Adversarially Robust Deep Learning Models for Secure Cyber-Physical Systems: A Multi-Layer Defense Framework for Real-Time Threat Mitigation
The Critical Review of Social Sciences Studies · 19.02.2026 · 4 (1) · 1601-1614
DOI: 10.59075/6g3beg35 ↗Kurz gefasst
Fachartikel in der Zeitschrift The Critical Review of Social Sciences Studies, online erschienen am 19.02.2026. Verfasst von Sakeena Parveen und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Cyber-physical systems (CPS) are increasingly dependent on deep learning models for real-time monitoring, control, and decision-making, making them vulnerable to adversarial attacks that can compromise both computational and physical operations. This study investigates the development of adversarially robust deep learning models integrated within a multi-layer defense framework to enhance security and ensure real-time threat mitigation in CPS. A quantitative experimental approach was employed, using Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks trained on clean and adversarially perturbed datasets generated using Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD). Data preprocessing, feature extraction, and adversarial training were combined with system-level defenses such as secure communication, redundancy mechanisms, and anomaly detection to form a comprehensive multi-layer framework. Statistical analysis, including ANOVA and regression, was conducted to evaluate performance and robustness across various experimental setups. Results indicate that standard CNN and LSTM models achieved high accuracy on clean data (CNN: 96.2%, LSTM: 94.7%) but were highly vulnerable to adversarial attacks (CNN: 68.5%, LSTM: 86.1%). Incorporating adversarial training and multi-layer defenses improved robustness significantly, with accuracy above 90%, F1-scores above 0.90, detection rates up to 96.1%, and latency maintained below 30 ms, demonstrating reliable real-time protection. These findings confirm that a holistic, layered security approach effectively mitigates adversarial threats, preserves operational integrity, and supports the deployment of trustworthy AI-driven CPS.
Zitieren
APA
Parveen, S., Waqar, F., Zahra, K., & Ali, A. (2026). Adversarially Robust Deep Learning Models for Secure Cyber-Physical Systems: A Multi-Layer Defense Framework for Real-Time Threat Mitigation. The Critical Review of Social Sciences Studies, 4(1), 1601-1614. https://doi.org/10.59075/6g3beg35
Vancouver
Parveen S, Waqar F, Zahra K, Ali A. Adversarially Robust Deep Learning Models for Secure Cyber-Physical Systems: A Multi-Layer Defense Framework for Real-Time Threat Mitigation. The Critical Review of Social Sciences Studies. 2026;4(1):1601-1614. doi:10.59075/6g3beg35
BibTeX
@article{Parveen2026,
author = {Parveen, Sakeena and Waqar, Farwa and Zahra, Kanza and Ali, Afrasiyab},
title = {Adversarially Robust Deep Learning Models for Secure Cyber-Physical Systems: A Multi-Layer Defense Framework for Real-Time Threat Mitigation},
journal = {The Critical Review of Social Sciences Studies},
year = {2026},
volume = {4},
number = {1},
pages = {1601--1614},
doi = {10.59075/6g3beg35}
}
EUVIMED