Volume 35, Issue 1, 2026
DOI: 10.53555/03276716.2026.01
FLASH-BPH: A Framework For Scalable And Trustworthy Human-Centric Cyber-Physical Systems
Abstract
Industry 4.0's rapid development has brought into sharp focus the critical need to balance cyber-physical systems (CPS) efficacy with potential risks of human errors. This work introduces FLASH-BPH, a human-centric framework aimed at integrating automation with human supervision while maintaining privacy, scalability, and ethical standards. FLASH-BPH incorporates Human-in-the-Loop (HITL) validation as a fundamental component, allowing domain experts to assess and enhance AI-generated judgements in real time. To address these concerns globally, this study proposes a novel framework that integrates split learning with federated learning (FL) and edge processing. Separation of Concerns is ensured by dividing deep neural networks between edges and servers such that only intermediate activations are supplied for server-side processing and raw sensor data remains on-device. The system incorporates blockchain-auditable procedures, human-centric consent methods, and Federated Learning (FL) to enable collaborative model training across dispersed users without central data aggregation focusing on decentralization of data. Tested on a real-world wearable sensor dataset, the hybrid split-FL method saves better on on-device computing and matches accuracy with centralized alternatives, proving its suitability for low-resource IoT situations. Activation encryption and adaptive model personalization mitigate non-independent and identically distributed data issues and leakage risk, while stress testing confirm scalability in diverse network settings. By offering a scalable, privacy-first paradigm for IoT driven applications, this work advances edge processing by combining personalized split learning with federated normalization in human-in-the-loop CPS.
Keywords
Split Learning, Federated learning, FLASH-BPH, Block-chain-auditable procedures, Human-in-the loop, Cyber Physical Systems