predictive control strategy employing active thermal management (ATM) is applied to the drive system, enabling a detailed analysis of the impact of OCF modes on the junction temperature of mosfets. Moreover, a novel feature extraction dataset is introduced, leveraging resistive/conduction loss data from the ATM scheme without requiring signal preprocessing. The proposed reliability score quantification and the dataset's diagnostic potential are validated using various data-driven classifiers. The most reliable classifier, achieving a 99.95% diagnosis accuracy, is further tested under diverse operating conditions using a control hardware-in-the-loop setup on an OPAL-RT testbed and Raspberry Pi.

Real-time cosimulation of geographically dispersed laboratories enables extensive system simulations but faces significant challenges from communication delays, impacting accuracy and stability. This issue is crucial in real-time power system cosimulation, where delays can disrupt synchronism and hinder dynamic analyses. This article proposes model-free predictive delay compensation methods as viable alternative signal transformation-based methods. This study explores the frequency domain stability of a model-free framework for delay prediction and compensation at power system interfaces using the ideal transformer method as interface algorithm. Similarly, time-domain implementation reveals signal amplitude magnification, addressed by introducing a delay- and frequency-dependent normalizing factor. This framework adapts interface coupling signals, enabling real-time parameter tuning without complex processing or system models, enhancing co-simulation accuracy and stability for distributed power system analyses.

Electric vehicle (EV) charging infrastructure hardware–software solutions and communication protocols have inherent vulnerabilities against cyberattacks. Due to the wide range of back doors and infiltration possibilities, there is an important need for solutions that can maintain critical service continuity during incidents. This study proposes a dynamic capacity sharing method for effective use of the constrained grid capacity between neighboring charging clusters in distribution grids when the communication link between the clusters’ operators and the grid operator is disrupted due to hardware faults or cyberattacks. The performance of the developed solution is thoroughly investigated in a Denial-of-Service cyberattack scenario that may take place at different times of the day in realistic scenarios involving residential demand and stochastic EV charging behavior. The analyses validated the effectiveness of the proposed method in improving the deteriorated service level per charging cluster and better utilization of an overall constrained capacity.

This article presents an innovative approach to improve hybrid powertrains’ cyber-physical security for high-speed, tracked, off-road vehicles against subtle and sophisticated cyberattacks. While cyber threats against hybrid tracked vehicles (HTVs) have been acknowledged, exploring advanced cyberattacks targeting battery lifetime and energy efficiency within the powertrain remains underexplored. This article proposes a dual-pronged intelligent methodology to construct novel FDIAs targeting vehicle control decisions from the energy management system (EMS). The optimal cyberattack strategy by a minimally informed adversary is formulated as a partial observable Markov decision process (POMDP), employing deep reinforcement learning (DRL) for online learning and attacking. Simultaneously, a comprehensive reward system is devised that incorporates sniffing features to elevate the stealthiness and efficacy of the cyberattacks. This augmentation ensures that the orchestrated attack vectors remain inconspicuous to human drivers. Evaluation metrics are formulated to assess the impact and stealth characteristics of the cyberattack. Furthermore, a sliding window-based controller data monitoring scheme combining iForest and dynamic time warping (DTW) algorithms is proposed to automatically detect the stealthy FDIAs in real-time, ensuring secure control of energy-efficient powertrain systems. Through the proposed metrics and detection module, the article thoroughly examines the impact of cyberattacks on the energy consumption of the HTV. It also provides vital insights for defending vehicles that operate in austere environments against sophisticated controller attacks.

Inverter-based electric drive systems (EDS) in electric vehicles (EVs) are susceptible to various common cyber and physical anomalies (CPAs), such as cyber-attacks (CAs) and open-circuit faults (OCFs) in power switches. Most existing data-driven diagnostic schemes for motor drives, solely rely on residual and 3-phase output current signals and are prone to misclassification due to overlapping data features associated with these anomalies. Moreover, these methods concentrate on detecting either CAs or OCFs independently, lacking a unified approach that can be universally applied to both types of anomalies. Therefore, in this work, a physics-informed machine learning (PIML) approach is proposed which can effectively distinguish between CAs and OCFs in EDS. In this regard, the unique features associated with stator current, resulting from various CPAs, are used to estimate the stator voltage characteristics through a synchronous motor mathematical model. The proposed voltage-based features characterizing better correlation via distinctive EDS transients in response to these anomalies are further used as the system’s prior information by a machine learning (ML) classifier. Moreover, active-thermal management (ATM)-based model predictive control (MPC) is utilized in the control layer due to its benefits in device-level thermal cycling and power quality. In addition, various variations in electrical, mechanical, and thermal characteristics due to CPAs in EDS have also been analyzed. The presented approach is evaluated for the US06 standard drive cycle in real-time via controller-hardware-in-the-loop (CHIL) experiment for various case study scenarios, resulting in a classification accuracy of 98.92%.