The Intelligence of Things

Industry 4.0 is predicated on some fundamental technologies such as the IOT (Internet of Things) , which is the network of devices that connects, collects, and shares data across a vast web of devices or “things,”. There is also, however, AI and its subset Machine Learning (ML), which on the other hand consumes massive amounts of data, performs analyses and extracts patterns from it in order to learn and improve on a given task.

In the fast-pace and continually developing world of Industry 4.0, the connecting, collecting, and sharing data across the organisations is still a fundamental objective. But the handling of data is only one step in the value chain as we still need to process, analyse, and extract value from it. However, for that we need AI as it is an essential form of analytics to make rapid decisions and uncover deep insights as it “learns” from massive volumes of IoT data. The convergence of these two enabling technologies such as Artificial intelligence (AI) and the Internet of Things (IoT), has come about to become a new discipline - the  Artificial Intelligence of Things.

The convergence of disciplines, IoT and AI, into the Artificial Intelligence of Things (AIoT) allows us to make the best of both worlds by connecting and collecting vast amounts of real world data and then rapidly analysing and surfacing deep insights as the AI “learns” from the massive volumes of IoT data. Well, at least, that’s the theory.

However there are issues. This is because, although the merging of the purpose of IoT and AI are seemingly mutually compatible and highly desirable, their underpinning  technologies are anything but.  

This is because many machine learning models demand high-performance computing power and are computationally expensive to train, and running inference on them will be also quite expensive. Therefore, if we are to benefit from the convergence of IoT with AI in some sort of feasible integration as the Intelligence of Things, we will need to find a solution that allows machine learning models to run inference on smaller, more resource-constrained devices. And that isn’t going to be easy. The overarching problem is how do we compress machine learning models to fit into small Edge devices, let alone resource-constrained IoT devices. But also why would we want to?

Well, if we consider the issues we currently face handling sensor data, in collecting and processing the data in manufacturing, there's essentially going to be a bottleneck where a machine tool is sending us real-time data but our device is unable to send all of that data to the processing analytical engines, perhaps in the cloud, immediately, so this necessitates packet queuing and latency. Then it makes sense to put some of the intelligence and analytical competency on the device, rather than depending on having all the intelligence running on a centralised cloud service.

The obvious solution is to deploy the ML at the edge as some edge devices are powerful enough that they're able to run these heavy lifting algorithms, like deep learning models. For example; TensorFlow light, which is basically a compressed solution of Google’s flagship Tensorflow, is designed to be used on a mobile phone. But many mobile phones are insanely powerful so the objective is to go further and compress AI models to run on truly resource-constrained IoT devices.

The Benefits of the Intelligence of Things

The benefits of running the ML at the edge or even better on the IoT device itself are that you can eliminate latency issues and operate critical automation processes in real-time:

Latency: The data does not need to be transferred to a server for inference because the model operates on the edge devices. Data transfers to the cloud typically take 10’s of milliseconds , which obviously causes a delay. In manufacturing and machine monitoring you may need an emergency shutdown response in microseconds.

Reduced bandwidth: ML inference does not require internet bandwidth as the on-device sensors capture data and process it on the device. This means there is no raw sensor data constantly being communicated upstream to a server.

Data privacy: The data is not kept on servers because the model runs on the edge or the device, which increases the guarantee of data privacy.

The Intelligence of Things is a burgeoning field of interest in Industry 4.0, its finding new audiences, as technologists look for process and automation efficiencies and ways to push intelligence to the edge and even on to the IoT device. In the following articles we will dive deeper into the practicalities of the Intelligence of Things, and how we can use this new analytical model to our benefit in Industry 4.0.  

‍

About the Author

Written by Alasdair Gilchrist, a technology polymath that has spent 35 years in the field of technology spanning roles as a computer technician, data analyst, network engineer, security professional, OT data engineers, cloud native developer as well as serving as an IT Director.