A Secure and Verifiable Federated Learning Framework for Industrial Internet Systems

Abstract

The Industrial Internet of Things (IIoT) generates massive data for intelligent decisions like predictive maintenance in smart manufacturing. However, this data often contains sensitive information, requiring robust privacy protection under regulations such as GDPR and ISO/IEC 29100. Centralized training transmits raw data, incurring heavy communication overhead-conflicting with limited IIoT bandwidth-and increasing breach risks. Federated Learning (FL), where “models move, data remains local,” suits IIoT collaborative training. Yet existing privacypreserving FL (PPFL) schemes have shortcomings. Plain homomorphic encryption (PHE) lacks flexibility and incurs high overhead. Multi-key homomorphic encryption (MKHE) partially addresses these issues but fails to resolve ciphertext expansion and communication costs. Neither approach handles dynamic device membership, common in IIoT. Solutions integrating both privacy and verifiable aggregation remain scarce in dynamic settings. To address these challenges, this paper proposes a tailored FL framework with three key innovations: 1)Optimized MKHE: Reduces overhead via parameter pruning. 2)Dynamic membership management: Enables seamless device join and exit. 3)HS-FC integration: Achieves constant-time verifiable aggregation.