Automated deep learning-based tailoring of error detectors for digital twin-aided cyber-physical systems

Read full text Summarise with VIRAT
Authors: Tagir Fabarisov
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This dissertation presents a methodology for tailoring Deep Learning-based Anomaly Detection (AD) systems to specific Cyber-Physical Systems (CPS) by leveraging their digital twins. In contrast to system-agnostic approaches, the framework integrates system-specific knowledge at each stage of an iterative process. The methodology consists of three interconnected phases: (i) identification of critical faults, (ii) optimization of data access points, and (iii) deployment of error detection models. Fault injection experiments and reinforcement learning guide each phase. By tailoring the AD system to the target CPS, the approach addresses high dimensionality, multi-modality, and complex fault interactions while improving computational efficiency and robustness. The effectiveness of the proposed methodology was first demonstrated on a MATLAB/Simulink-based reference model emulating an engine system in closed-loop control. This controlled environment enabled validation of each phase, showing significant improvements in anomaly detection precision and fault coverage compared to system-agnostic approaches. The methodology was then applied to a real-world exoskeleton CPS equipped with a digital twin. The AD system was tailored to the hardware architecture, operational constraints, and critical fault parameters of the exoskeleton. Comprehensive evaluations demonstrated improved computational efficiency while maintaining detection accuracy. The reinforcement learning-driven fault injection framework and optimized access point selection reduced data requirements without compromising detection performance. The results confirm the effectiveness of the methodology in addressing system-specific CPS challenges. This dissertation provides a structured framework for designing robust and scalable AD systems tailored to diverse CPS applications.

Fields of Study & Subjects

Fault Detection and Control Systems Anomaly Detection Techniques and Applications Software System Performance and Reliability

Publication details

Publication date
1 January 2026
Language
EN
Document type
Thesis