A fully homomorphic encryption federated learning architecture for privacy preserving in industrial internet of things

We are currently entering the fifth revolution of industry - Industry 5.0. IIoT is the domain where massive quantities of data are flourished by the associated devices in an industry on a daily basis. To realize industry 4.0, the Industrial internet of things is considered a prominent one. Federated Learning, also known as collaborative learning, employs a decentralized approach in its applicability while maintaining data privacy, but many existing frameworks struggle with handling privacy of gradients which are transferred to Federated servers. Unlike conventional approaches, proposed fully homomorphic encryption based Federated Learning-FHEEFL ensures that raw gradients never leave local IoT nodes; instead, only updates or changes in model secured with encryption techniques are transmitted. The framework achieves a very strong privacy as well as data security. FHEEFL is intended for lightweight to moderate-capacity models characteristically labouring in Edge-IIoTset analytics, where privacy guarantees must be well-adjusted with computational feasibility. While CKKS-based encrypted aggregation incurs additional overhead, the framework establishes practical applicability for privacy-critical industrial tasks under realistic resource constraints. It is proven as a privacy-centric federated learning solution, setting a new benchmark in tackling key challenges in data security and privacy. The proposed method is implemented using Edge-IIoTset dataset. The proposed fully homomorphic encryption based Federated Learning-FHEEFL method is tested in IIOT scenarios.•FHEEFL method provides better privacy with better memory usage, CPU usage and throughput parameters.•Performance is analysed with 4 variations of models- Tiny, small, medium and large.