DIGITAL TWIN ARCHITECTURE FOR IOT: EDGE PREPROCESSING, STREAMING EVENTS, AND RELIABILITY SLO/SLA

The article proposes a digital twin architecture for an Internet of Things (IoT) environment, focused on stable operation under increasing load and degradation of telemetry quality. The architecture includes a data collection loop from devices and gateways, a message broker with redelivery and loss/duplicate control functions, streaming event processing, a digital twin core, and services for diagnostics, visualization, and (optionally) control. To formalize reliability, an aggregated data quality metric Q∈[0,1] is introduced, taking into account completeness, outliers, drift, distribution skew, and source trust, as well as a degradation rule that switches the system into normal/degraded/offline modes depending on thresholds Q_"high" ,Q_"low" and the SLO compliance indicator I_"SLO" . The Model Manager module implements the selection of an active model from the set M={M_"phys" ,M_"ml" ,M_"hyb" } depending on Q and I_"SLO" , ensuring a trade-off between accuracy and stability. Experimental evaluation at loads of 100–2000 msg/s shows the advantage of the edge+quality mode: a reduction in p50/p95/p99 latency by 20–32%, a reduction in message losses by up to 50% and duplicates by up to 50%, as well as an increase in availability by up to 0.3 percentage points. It is shown that the proposed formalization of SLO and data quality extends the range of stable system operation and increases the predictability of digital twin behavior under adverse telemetry transmission and processing conditions.