3DFed: A Secure Federated Learning-based System for Fault Detection in 3D Printer Industry

Abstract

This paper proposes a secure federated learning (FL) approach for 3D printer fault detection, namely 3DFed. Most current 3D fault detection systems were developed with a centralized learning approach, which is less efficient for large-scale deployment due to limited data for training. The FL-based system could be exploited to further increase fault detection accuracy while maintaining high performance by using the FedAvg algorithm. The 2D convolutional neural network (CNN) was used to extract data features from an image array. To further improve security in the simulation work, a certificate authority (CA) was added to maintain secure communication between the FL server and clients. The suggested 3DFed with the proposed CNN-based model can deliver high classification accuracy while preserving minimum time-cost, according to a thorough performance evaluation. Also covered in depth is how total client variance affects the learning process.