Physics-Informed Graph Digital Twin with Reinforcement Learning and Sensor Data Fusion for Predictive Maintenance Optimization in Smart Manufacturing

View at publisher Summarise with VIRAT
Authors: Md Mahamudur Rahaman Shamim, Md. Nuruzzaman, M. Muklasur Rahman Opu, Mohammad Mostafijur Rahman, Md. Rajib Ahmed, Sudoy Kumer Ghosh
Cited by: 0
References: 20

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 paper introduces a new Physics-Informed Graph Digital Twin framework combining the graph neural networks (GNNs) and the reinforcement learning (RL) to discover intelligent maintenance scheduling in the manufacture of electrical equipment. An operation of the system is as follows: the system trains complex interdependencies among assets using multi-sensor data, such as MachineID, TempC, Vibrationmms, PressureBar, AcousticdB, InspectionDurationmin, DowntimeCostUSD, TechnicianAvailabilitypct, FailureProb, and MaintenancePriority. The prediction of the trained model is used to direct a reinforcement learning scheduler to optimize periods of inspection and allocation of resources that brings significant changes in the accuracy of maintenance (88.18) and the cost of downtime (50.7) improvements. Learned graph structures, uncertainty maps and reward trajectories can be visualized, and they are interpretable and operationally reliable. The proposed hybrid framework integrates data-driven intelligence and physical consistency that provides an expandable and understandable method of real-time decisionmaking in the industrial IoT and edge-enabled smart factory settings.

Fields of Study & Subjects

Digital Transformation in Industry Reliability and Maintenance Optimization Machine Fault Diagnosis Techniques

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
18–24
Publication date
21 January 2026
Language
EN