علي اثني عشري

Cyber-Physical Systems (CPSs) represent the core driving force of the Fourth Industrial Revolution. However, many of these systems, such as autonomous vehicles, are inherently safety-critical, requiring exceptionally high levels of system reliability an‎d fault tolerance. Traditional fault-tolerance mechanisms, which predominantly rely on hardware redundancy, address faults by incorporating backup replicas of physical components. However, this conventional approach inevitably increases hardware cost, energy consumption, an‎d system footprint. Furthermore, as CPSs evolve toward higher levels of complexity an‎d integration, such hardware-centric redundancy mechanisms exacerbate power density an‎d thermal stress, ultimately degrading system reliability. To address these limitations, Digital Twin (DT) technology has emerged as a transformative paradigm. At its core, this paradigm replaces physical backup units with precise software-based models, thereby transitioning fault-tolerant CPS design from a static, hardware-dependent approach to a dynamic, knowledge-driven one, where reliability is ensured through intelligent data analytics rather than physical duplication. This research introduces a comprehensive two-stage framework, termed PhyTwin, to realize this vision. In the first stage, an intelligent DT component is developed based on Physics-Informed Neural Networks (PINNs). By fusing fundamental physical laws with deep learning capabilities, the proposed model enables accurate reconstruction of faulty sensor signals. Experimental results obtained from an autonomous vehicle case study confirm the superior accuracy an‎d robustness of the physics-informed model compared to a conventional neural network across all fault-injection scenarios. In the second stage, a semi-optimal orchestration platform is introduced, employing a Genetic Algorithm (GA) to manage the replacement of hardware redundancies with digital twins. By formulating an‎d solving a multi-objective optimization problem, the platform identifies the optimal configuration that simultaneously maximizes system reliability an‎d minimizes energy consumption. eva‎luation results demonstrate that the proposed PhyTwin platform not only enhances system reliability over the fully hardware-based baseline but also reduces energy consumption by up to 50% in the best case an‎d 21% in the worst case.

سامانه هاي رايا-فيزيكي , همزاد ديجيتال , بهبود قابليت اطمينان , كاهش مصرف انرژي , شبكه هاي عصبي آگاه از فيزيك

Cyber-Physical Systems (CPSs) , Digital Twin , Reliability Improvement , Energy Consumption Reduction , Physics-Informed Neural Networks (PINNs)

Ali Asnaashari

Dr. Amir Mahdi Hosseini Monazzah