arXiv subjects

Davor Svetinovic

Publications and source records attributed to Davor Svetinovic.

15 recordsLinked to original sources

arXiv2026

An AI-Based Supervisory Measurement Integrity Validation Layer for Cyber-Resilient AC/DC Protection in Inverter-Based Microgrids

Line current differential relays (LCDRs) are measurement-driven relays that rely on time-synchronized multi-phase current waveforms to infer internal faults in AC and DC power networks. In inverter-based microgrids, however, the increasing reliance on digitally communicated measurements exposes LCDRs to false-data injection attacks (FDIAs), in which adversaries manipulate remote measurement streams to create protection-triggering yet physically inconsistent current trajectories. This paper addresses this emerging measurement integrity problem by introducing a measurement integrity validation scheme that operates as a supervisory instrumentation layer for modern LCDRs. The proposed scheme interprets short windows of synchronized instantaneous current measurements recorded during relay operation and assesses their physical consistency to distinguish genuine fault-induced trajectories from cyber-manipulated measurement streams. A recurrent neural network is trained offline using only relay-available current measurements and exploits the temporal structure of differential current waveforms, which remains informative in inverter-dominated systems where current magnitude is no longer a reliable observable. The method requires no additional sensors, auxiliary protection elements, or prior knowledge of network topology, and is applicable to both AC and DC LCDRs without structural modification. The proposed measurement validation scheme is evaluated on an islanded inverter-based microgrid under a comprehensive set of fault and FDIA scenarios, demonstrating high detection accuracy while preserving relay dependability. Hardware-in-the-loop validation using an OPAL-RT real-time simulator confirms that the scheme satisfies protection timing constraints and can operate in real time under realistic operating conditions.

cs.CR↗

arXiv2026

Towards Cognitive Defect Analysis in Active Infrared Thermography with Vision-Text Cues

Active infrared thermography (AIRT) is currently witnessing a surge of artificial intelligence (AI) methodologies being deployed for automated subsurface defect analysis of high performance carbon fiber-reinforced polymers (CFRP). Deploying AI-based AIRT methodologies for inspecting CFRPs requires the creation of time consuming and expensive datasets of CFRP inspection sequences to train neural networks. To address this challenge, this work introduces a novel language-guided framework for cognitive defect analysis in CFRPs using AIRT and vision-language models (VLMs). Unlike conventional learning-based approaches, the proposed framework does not require developing training datasets for extensive training of defect detectors, instead it relies solely on pretrained multimodal VLM encoders coupled with a lightweight adapter to enable generative zero-shot understanding and localization of subsurface defects. By leveraging pretrained multimodal encoders, the proposed system enables generative zero-shot understanding of thermographic patterns and automatic detection of subsurface defects. Given the domain gap between thermographic data and natural images used to train VLMs, an AIRT-VLM Adapter is proposed to enhance the visibility of defects while aligning the thermographic domain with the learned representations of VLMs. The proposed framework is validated using three representative VLMs; specifically, GroundingDINO, Qwen-VL-Chat, and CogVLM. Validation is performed on 25 CFRP inspection sequences with impacts introduced at different energy levels, reflecting realistic defects encountered in industrial scenarios. Experimental results demonstrate that the AIRT-VLM adapter achieves signal-to-noise ratio (SNR) gains exceeding 10 dB compared with conventional thermographic dimensionality-reduction methods, while enabling zero-shot defect detection with intersection-over-union values reaching 70%.

cs.CV↗

arXiv2025

Masked Sequence Autoencoding for Enhanced Defect Visualization in Active Infrared Thermography

Active infrared thermography (AIRT) became a crucial tool in aerospace non-destructive testing (NDT), enabling the detection of hidden defects and anomalies in materials by capturing thermal responses over time. In AIRT, autoencoders are widely used to enhance defect detection by reducing the dimensionality of thermal data and improving the signal-to-noise ratio. However, traditional AIRT autoencoders often struggle to disentangle subtle defect features from dominant background responses, leading to suboptimal defect analysis under varying material and inspection conditions. To overcome this challenge, this work proposes a Masked CNN-Attention Autoencoder (AIRT-Masked-CAAE) that integrates convolutional feature extraction with attention mechanisms to capture both local thermal patterns and global contextual dependencies. The AIRT-Masked-CAAE introduces a masked sequence autoencoding strategy, where the network learns to infer missing thermal responses from surrounding contextual cues, while suppressing background redundancy. In addition, the proposed masked sequence autoencoding approach enables training on only a subset of the thermal sequence, while providing generalizable latent representations and reducing training time by a factor of 30. The AIRT-Masked-CAAE framework was evaluated using specimens made of PVC, CFRP, and PLA. The results demonstrate that the AIRT-Masked-CAAE surpasses state-of-the-art AIRT autoencoders in terms of contrast, signal-to-noise ratio (SNR), and metrics based on neural networks.

eess.SP↗

arXiv2025

Large Language Model-Based Framework for Explainable Cyberattack Detection in Automatic Generation Control Systems

The increasing digitization of smart grids has improved operational efficiency but also introduced new cybersecurity vulnerabilities, such as False Data Injection Attacks (FDIAs) targeting Automatic Generation Control (AGC) systems. While machine learning (ML) and deep learning (DL) models have shown promise in detecting such attacks, their opaque decision-making limits operator trust and real-world applicability. This paper proposes a hybrid framework that integrates lightweight ML-based attack detection with natural language explanations generated by Large Language Models (LLMs). Classifiers such as LightGBM achieve up to 95.13% attack detection accuracy with only 0.004 s inference latency. Upon detecting a cyberattack, the system invokes LLMs, including GPT-3.5 Turbo, GPT-4 Turbo, and GPT-4o mini, to generate human-readable explanation of the event. Evaluated on 100 test samples, GPT-4o mini with 20-shot prompting achieved 93% accuracy in identifying the attack target, a mean absolute error of 0.075 pu in estimating attack magnitude, and 2.19 seconds mean absolute error (MAE) in estimating attack onset. These results demonstrate that the proposed framework effectively balances real-time detection with interpretable, high-fidelity explanations, addressing a critical need for actionable AI in smart grid cybersecurity.

cs.CR↗

arXiv2025

PCA-Guided Autoencoding for Structured Dimensionality Reduction in Active Infrared Thermography

Active Infrared thermography (AIRT) is a widely adopted non-destructive testing (NDT) technique for detecting subsurface anomalies in industrial components. Due to the high dimensionality of AIRT data, current approaches employ non-linear autoencoders (AEs) for dimensionality reduction. However, the latent space learned by AIRT AEs lacks structure, limiting their effectiveness in downstream defect characterization tasks. To address this limitation, this paper proposes a principal component analysis guided (PCA-guided) autoencoding framework for structured dimensionality reduction to capture intricate, non-linear features in thermographic signals while enforcing a structured latent space. A novel loss function, PCA distillation loss, is introduced to guide AIRT AEs to align the latent representation with structured PCA components while capturing the intricate, non-linear patterns in thermographic signals. To evaluate the utility of the learned, structured latent space, we propose a neural network-based evaluation metric that assesses its suitability for defect characterization. Experimental results show that the proposed PCA-guided AE outperforms state-of-the-art dimensionality reduction methods on PVC, CFRP, and PLA samples in terms of contrast, signal-to-noise ratio (SNR), and neural network-based metrics.

eess.IV↗

arXiv2025

Maximal Extractable Value in Decentralized Finance: Taxonomy, Detection, and Mitigation

Decentralized Finance (DeFi) leverages blockchain-enabled smart contracts to deliver automated and trustless financial services without the need for intermediaries. However, the public visibility of financial transactions on the blockchain can be exploited, as participants can reorder, insert, or remove transactions to extract value, often at the expense of others. This extracted value is known as the Maximal Extractable Value (MEV). MEV causes financial losses and consensus instability, disrupting the security, efficiency, and decentralization goals of the DeFi ecosystem. Therefore, it is crucial to analyze, detect, and mitigate MEV to safeguard DeFi. Our comprehensive survey offers a holistic view of the MEV landscape in the DeFi ecosystem. We present an in-depth understanding of MEV through a novel taxonomy of MEV transactions supported by real transaction examples. We perform a critical comparative analysis of various MEV detection approaches, evaluating their effectiveness in identifying different transaction types. Furthermore, we assess different categories of MEV mitigation strategies and discuss their limitations. We identify the challenges of current mitigation and detection approaches and discuss potential solutions. This survey provides valuable insights for researchers, developers, stakeholders, and policymakers, helping to curb and democratize MEV for a more secure and efficient DeFi ecosystem.

cs.CR↗

arXiv2025

Multi-Modal Attention Networks for Enhanced Segmentation and Depth Estimation of Subsurface Defects in Pulse Thermography

AI-driven pulse thermography (PT) has become a crucial tool in non-destructive testing (NDT), enabling automatic detection of hidden anomalies in various industrial components. Current state-of-the-art techniques feed segmentation and depth estimation networks compressed PT sequences using either Principal Component Analysis (PCA) or Thermographic Signal Reconstruction (TSR). However, treating these two modalities independently constrains the performance of PT inspection models as these representations possess complementary semantic features. To address this limitation, this work proposes PT-Fusion, a multi-modal attention-based fusion network that fuses both PCA and TSR modalities for defect segmentation and depth estimation of subsurface defects in PT setups. PT-Fusion introduces novel feature fusion modules, Encoder Attention Fusion Gate (EAFG) and Attention Enhanced Decoding Block (AEDB), to fuse PCA and TSR features for enhanced segmentation and depth estimation of subsurface defects. In addition, a novel data augmentation technique is proposed based on random data sampling from thermographic sequences to alleviate the scarcity of PT datasets. The proposed method is benchmarked against state-of-the-art PT inspection models, including U-Net, attention U-Net, and 3D-CNN on the Université Laval IRT-PVC dataset. The results demonstrate that PT-Fusion outperforms the aforementioned models in defect segmentation and depth estimation accuracies with a margin of 10%.

cs.CV↗

arXiv2024

Unmasking Covert Intrusions: Detection of Fault-Masking Cyberattacks on Differential Protection Systems

Line Current Differential Relays (LCDRs) are high-speed relays progressively used to protect critical transmission lines. However, LCDRs are vulnerable to cyberattacks. Fault-Masking Attacks (FMAs) are stealthy cyberattacks performed by manipulating the remote measurements of the targeted LCDR to disguise faults on the protected line. Hence, they remain undetected by this LCDR. In this paper, we propose a two-module framework to detect FMAs. The first module is a Mismatch Index (MI) developed from the protected transmission line's equivalent physical model. The MI is triggered only if there is a significant mismatch in the LCDR's local and remote measurements while the LCDR itself is untriggered, which indicates an FMA. After the MI is triggered, the second module, a neural network-based classifier, promptly confirms that the triggering event is a physical fault that lies on the line protected by the LCDR before declaring the occurrence of an FMA. The proposed framework is tested using the IEEE 39-bus benchmark system. Our simulation results confirm that the proposed framework can accurately detect FMAs on LCDRs and is not affected by normal system disturbances, variations, or measurement noise. Our experimental results using OPAL-RT's real-time simulator confirm the proposed solution's real-time performance capability.

eess.SY↗

arXiv2024

Enhancing Power Quality Event Classification with AI Transformer Models

Recently, there has been a growing interest in utilizing machine learning for accurate classification of power quality events (PQEs). However, most of these studies are performed assuming an ideal situation, while in reality, we can have measurement noise, DC offset, and variations in the voltage signal's amplitude and frequency. Building on the prior PQE classification works using deep learning, this paper proposes a deep-learning framework that leverages attention-enabled Transformers as a tool to accurately classify PQEs under the aforementioned considerations. The proposed framework can operate directly on the voltage signals with no need for a separate feature extraction or calculation phase. Our results show that the proposed framework outperforms recently proposed learning-based techniques. It can accurately classify PQEs under the aforementioned conditions with an accuracy varying between 99.81%$-$91.43% depending on the signal-to-noise ratio, DC offsets, and variations in the signal amplitude and frequency.

cs.LG↗

arXiv2024

SDXL Finetuned with LoRA for Coloring Therapy: Generating Graphic Templates Inspired by United Arab Emirates Culture

A transformative approach to mental health therapy lies at the crossroads of cultural heritage and advanced technology. This paper introduces an innovative method that fuses machine learning techniques with traditional Emirati motifs, focusing on the United Arab Emirates (UAE). We utilize the Stable Diffusion XL (SDXL) model, enhanced with Low-Rank Adaptation (LoRA), to create culturally significant coloring templates featuring Al-Sadu weaving patterns. This novel approach leverages coloring therapy for its recognized stress-relieving benefits and embeds deep cultural resonance, making it a potent tool for therapeutic intervention and cultural preservation. Specifically targeting Generalized Anxiety Disorder (GAD), our method demonstrates significant potential in reducing associated symptoms. Additionally, the paper delves into the broader implications of color and music therapy, emphasizing the importance of culturally tailored content. The technical aspects of the SDXL model and its LoRA fine-tuning showcase its capability to generate high-quality, culturally specific images. This research stands at the forefront of integrating mental wellness practices with cultural heritage, providing a groundbreaking perspective on the synergy between technology, culture, and healthcare. In future work, we aim to employ biosignals to assess the level of engagement and effectiveness of color therapy. A key focus will be to examine the impact of the Emirati heritage Al Sadu art on Emirati individuals and compare their responses with those of other nationalities. This will provide deeper insights into the cultural specificity of therapeutic interventions and further the understanding of the unique interplay between cultural identity and mental health therapy.

cs.HC↗

arXiv2023

Learning-Based Detection of Malicious Volt-VAr Control Parameters in Smart Inverters

Distributed Volt-Var Control (VVC) is a widely used control mode of smart inverters. However, necessary VVC curve parameters are remotely communicated to the smart inverter, which opens doors for cyberattacks. If the VVC curves of an inverter are maliciously manipulated, the attacked inverter's reactive power injection will oscillate, causing undesirable voltage oscillations to manifest in the distribution system, which, in turn, threatens the system's stability. In contrast with previous works that proposed methods to mitigate the oscillations after they are already present in the system, this paper presents an intrusion detection method to detect malicious VVC curves once they are communicated to the inverter. The proposed method utilizes a Multi-Layer Perceptron (MLP) that is trained on features extracted from only the local measurements of the inverter. After a smart inverter is equipped with the proposed method, any communicated VVC curve will be verified by the MLP once received. If the curve is found to be malicious, it will be rejected, thus preventing unwanted oscillations beforehand. Otherwise, legitimate curves will be permitted. The performance of the proposed scheme is verified using the 9-bus Canadian urban benchmark distribution system simulated in PSCAD/EMTDC environment. Our results show that the proposed solution can accurately detect malicious VVC curves.

eess.SY↗

arXiv2023

BaseSAP: Modular Stealth Address Protocol for Programmable Blockchains

Stealth addresses represent an approach to enhancing privacy within public and distributed blockchains, such as Ethereum and Bitcoin. Stealth address protocols generate a distinct, randomly generated address for the recipient, thereby concealing interactions between entities. In this study, we introduce BaseSAP, an autonomous base-layer protocol for embedding stealth addresses within the application layer of programmable blockchains. BaseSAP expands upon previous research to develop a modular protocol for executing unlikable transactions on public blockchains. BaseSAP allows for developing additional stealth address layers using different cryptographic algorithms on top of the primary implementation, capitalizing on its modularity. To demonstrate the effectiveness of our proposed protocol, we present simulations of an advanced Secp256k1-based dual-key stealth address protocol. This protocol is designed on top of BaseSAP and is deployed on the Goerli and Sepolia test networks as the first prototype implementation. Furthermore, we provide cost analyses and underscore potential security ramifications and attack vectors that could affect the privacy of stealth addresses. Our study reveals the flexibility of the BaseSAP protocol and offers insight into the broader implications of stealth address technology.

cs.CR↗

arXiv2023

Time to Bribe: Measuring Block Construction Market

With the emergence of Miner Extractable Value (MEV), block construction markets on blockchains have evolved into a competitive arena. Following Ethereum's transition from Proof of Work (PoW) to Proof of Stake (PoS), the Proposer Builder Separation (PBS) mechanism has emerged as the dominant force in the Ethereum block construction market. This paper presents an in-depth longitudinal study of the Ethereum block construction market, spanning from the introduction of PoS and PBS in September 2022 to May 2023. We analyze the market shares of builders and relays, their temporal changes, and the financial dynamics within the PBS system, including payments among builders and block proposers -- commonly referred to as bribes. We introduce an MEV-time law quantifying the expected MEV revenue wrt. the time elapsed since the last proposed block. We provide empirical evidence that moments of crisis (e.g. the FTX collapse, USDC stablecoin de-peg) coincide with significant spikes in MEV payments compared to the baseline. Despite the intention of the PBS architecture to enhance decentralization by separating actor roles, it remains unclear whether its design is optimal. Implicit trust assumptions and conflicts of interest may benefit particular parties and foster the need for vertical integration. MEV-Boost was explicitly designed to foster decentralization, causing the side effect of enabling risk-free sandwich extraction from unsuspecting users, potentially raising concerns for regulators.

cs.NI↗

arXiv2023

Blockchain Censorship

Permissionless blockchains promise to be resilient against censorship by a single entity. This suggests that deterministic rules, and not third-party actors, are responsible for deciding if a transaction is appended to the blockchain or not. In 2022, the U.S. Office of Foreign Assets Control (OFAC) sanctioned a Bitcoin mixer and an Ethereum application, putting the neutrality of permissionless blockchains to the test. In this paper, we formalize quantify and analyze the security impact of blockchain censorship. We start by defining censorship, followed by a quantitative assessment of current censorship practices. We find that 46% of Ethereum blocks were made by censoring actors that intend to comply with OFAC sanctions, indicating the significant impact of OFAC sanctions on the neutrality of public blockchains. We further uncover that censorship not only impacts neutrality, but also security. We show how after Ethereum's move to Proof-of-Stake (PoS) and adoption of Proposer-Builder Separation (PBS) the inclusion of censored transactions was delayed by an average of 85%. Inclusion delays compromise a transaction's security by, e.g., strengthening a sandwich adversary. Finally we prove a fundamental limitation of PoS and Proof-of-Work (PoW) protocols against censorship resilience.

cs.CR↗

arXiv2022

Towards Situational Aware Cyber-Physical Systems: A Security-Enhancing Use Case of Blockchain-based Digital Twins

The complexity of cyberattacks in Cyber-Physical Systems (CPSs) calls for a mechanism that can evaluate critical infrastructures' operational behaviour and security without affecting the operation of live systems. In this regard, Digital Twins (DTs) provide actionable insights through monitoring, simulating, predicting, and optimizing the state of CPSs. Through the use cases, including system testing and training, detecting system misconfigurations, and security testing, DTs strengthen the security of CPSs throughout the product lifecycle. However, such benefits of DTs depend on an assumption about data integrity and security. Data trustworthiness becomes more critical while integrating multiple components among different DTs owned by various stakeholders to provide an aggregated view of the complex physical system. This article envisions a blockchain-based DT framework as Trusted Twins for Securing Cyber-Physical Systems (TTS-CPS). With the automotive industry as a CPS use case, we demonstrate the viability of the TTS-CPS framework in a proof of concept. To utilize reliable system specification data for building the process knowledge of DTs, we ensure the trustworthiness of data-generating sources through integrity checking mechanisms. Additionally, the safety and security rules evaluated during simulation are stored and retrieved from the blockchain, thereby establishing more understanding and confidence in the decisions made by the underlying systems. Finally, we perform formal verification of the TTS-CPS.

cs.CR↗