Journal of Engineering Research and Reports

Privacy-Preserving Zero-Day Malware Detection in Encrypted Traffic Using Temporal-Structural Transformers

Michelle Raya de Luis, Zohaib Ali, Najlaa Jannah, Muhammad Arslan — Fri, 04 Sep 2026 00:00:00 +0000

End-to-end encryption protocols such as TLS 1.3 and QUIC strengthen user privacy but reduce the visibility available to conventional payload-based network-security inspection. This study presents a privacy-preserving Temporal-Structural Transformer (TST) framework for zero-day malware detection in encrypted network traffic without decrypting packet payloads. The framework combines transformer-based temporal modelling of packet sequences with graph attention mechanisms for capturing structural relationships among flows. Observable metadata, including packet size, direction, inter-arrival time, flow statistics, and available handshake attributes, are used to represent encrypted communication behavior. A contrastive-learning objective is incorporated with classification loss to improve generalization to previously unseen malware families. The framework is evaluated using multiple benchmark datasets containing TLS 1.2, TLS 1.3, and QUIC traffic, with zero-day testing, ablation analysis, robustness testing, cross-validation, and real-time performance assessment. The proposed model achieves an F1-score of 0.93 in the overall evaluation and 0.934 in the zero-day setting, with a zero-day ROC-AUC of 0.95. Ablation results indicate contributions from the temporal, graph, and contrastive-learning components, while robustness tests show limited degradation under the evaluated perturbations. Reported inference latency is below 10 ms per flow, with throughput above 10,000 flows per second. These findings support metadata-based temporal and structural modelling as a privacy-preserving approach to zero-day malware detection in encrypted traffic.

An Integrated IoT-Edge Computing Framework for Advanced Fault Diagnosis and Self-Healing in 132 kV Transmission Networks

Inyene U. Robert, Nseobong I. Okpura, Kufre M. Udofia — Fri, 11 Sep 2026 00:00:00 +0000

The increasing complexity of modern power systems and the growing penetration of renewable energy necessitate autonomous, cyber-resilient fault management to minimise outage durations. Traditional fault detection methods and centralised cloud-centric architectures suffer from high latency, communication bottlenecks, and significant cybersecurity vulnerabilities. To address these challenges, this paper proposes an integrated fault diagnosis and self-healing framework for 132 kV transmission networks utilising Internet of Things (IoT) sensors, edge computing, artificial intelligence (AI), and lightweight cybersecurity protocols. The methodology employs a discrete wavelet packet transform (DWPT) for feature extraction, paired with an 8-bit integer-quantized artificial neural network (ANN) deployed on edge devices for rapid fault classification. A low-latency fault location, isolation, and service restoration (FLISR) mechanism is orchestrated at the network edge using a mixed-integer linear programming (MILP) solver accelerated by McCormick envelopes. Furthermore, a co-optimised cybersecurity layer incorporating the Elliptic Curve Integrated Encryption Scheme (ECIES), Hash-based Message Authentication Code (HMAC), and a random forest intrusion detection system (IDS) ensures data integrity. Experimental validation on a modified IEEE 34-node test system demonstrates that the proposed secure edge framework achieves 97.3% fault classification accuracy and a total fault-to-restoration time of 198 ms, representing a 78% improvement over cloud-based architectures. The integrated security layer successfully detects 94% of false data injection attacks and 91% of replay attacks with a minimal latency overhead of 26 ms, demonstrating that robust cyber-physical protection can be achieved without compromising real-time power system protection.

Design, Simulation, and Practical Realization of a Compact 2.4 GHz Printed Circular Monopole Antenna for RF Energy Harvesting

Ahmed Abdul-Kadhem Salih, Watheq A. Neamah, Kerrar H. A. Abdulrahem — Mon, 14 Sep 2026 00:00:00 +0000

Radio-frequency energy harvesting requires a receiving antenna that combines compact dimensions with effective impedance matching at the target frequency. This study presents the design, simulation, fabrication, and experimental evaluation of a single-band printed circular monopole antenna intended for operation at 2.4 GHz within the Wi-Fi/ISM band. The antenna employs a circular radiating element, a microstrip feed line, and a partial ground plane on a CEM-1 (FR4) substrate, with overall dimensions of 70 × 50 mm². The design was modelled and evaluated using CST Microwave Studio, and its performance was assessed in terms of voltage standing wave ratio, reflection coefficient, bandwidth, surface-current distribution, radiation pattern, and gain. The simulated antenna exhibited a VSWR of 1.0027 and an S11 reflection coefficient of −31.608 dB at 2.4 GHz. The reported impedance bandwidth was approximately 188 MHz. Surface-current analysis showed strong current concentration near the feed region followed by a broadly symmetrical distribution over the circular radiator. The simulated radiation characteristics were stable in the principal planes, with a quasi-omnidirectional three-dimensional pattern. A fabricated prototype was subsequently assessed using a LiteVNA vector network analyser. The measured reflection coefficient was approximately −28 dB at the target frequency, indicating close correspondence with the simulated resonance. Overall, the results demonstrate the practical realisation of the proposed compact receiving antenna for 2.4 GHz RF energy-harvesting applications.

Finite Element Analysis of Structural Test Pedestals under Different Loading Combinations

Jianing Wang — Tue, 15 Sep 2026 00:00:00 +0000

Structural test pedestals form part of the reaction system used in large-scale structural testing, and their response may vary with loading-point arrangement and loading direction. This study compares two displacement-controlled loading combinations for an existing reinforced-concrete test pedestal using a three-dimensional finite element model comprising top and bottom slabs, columns, cross-shaped columns, loading pads, and reinforcement cages. Concrete is represented using a damage-plasticity formulation, while steel members and reinforcement are modelled as elastoplastic materials. Tie constraints are applied between concrete components and between the loading pads and top slab, with reinforcement embedded in the concrete. Both loading cases use a target displacement of 6 mm. Case 1 applies downward displacement through two pads, whereas Case 2 applies upward displacement at the middle pad and downward displacement at two side pads. The numerical response shows that Case 2 develops the higher reaction up to approximately 4.5 mm displacement, after which Case 1 becomes higher. At 6 mm, approximate reactions read from the source curve are 15 MN for Case 1 and 13 MN for Case 2. Equivalent-stress contours indicate stress concentrations around the pad load-transfer zones and adjacent connections. The comparison demonstrates that loading arrangement and direction influence the global reaction response and local load-transfer pattern.

Wavelet Transform and Artificial Neural Network-based Fault Detection and Classification for Nigerian 330 kV Transmission Network

Thu, 17 Sep 2026 00:00:00 +0000

Reliable fault detection and classification are essential for improving the security and operational stability of transmission networks. However, existing techniques often depend on effective feature extraction and may experience reduced performance under varying fault conditions. This study presents an intelligent fault detection and classification approach for the Nigerian 330 kV transmission network using the Wavelet Transform (WT) and Artificial Neural Network (ANN). The transmission system was modelled in MATLAB/Simulink R2021b, in which twelve fault conditions were simulated at different fault locations and fault resistances. The Daubechies-4 (db4) wavelet was employed to extract maximum current coefficients, which served as input features for ANN training and testing. A total of 372 fault and no-fault datasets were generated, with 279, 37, and 56 samples used for training, testing, and validation, respectively. The fault detection model (4-10-1-1) achieved a best validation mean squared error (MSE) of 2.5579 × 10⁻¹¹ with a regression coefficient of R = 1.0000, while the fault classification model (4-20-4-4) produced a best validation MSE of 2.3021 × 10⁻⁴ and regression coefficients of 0.99945, 0.99905, and 0.99953 for the training, testing, and validation datasets, respectively. These results demonstrate that the proposed WT-ANN framework provides fast, accurate, and reliable fault detection and classification, making it a promising intelligent technique for modern transmission line protection.

Deformation Characteristics of a Long Deep Metro Excavation in Ningbo Soft Clay Using the HSS Model

Ying Qiao — Thu, 17 Sep 2026 00:00:00 +0000

Huishi Road Station on Ningbo Metro Line 6 was investigated to assess the deformation behaviour of a long, deep excavation in Ningbo soft clay. A three-dimensional PLAXIS model was developed to simulate staged excavation, diaphragm walls, reinforced-concrete struts, and bottom-soil improvement. Soil behaviour was represented using the Hardening Soil model with small-strain stiffness (HSS), and numerical predictions were compared with wall-inclinometer and ground-settlement monitoring. The measured wall-deflection profiles were generally bow-shaped, and deformation was greater in the middle of the long side than near the excavation ends, indicating a marked spatial effect. At the final excavation stage, maximum wall deflection at 48 inclinometer locations ranged from 0.23%H to 1.01%H, with a mean of 0.62%H. The model predicted a maximum wall displacement of 132.5 mm, approximately 0.8%H, while the mean absolute error for four representative monitoring sections was 12.3%. The maximum measured ground settlement was approximately 0.16%H, with an influence range of about 3H. Numerical settlement was larger and extended to approximately 3.5H. The results indicate that the adopted HSS parameters reproduce the principal wall-deformation pattern reasonably well, whereas settlement predictions are more conservative. Accordingly, wall deflection is suitable as a primary calibration quantity, while predicted settlement can provide a conservative reference for construction-stage deformation assessment in comparable Ningbo soft-clay excavations.

Welding Technologies for Metallic Bipolar Plates in Proton Exchange Membrane Fuel Cells: A Critical Review of Processes, Joint Performance, and Manufacturing Readiness

Yueyang Liu — Mon, 07 Sep 2026 00:00:00 +0000

Metallic bipolar plates (BPPs) enable thin, lightweight, and high-throughput proton exchange membrane fuel cell (PEMFC) stacks, but their industrial value depends on joining two formed foils without sacrificing gas tightness, dimensional accuracy, electrical performance, or corrosion durability. This article presents a critical narrative review of welding and related joining technologies for metallic BPPs. Literature was evaluated through 31 July 2026 and interpreted using three evidence levels: direct validation on BPP assemblies, BPP-relevant thin-foil analogue studies, and transferable evidence from general microjoining. Laser welding has the strongest combined evidence because it offers localised heat input, high path flexibility, and compatibility with automation; however, its process window is narrowed by foil thickness, inter-sheet gap, variable heat sinking, coating condition, and high-speed melt-flow instabilities. Resistance seam welding provides direct evidence of leak-tight stainless-steel BPP joining and may be attractive where two-sided electrode access is available, although electrode wear, indentation, and complex-path capability require closer attention. Published BPP-specific validation of tungsten inert gas/microarc, electron-beam, ultrasonic, friction-stir-based, brazing, and adhesive routes remains insufficient for claims of high-volume readiness. Across all processes, the most consequential unresolved issues are the absence of harmonised seam-leak qualification, limited long-term fatigue and weld-zone corrosion data, uncertain welding-coating process sequences, and the lack of representative public datasets for in-line quality prediction. A qualification framework is proposed that links process parameters and weld morphology to mechanical, sealing, electrical, electrochemical, and durability outcomes. The review concludes that future progress depends less on isolated peak weld strength than on full-plate robustness, traceable defect detection, coating-aware process integration, and statistically demonstrated manufacturing capability.

Piping Systems Design and Analysis: A Review of Codes, Computational Methods and Emerging Digital Practice

Tue, 08 Sep 2026 00:00:00 +0000

Piping systems are used to transport fluids in power generation, the oil and gas sector, chemical processing and manufacturing industries, and water-delivery facilities. The design of these systems has a substantial influence on the safety and efficiency of industrial operations. This review brings together established piping design and analysis principles with research from the last five years on computational tools, monitoring technologies, and materials. Design standards, their application and accessibility, and flexibility and stress requirements are examined. The review highlights four major areas that are changing the field: the use of algorithms for improved route planning, digital twins integrated with real-time sensor data, machine learning for predicting corrosion and failures, and material challenges associated with composites and the use of steels in hydrogen pipelines. Recent studies show that conventional stress-analysis software can determine whether piping systems satisfy safety requirements, while experiments on components such as bellows expansion joints can improve understanding of their stress and strain capacity during design. The review also examines how machine learning is used to improve piping-condition prediction from available data. Research further indicates that hydrogen embrittlement requires careful monitoring and assessment before existing pipelines are repurposed. The novelty of this review lies in integrating established piping design and analysis codes and principles with emerging digital engineering practices, including computational tools, real-time monitoring, machine learning, digital twins, and advanced materials for hydrogen service. It addresses the fragmented nature of the literature and highlights practical challenges related to cost, technical skills, and data access, particularly in smaller engineering organisations.

Spare Parts and Materials Standardisation in Pressure Vessel Maintenance: A Critical Review of the Evidence for Inventory Cost Reduction

Henry Chukwuemeka Olisakwe, Nwakamma Ninduwezuor-Ehiobu, Paul Chibuike Obetta — Fri, 11 Sep 2026 00:00:00 +0000

Pressure vessels are among the most capital-intensive and safety-critical assets in process, energy and marine industries, and the materials and components consumed in maintaining them account for a substantial and persistent share of operating expenditure. Standardisation of spare parts and construction materials is widely promoted as a route to lower inventory cost, yet the supporting evidence is distributed across separate research communities that rarely address one another. This critical narrative review examines whether, and under what conditions, standardisation reduces the cost of holding and replenishing pressure vessel maintenance materials without eroding integrity assurance. Literature was identified through structured searching of open scholarly indexes, supplemented by backward and forward citation tracking and by consultation of authoritative institutional documents, with a defined search period closing on 3 July 2026. Sources were appraised for design adequacy, transparency, external validity and the alignment between reported findings and the claims they are used to support. The synthesis develops five layers of standardisation, spanning design, materials, components, data and process, and traces the causal pathways by which each is proposed to influence inventory cost. The evidence is strongest for data standardisation and for multi-criteria classification linked to stocking policy, where independent case studies report inventory value reductions of roughly fifteen to seventy per cent alongside maintained or improved availability. Evidence for materials standardisation in pressure-retaining service is markedly weaker, and is constrained by divergent allowable-stress bases between major construction codes, by damage mechanisms that are specific to alloy and environment, and by the absence of studies that quantify inventory consequences of material rationalisation decisions. Claims that additive manufacturing dissolves physical inventory are only partially supported; the most rigorous process-industry assessment identifies economic viability for approximately one per cent of stock keeping units. Principal unresolved questions concern the safety cost of substitution, the transferability of commonality models to code-governed components, and the near-total absence of longitudinal cost evidence from pressure vessel populations. Standardisation is best understood as a conditional lever whose returns depend on data quality, criticality structure and regulatory constraint rather than as a general economy.

Artificial Intelligence-Based Alzheimer’s Disease Classification Using Brain MRI: A Review of Machine Learning, Deep Learning and Emerging Intelligent Approaches

S. Deepa, A. Davincy Merline Sharmya, M. Gnanaprakash, D. Periyaazhagar — Fri, 11 Sep 2026 00:00:00 +0000

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that affects memory, reasoning, language, and other cognitive functions. Early identification is important because structural brain changes may occur before severe clinical symptoms become evident. Magnetic Resonance Imaging (MRI) provides detailed anatomical information and has therefore become an important source for computer-aided Alzheimer’s disease assessment. Artificial intelligence (AI) techniques have increasingly been applied to MRI-based classification, progressing from handcrafted features and conventional machine-learning methods to deep-learning approaches. Traditional methods use morphological, textural, statistical, and frequency-domain features with classifiers such as Support Vector Machine, Random Forest, and Artificial Neural Networks. Deep-learning models, including Convolutional Neural Networks (CNNs), three-dimensional CNNs, transfer-learning architectures, ResNet, DenseNet, and EfficientNet, can automatically learn disease-related image representations. Recent approaches, including Vision Transformers, Explainable Artificial Intelligence (XAI), multimodal learning, federated learning, and self-supervised learning, aim to improve global feature representation, interpretability, data utilisation, privacy, and generalisation. This review examines the development of AI techniques for MRI-based Alzheimer’s disease classification, including datasets, preprocessing, feature extraction, deep learning, and emerging methods. A comparative discussion highlights their strengths and limitations, while major challenges such as limited datasets, class imbalance, scanner variability, computational complexity, and insufficient external validation are discussed. Future research should focus on robust, explainable, multimodal, privacy-aware, and clinically validated AI systems.

Counterfactual Explanations for Sequential Purchase Decision Modeling in E-Commerce: A Systematic Survey

Jerome Jayanathan Needhipathi — Sat, 12 Sep 2026 00:00:00 +0000

The growing complexity of e-commerce platforms necessitates recommendation systems that provide high-quality predictions while maintaining transparent and interpretable algorithms with opportunities for recourse. As users proceed through different stages of sequential interaction, from browsing and comparison to adding products to a basket and finally proceeding to checkout, the interpretation of the sequential decision-making process becomes more complicated than the interpretation of static recommendations. Based on a qualitative literature-selection approach applied to key electronic databases, this paper provides a survey of the existing primary research in the domains of XAI, causal reasoning, sequential recommendation, and algorithmic recourse for sequential purchase decisions in e-commerce. A taxonomy of the existing literature in the domain of interest is provided into five main approaches: optimization, search, generative, sequential decision, and causal relationships. The key constraints in e-commerce that an explanation technique must satisfy for successful practical application are analysed: temporal non-reversibility, changing catalogue, multi-level behaviour hierarchy, alignment of users' short- and long-term intentions, and KPIs. The comparative analysis shows that no existing paradigm meets all domain constraints. A conceptual model of the hybrid counterfactual reasoning system is proposed and further research directions are highlighted: ultra-low latency counterfactual inference, integration with LLMs, cold-start counterfactual recourse, and counterfactual robustness under concept drift.

On-Device Privacy and Zero-Knowledge Storage in Mobile Applications: A Review with Attention to Cross-Platform React Native Development

Tue, 15 Sep 2026 00:00:00 +0000

Mobile applications hold the most intimate data that most people generate—messages, health signals, finances, locations, and journals—and the architectural question of where such data live in plaintext has become the defining privacy decision of app design. On-device privacy names one pole of the answer: data is processed, stored, and protected where it originates, and services are designed so that servers hold ciphertext they cannot read, an architecture popularly called zero-knowledge storage. This review synthesizes the research literature relevant to building such applications, with specific attention to the cross-platform reality in which much of the mobile ecosystem, prominently React Native applications, is actually built. It first fixes terminology, distinguishing the cryptographic meaning of zero knowledge from the architectural marketing usage, and organizes the threat models—curious servers, device compromise, network adversaries, and legal compulsion—against which designs are meaningfully compared. It then reviews the platform security foundations, hardware-backed keystores, trusted execution, biometric gating, and their documented limits; encrypted local storage, from key derivation through database encryption to searchable and oblivious techniques and their leakage trade-offs; end-to-end encrypted synchronization and the protocol lineage that made forward secrecy deployable; and on-device privacy-preserving computation, including local inference, differential privacy, and federated learning, with the attack literature that bounds their guarantees. The cross-platform layer receives dedicated treatment: the JavaScript runtime, bridge architecture, dependency supply chain, and the misuse findings of the mobile security measurement literature as they bear on React Native designs. The review closes with the recurring gaps—usable key recovery, metadata protection, and verification of zero-knowledge claims—and identifies directions for research and practice.

Automated Validation in Embedded IoT Development: A Review of Continuous Integration and Delivery across the Fidelity Ladder

Fri, 18 Sep 2026 00:00:00 +0000

Embedded software for the Internet of Things is developed under pressures that pull in opposite directions: the delivery cadence of connected products, sustained by over-the-air update channels, demands the automation of continuous integration and delivery, while the software's defining property—correct behaviour on constrained hardware interacting with physical signals and unreliable networks—is exactly what conventional automated pipelines exercise least. A substantial literature, scattered across software engineering, embedded systems, testing, and security venues, now addresses this tension, and this paper reviews it. The review organises automated validation for embedded IoT development as a fidelity ladder that pipelines climb: static analysis and compilation-time checking, whose industrial-scale evidence is strong and whose embedded-specific value lies in catching the defect classes hardware testing finds expensively; host-based unit and component testing with hardware abstraction, where adequacy measurement and property-based techniques transfer directly; simulation and software-in-the-loop validation, including digital-twin approaches that make plant models maintained assets; emulation and firmware re-hosting, the enabling layer for dynamic analysis and fuzzing of embedded binaries, whose distinctive failure modes the security literature has characterised; hardware-in-the-loop testing, the automotive-proven apex whose integration into continuous pipelines the empirical literature identifies as the field's persistent gap; and field-level validation through staged rollout and telemetry against fleets. Cross-cutting concerns are reviewed in depth: test selection and prioritisation under slow suites, flaky-test management under physical variability, environment-as-code for reproducible test infrastructure, model-based test generation, fault injection for dependability claims, and the validation of the update channel itself. The review weighs the empirical evidence on continuous practices in embedded organisations, identifies areas where evidence is thin—benchmarks, cross-level coverage reasoning, and pipeline-scale evaluation—and closes with directions for research and practice.