Toward privacy-Aware human digital twins: A multi-Layer architecture

The concept of the Human Digital Twin (HDT) is gaining increasing attention as a means of modelling individuals in real time using physiological and contextual data. While Digital Twins (DTs) have been widely adopted in Industry 4.0 to optimize machines and processes, the modeling of human operators as digital entities has seen limited adoption. In Industry 5.0, the well-being of human workers becomes integral to production, and HDTs play a critical role by enabling personalized support and improved human-machine interaction. The development of sensor-rich wearable devices has paved the way for the real-time digital modeling of human beings. However, deploying HDT systems in real-world settings introduces significant challenges, including a lack of interoperability among heterogeneous devices, data fragmentation across vendor ecosystems, and concerns related to privacy and user control. This work proposes a multi-layer HDT architecture composed of Edge, Fog, and Cloud layers. The Edge layer, implemented as a mobile app, enables users to collect, process, and manage data locally. The Fog layer offers intermediate storage and computing capabilities, while the Cloud layer manages global model repositories, supporting recommendation and remote inference. RESTful APIs ensure secure and standardized communication across layers. To validate the architecture, we implemented a real-world Human Activity Recognition (HAR) scenario using wearable sensor data. A Convolutional Neural Network (CNN) was trained to recognize five distinct activities. Additionally, Human-in-the-Loop (HITL) mechanisms enable ondevice model personalization, allowing adaptation to individual users and improving accuracy over time. This contribution moves toward deployable, privacy-aware, and open HDT systems suited for real-world use.