What an absolutely beautiful paper: Wang et al, Green Industrial Internet of Things Architecture: An Energy-Efficient Perspective.

Hierarchical Industrial IoT architecture

I recommend checking it out just because its design is stunning. Of course, the topic is also very interesting - CPEs and Internet of Things (IoT) devices often have very low energy requirements. On the one hand, such devices are inherently small and require less energy. On the other hand, there is an increasing energy usage due to a growing number of IoT devices used in an industrial setting. The paper proposes a communication architecture that aims to be energy-efficient, thus coining the term "Green Industrial IoT".

More specifically, the authors propose a hierarchical architecture that consists of a sense layer, a gateway layer and a control layer, as seen on the figure on the right. Sensor devices are for data collection. They send data to a gateway node. Note that in the scenario of the paper, sensors are not allowed to talk to each other but have to go through a gateway node. Gateway nodes may have a set of multiple sensor nodes they manage and act as routers to further send sensor data to control nodes. It is also very interesting that control nodes balance out the sensor nodes and assign them to the gateway nodes. With this, the authors argue that their approach has less resource utilization and less energy consumption.

Smart scheduling for Industrial IoT

Industrial IoT (IIoT) uses RFID chips for controlling and monitoring factories. For instance, one could monitor when each product goes through a specific stage in the manufacturing process and, based on that, seek improvements in the process, or simply collect statistics. Or, if a new truck with a load comes in, RFID sensors could send an alert and start an unloading process. There are a lot of possibilities - Siemens has a large set of products that allow all kinds of IIoT solutions for automatization in a factory setting.

However, communication among these small IIoT devices is energy-intensive. The reasons are multifold: the sensors themselves use a huge amount of energy, the wireless sensor network needs a lot of energy to keep up the network, and the topology plays a major role. So the main idea of the authors is to buffer the communication using sleep periods and predict how long these periods should be.

Each communication attempt should be carefully scheduled. By predicting when a communication is needed and letting the sensor devices sleep in the meantime, total energy consumption can be lowered. Specifically, gateway nodes that get a communication request decide whether to wake a sensor node using the following simple mechanism: has the predicted sleep interval expired (i.e. does the gateway node predict that the sensor node should now be awake)? Is the communication/query request critical? The figure on the left shows the exact mechanism.

The sleep period is predicted simply by taking an exponential average of the empirical historical sleep intervals that the gateway node has experienced for the given sensor node.

This works, since the authors show an improvement in terms of energy consumption, but it does also easily open itself to the idea of using Machine Learning (ML) to predict the sleep period in a more sophisticated manner. One of the authors specializes in ML, so we may see some follow up work in the near future.

References:

Wang et al., Green Industrial Internet of Things Architecture: An Energy-Efficient Perspective
* Figures are directly from Wang et al.