Litcius/Paper detail

Fault Detection in Predictive Maintenance of Industry 4.0 using a Lightweight Hybrid model in the FL System

Lakshmaiya N

20256 citationsDOI

Abstract

Industry 4.0 is the revolution in the industrial setting where the sharing of information securely takes place using the decentralised system of the Federated Learning approach. Federated learning models suffer from a heavy training model that is distributed among global and local models. Predictive maintenance is a crucial application of Industry 4.0, where fault detection is the primary concern, necessitating immediate action. The efficient, lightweight models, integrated with both local and global models, enable the system to function at a faster rate. Thus, the proposed work utilises the ResNet 18, a shallower model of the residual network, for feature extraction, and the Self-attention model for classification purposes in both local and global models. The proposed method achieved 94.8% global accuracy, with effective fault detection of weld images from the Kaggle dataset. The proposed method outperforms the other FL systems by 10% compared to the standard ResNet and 17% compared to CNN models.

Topics & Concepts

Computer scienceFault detection and isolationResidualFeature (linguistics)Residual neural networkFault (geology)Artificial intelligenceData miningFunction (biology)Artificial neural networkWork (physics)Deep learningReal-time computingMachine learningReliability engineeringFeature extractionPredictive maintenanceData modelingComponent (thermodynamics)Industry 4.0Information sharingFault modelSystem modelAdvanced Neural Network ApplicationsDigital Transformation in IndustryAdversarial Robustness in Machine Learning