Industrial Digital Twins (DTs) increasingly rely on on-device learning to keep virtual replicas synchronized with heterogeneous physical assets. Severe client heterogeneity and tight edge budgets challenge standard federated learning (FL), motivating Clustered Federated Learning (CFL) that maintains one model per group of similar devices. This paper presents a laptop-runnable, reproducible benchmark that fairly compares six CFL algorithms (IFCA, FedClust, FedLC, DBSCAN-FL, K-Means-FL, Hierarchical-FL) under identical settings. Beyond accuracy, we log clustering quality, convergence, and system cost (communication volume, runtime) to reflect DT constraints. Using non-IID Fashion-MNIST as a lightweight stand-in for vision/inspection tasks, we find DBSCAN-FL provides the best balance: best accuracy (85.50%±0.85), fast convergence (14.4± 6.9 rounds), and low communication (600.6±97.6 MB), while IFCA underperforms across metrics. We discuss how resource-aware CFL can partition fleets by process, line, or usage profile to improve twin fidelity under tight bandwidth and latency budgets.

IoRT systems, which operate with microcontroller-class processors and limited memory (≤ 2MB), require real-time DDoS detection. However, existing federated learning approaches cannot simultaneously achieve the necessary model compression, DP, and communication efficiency for resource-constrained deployments. This paper presents a novel federated learning framework that resolves these conflicting requirements through an integrated QAT approach with DP-based clustering.Our approach achieves 4.0x model compression (0.52MB to 0.13MB) with only 0.79% accuracy degradation compared to uncompressed federated baselines, while maintaining F1-scores (0.998). Evaluation against methods such as FedProx, DeepShield, and AUWPAE on CICIoT2023 demonstrates the advantage of our approach: 433x faster inference than centralized ensemble methods (0.15ms vs. 65ms) with a 75% reduction in federated communication overhead. The DP_Clustering successfully handles non-IID data heterogeneity while providing formal privacy guarantees unavailable in centralized approaches.

Federated learning (FL) enables collaborative training on decentralized data, but sharing model updates exposes clients to inference attacks. We introduce a lightweight, open benchmark for an apples-to-apples comparison of orthogonal privacy-enhancing technologies (PETs) under identical, resource-constrained conditions. While our framework supports homomorphic encryption (HE), this study focuses empirically on sample-level differential privacy (DP) and secure aggregation (SA) on a non-IID Fashion-MNIST setup with a shared CNN and optimizer.Across identical settings, we log model utility, communication, computation, and formal privacy. We observe a clear trade-off: the non-private FedAvg baseline reaches 86.50 % test accuracy in 6 rounds to convergence and 202.00 MB total communication. DP (final (ε, δ) = (1.86, 10−5)) drops to 76.30 % accuracy, requires 120 rounds to convergence, and transmits 4.05 GB. SA preserves high utility (82.50 %) with 7 rounds to convergence and 124.00 MB total traffic, adding only modest cryptographic overhead while shielding individual updates from a curious server. Taken together, our unified benchmark and results pro-vide actionable guidance for privacy choices in cyber-physical deployments where latency, bandwidth, and energy are tight.

This paper presents a framework for securing blockchain-based IoT systems by integrating Physical Unclonable Functions (PUFs) and Zero-Knowledge Proofs (ZKPs) within a Hyperledger Fabric environment. Our approach leverages PUFs for robust device authentication and ZKPs for privacy-preserving transaction processing, addressing key challenges of security, privacy, and scalability in IoT systems. The framework’s architecture utilizes Hyperledger Fabric’s modular design and private channels to enhance scalability. Off-chain experimental results demonstrate the framework’s feasibility, with compact proof sizes (median 805 bytes) and efficient processing times (average 2,800 ms end-to-end). A comprehensive security analysis shows the framework’s resilience against various attacks, including device impersonation and data tampering. This work provides a foundation for secure and scalable blockchain-based IoT systems, with directions for future on-chain implementation and optimization for resource-constrained devices.

This paper addresses the challenge of Distributed Denial of Service (DDoS) attacks in the Internet of Robotic Things (IoRT) using a federated learning approach. We investigate the performance of Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRUs) for DDoS detection in IoRT systems. Our models are evaluated using the CICDDoS2019 dataset. The CNN-based model achieves the highest performance with an accuracy of 0.9810 and an F1-score of 0.9800, outperforming LSTM and GRU-based models. We analyze the models’ convergence properties and discuss their suitability for resource-constrained IoRT devices. Our results demonstrate the potential of federated learning for enhancing IoRT security while highlighting the trade-offs between model performance and efficiency.

In this work-in-progress study, we describe a development of an instrument aimed at capturing engineering faculty's perceptions of and confidence in teaching laboratory intensive online courses. Whether they want to or not, engineering faculty are being asked to design and deliver laboratory intensive courses remotely. Among the reasons are the most recent Covid-19 pandemic and the increasing number of programs and degrees offered in distance or on the Internet. A typical engineering education laboratory environment has been substantially evolved because of the advancements in communication technologies. Virtual realities are the most recent inventions where students and teachers can interact with one another similar to how they would interact in physical laboratory environments. Most of the tools and machines used in the laboratories will soon be simulated in those environments. It is not well studied and documented in the literature what the engineering faculty perceive about the quality of their teaching and what their self-efficacies in designing laboratory intensive online courses are. After the Covid-19 pandemic lock-down, majority of the engineering faculty have experienced teaching lab-intensive courses online. Those experiences led the engineering faculty develop some perceptions about the feasibility or the quality of their teaching, and beliefs about their capabilities and skills to offer the courses. To capture those perceptions and capabilities, a valid and reliable instrument should be developed. In this paper, we report and discuss the findings from a study aimed at developing a valid and reliable instrument to capture engineering faculty's perception of and self-efficacy in teaching laboratory intensive online courses. The survey items were specifically written to capture engineering faculty members' experiences with teaching laboratory intensive online courses and their beliefs and confidence in teaching them effectively. The newly designed instrument was administered to a group of faculty on campus. We ran tests to compare the faculty's demographics and their average survey scores and reported the significant findings.

Adversarial Training is a proven defense strategy against adversarial malware. However, generating adversarial malware samples for this type of training presents a challenge because the resulting adversarial malware needs to remain evasive and functional. This work proposes an attack framework, EGAN, to address this limitation. EGAN leverages an Evolution Strategy and Generative A dversarial Network to select a sequence of attack actions that can mutate a Ransonware file while preserving its original functionality. We tested this framework on popular AI-powered commercial antivirus systems listed on VirusTotal and demonstrated that our framework is capable of bypassing the majority of these systems. Moreover, we evaluated whether the EGAN attack framework can evade other commercial non-AI antivirus solutions. Our results indicate that the adversarial ransonware generated can increase the probability of evading some of them.

This work introduces a novel security framework for wireless sensor networks (WSN) based on dynamic duty cycle, which allows nodes to detect their compromised neighbors based on unanticipated fluctuations in network traffic send rate over time. Our framework was assessed by its ability to detect advanced WSN threats (e.g., active, passive, or both attacks). One of the benefits of this framework is that it reduces all threats to unanticipated power dissipation. In other words, the framework assumes any neighbor not conforming to predicted power levels has been communicating with an unauthorized node, and thus is compromised. This threat model is emulated by applying pseudo random but bound (large to small) power dissipations to arbitrary nodes. Simulation results demonstrated that this framework was effective in detecting and isolating compromised sensor nodes.

In recent years, machine-to-machine (M2M) networks, which do not require direct human intervention, is increasing at a rapid pace. Meanwhile, the need of a wireless platform with a vast coverage and low network deployment cost for controlling and monitoring these M2M networks has not yet been met. Mobile cellular networks (MCNs) and wireless sensor networks (WSNs) are emerging as two heterogeneous networks that can meet the challenges of M2M communication through network convergence. In this paper, we propose a model for network convergence between a Long Term Evolution-Advance (LTE-A) cellular network and a wireless sensor network. Quality of service (QoS) issues are assessed by a comparative study of the network delay in tight coupling and loose coupling LTE-A configurations. Simulation results indicate that the network delay in our proposed converged network is acceptable for various M2M applications.

Using a reputation-based trust framework for wireless sensor networks we introduce a mechanism that prevents the election of compromised or malicious nodes as cluster heads, through trust based decision making. We employ a secure cluster formation algorithm to facilitate the establishment of trusted clusters via pre-distributed keys. Reputation and trust is built over time and allow the continuation of trusted cluster heads elections. We performed an evaluation of our approach through simulations. The results indicate clear advantages of our approach in protecting the information of our network by preventing the election of untrustworthy cluster heads.

Cooperation among nodes is essential for the reliable routing of packets in large scale wireless sensor networks from nodes to base station. Most of the previous works have assumed a single governing authority with full cooperation among nodes. The assumption of node cooperation, however, cannot be applied to wireless sensor networks (WSNs) with more than one governing authority. In this paper, we introduce the concept of multi-class wireless sensor networks where each class is governed by a different authority. We study the evolution of cooperation in static and mobile multi-class wireless sensor networks using evolutionary game theory which has, to the best of our knowledge, never been attempted before. We then propose a novel localized distributive algorithm we call the patient grim strategy (PGS), and demonstrate that it provides a Nash equilibrium solution to the game theoretic problem of cooperation in multi-class static wireless sensor networks. Our simulation results show that in static multi-class WSNs populations playing the prisoner's dilemma, significant propensities to cooperate can evolve.