Learning on the Edge for Sensor Role Allocation in BLE IoT devices

Wireless Sensor Networks (WSNs) consist of distributed, resource-constrained sensor nodes that report on ambient environmental conditions and thus enable Building Automation (BA) systems to control technical systems. The embedded systems, in such WSNs, have to be rigorously optimised for power consumption because of their reliance on constrained energy sources (e.g., tiny batteries or energy harvested from the ambient). Despite significant progress in energy efficiency and usage, the potential for further improvement through intelligent allocation of sensors, in conjunction with the operation of the BA system has not been fully explored. This work aims to address this issue by using a machine learning model implemented using TinyML to intelligently allocate sensing tasks based on the needs of the automation system. Our system estimates the residual battery capacity of asynchronous sensor nodes using TinyML, enabling a centralised automation orchestrator to dynamically adjust the sensor node's duty cycling to meet BA sensing requirements. Preliminary results show improved efficiency in utilising the available energy and higher efficacy in fulfilling the tasks.