With the increasing use of Internet of Things (IoT) devices in various industries, the challenge of efficient use of energy and reducing the amount of device downtime is a key concern in sustainable and reliable systems usage. The paper deals with energy consumption optimization and the forecasting of device failures in the IoT systems in industrial automation. More precisely, we suggest the hybrid solution consisting of reinforcement learning (RL) to optimize energy consumption, Long Short-Memory (LSTM) networks to predictive maintenance, and edge-cloud offloading solutions to improve the performance of devices and their sustainability. We use the modeling of heterogeneous IoT devices in our methodology, but with different energy profiles, different workloads, and operational modes. We use RL to dynamically tweak devices modes, which saves energy and does not impact task completion. We combine predictive maintenance, in which we use LSTM to forecast device failures using historical data to give timely warnings on maintenance. Moreover, the edge-cloud offloading decisions are also integrated into our system to further optimize the energy consumption and optimize the efficiency of performing tasks. The principal results are that our method is highly energy efficient and device uptime, and reinforcement learning is the most energy saving method. The predictive maintenance model is an effective way of cutting on the downtime, as the correct prediction of the occurrence of failure results in the taking of preventive measures. Task offloading, especially when there are resource constraints, improves system reliability without affecting the performance. Such findings indicate that adaptive IoT systems could be used to optimize energy use and system performance and, therefore, make industrial automation more sustainable. Our solution is informative to the design of future IoT systems in which energy efficiency and predictable stability will be the most crucial attributes. Keywords: IoT Energy Optimization, Predictive Maintenance, Reinforcement Learning, LSTM Networks, Edge-Cloud Offloading.
Kobra et al. (Wed,) studied this question.