A Design for Digital Twin Based IDS System Using Split Learning Predictive Maintenance in Internet of Things Related Smart Manufacturing Systems

Abstract

The Internet of Industrial Things (IIoTs) is rapidly transforming our environments and services, but its security
vulnerabilities are increasingly apparent. Device identification is crucial for protection, and implementing appropriate security measures is essential as IoT devices grow in smart cities. These vulnerabilities pose risks to urban system integrity. To address this challenge, a novel model known as the Digital Twin in IDS Using Split Learning Predictive Maintenance in IoT in Smart Manufacturing (DISLIS) model detects cyberattacks in IIoT environments using the WUSTL-IIoT-2021 dataset, which includes over a million recordings from a SCADA testbed. An Effective Feature Selection Process (EFSP) for dimensionality reduction and combines baseline and advanced models, including a Dummy Classifier, Logistic Regression, and a deep learning architecture with dropout regularization, to ensure consistent classification performance. The Improved Cyberattack Detection Process (ICDP) employs a multi-layer probabilistic model based on Restricted Boltzmann Machines to enhance detection probability and learn complex data distributions. An enhanced Dung Beetle Optimization technique fine-tunes hyperparameters for optimal accuracy. A digital Twin model is developed to provide effective predictive maintenance for the IoT environment. The anomaly detection module utilizes XGBoost, Support Vector Machines (SVM), and Deep Convolutional Neural Networks (D-CNN) for effective classification. The Random-Access Channel (RAC) study evaluates network access in 5G-enabled IIoT systems, analyzing success and collision probabilities for reliable communication. The experimental analysis is dummy classifier, XGBoost, SVM, DCNN of confusion matrix’s, DISLIS accuracy, DISLIS loss calculations.