Dynamic computation offloading for wireless sensors in IIoT

RWTH Publications (RWTH Aachen)

Abstract

The growth of wireless Industrial Internet of Things (IIoT) sensor systems has introduced new possibilities for real-time monitoring, predictive maintenance, and flexible deployment in industrial environments. However, these wireless sensor platforms often operate under severe constraints related to energy availability, processing capacity, and network variability. These limitations pose significant challenges to maintain performance and longevity, especially under dynamic and unpredictable workloads. This thesis addresses the need for adaptive task orchestration by proposing a dynamic computation offloading framework tailored for wireless IIoT sensors. The proposed system is implemented on the Arduino Portenta H7 platform and integrates three key components: a custom energy profiling module, a real-time task scheduler, and a multi-objective particle swarm optimization (MOPSO) algorithm for offloading node selection. Offloading decisions are made using a decision tree trained on past optimization outcomes, allowing the node to respond in real-time to changing energy levels, workloads, and connectivity conditions. A detailed energy modeling framework is developed to capture both local and offloaded execution costs, including CPU switching activity, frequency scaling, and wireless transmission overheads. The system selectively offloads tasks only when the decision results in net energy savings while satisfying application-level latency constraints. The framework is evaluated in three representative IIoT use cases: predictive maintenance, condition monitoring, and environmental sensing. Experimental results show that the dynamic offloading approach reduces local energy consumption by up to 40%, adapts effectively to changing network and workload conditions, and maintains responsiveness without overloading the network. This work con-tributes a modular and deployable energy-aware orchestration framework that addresses core challenges in IIoT systems, enabling more intelligent, autonomous, and energy-efficient industrial sensor platforms.

Authors 1

1. Praveen Mohanram corresponding Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen