Intelligent Digital Twin Framework for Real-Time Structural Health Monitoring and Optimization of Mechanical Systems Using Reinforcement Learning

¹ Assoc. Prof., Dept. of CSE (Data Science), SVCE, Bengaluru, India, This email address is being protected from spambots. You need JavaScript enabled to view it.
² Asst. Prof., School of Electronics and Communication, REVA university, This email address is being protected from spambots. You need JavaScript enabled to view it.
³ Assoc. Prof., Dept. of Mathematics, Sapthagiri NPS University, Bangalore, This email address is being protected from spambots. You need JavaScript enabled to view it.
⁴ Project Cybersecurity Manager, Harman International, Bangalore, This email address is being protected from spambots. You need JavaScript enabled to view it.
⁵ Asst. Prof., Dept. of EEE, Dayananda Sagar College of Engineering, Bangalore, This email address is being protected from spambots. You need JavaScript enabled to view it.
⁶ Dept. of CSE (Data Science), SVCE, Bengaluru, India, This email address is being protected from spambots. You need JavaScript enabled to view it.

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

This paper presents an Intelligent Digital Twin (IDT) framework designed for real-time Structural Health Monitoring (SHM) and optimization of mechanical systems. The proposed approach integrates high-frequency sensor data with a physics-based digital model to provide accurate, continuous assessment of structural integrity. Sensor inputs from vibration, strain, and temperature measurements are preprocessed and transformed into damage-sensitive features, which are assimilated into a high-fidelity finite element digital twin using Kalman filtering for precise state estimation. An anomaly detection module evaluates residuals between measured and predicted responses to identify potential faults, while a reinforcement learning (RL) agent operates within the updated digital twin to learn optimal maintenance and control strategies that minimize structural degradation and operational costs. The framework is implemented and tested on a scaled mechanical testbed subjected to dynamic loading. Experimental results demonstrate significant performance improvements compared to conventional SHM methods, achieving 96.8% detection accuracy, 95.6% precision, 96.1% recall, and a 14.7% optimization gain, along with reduced response latency of 135 ms. These outcomes highlight the effectiveness of combining digital twin technology with RL-driven decision-making to create adaptive, proactive, and efficient SHM systems suitable for long-term industrial deployment.