https://www.mdu.se/

Integration of deep learning-based edge computing into industrial processes has enabled intelligent automation in the Industrial Internet of Things (IIoT). However, the deployment of deep learning inference models in embedded edge clusters remains a challenge due to energy constraints, communication latency, and computational limitations. This paper proposes an energy-aware task allocation framework for optimizing deep learning inference in IIoT environments utilizing a Embedded cluster with both Wi-Fi and Ethernet-based communication setups. We have implemented and evaluated the performance of four classical deep learning models (MobileNet SSD, Tiny YOLO, EfficientDet Lite, and Faster R-CNN) on the edge cluster and analyze their execution time, energy consumption, and network efficiency. The framework provides a scalable and energy-efficient solution for deploying deep learning inference on resource-constrained edge computing platforms. The is ongoing development for the Embedded-Edge test beds and Future work will explore heterogeneous edge device integration and 5G-enabled computing for further performance enhancements. Experimental results demonstrate that Ethernet-based communication improves task execution speed by 15-20% and reduces energy consumption by 12-18% compared to Wi-Fi. Additionally, a fault-tolerant task reallocation mechanism ensures uninterrupted operation in case of node failures. The findings suggest that efficient task scheduling and network optimization strategies can significantly enhance the real-time processing capability of IIoT applications on Embedded edge also.