The intersection of IoT and AI seems pretty obvious. AI automates decision processes based on large data sets. And IoT is one possible data source for AI, not the worst one even if challenging: IoT devices provide big amounts of structured data which can be learned from. But there is also more to it.
This article is about the reasons why we decided to extend our technological expertise to IoT and how we are doing it. It is a summary of the presentation I gave at the MWC19.
Most of the time, our data projects start with collecting data and exposing the result of its analysis to business users. In a second step we build an AI and automate decision processes where it make sense. We have for that some building blocks, our adaptive data services, like data storytelling (a new method to present an argumentation based on data to business users, management or even customers) or machine learning that we embed into applications and user interactions.
Machine learning is essentially applied in 2 steps:
- A first phase where machine learning algorithms are used to identify patterns in data sets and find correlations between some metrics, like for instance some temperature, pressure and vibration level with the technical health of a machine
- The acquired knowledge is applied in a second step in real-time to respond quickly to some measured conditions like in the case of face or object detection
Intelligent IoT is the result of the combination of IoT with AI. IoT devices generate a large quantity of streaming data. To realise the full potential of IoT, this data will need to get correlated with many other data sources. To do that, AI becomes critical to extract information from not only vast amounts but also from a large variety of data.
How did AI change the way we use machine or device data? Let’s look at some typical use cases. Today the main industrial use cases of the combination of AI and IoT are in avoiding downtime through predictive capabilities or in the improvement of operational efficiency.
Data from a production line can be connected with business data and get analysed to help optimise the production process and usage of the machines. For example, one of our customers in Germany could identify, through the analysis of sensor data, the reason why their large production machines where regularly producing bad results. A machine learning system correlated measures like temperature or vibration levels and characteristics of raw materials with those issues. They could optimise the configuration of those machines to improve performance.
Another use case is the pro-active maintenance of critical building assets where an AI can inform about issues which might occur in the near future within a specific building area or a piece of equipment. This can help schedule maintenance tasks more efficiently.
Beyond the optimisation of operations, products with embedded sensors can also collect data on how customers use them with the potential of giving manufacturers and their engineering teams new product design ideas.
We at incontext.technology are now extending our technological footprint to support AI on the edge. Why is that needed?
As we saw, IoT provides the data AI needs in order to make smart decisions. In return, some of these decisions can lead to actions performed by IoT devices such as robots or industrial machines.
Machine data can also be processed on-site, at the “edge”. Bringing reactive intelligence into the machine permits immediate action and helps reduce response cycles to specific situations. It can be used for instance to spot problems immediately before a batch of defective products leaves the production line.
That is where LTE-M was instrumental for us. It gave us 3 essential features which allowed us to extend our AI application development to IoT devices:
- longer battery life through lower energy consumption
- Bi-directional communication
- Short rollout time and low deployment costs
How does those features benefit our specific AI projects?
For most of our AI projects we relied on machine data as provided by our customers. LTE-M gave us the possibility to extend our value proposition through a real-time data acquisition system with IoT devices that can be deployed simply. We can also communicate with those devices and steer them remotely. The benefit of this approach is obvious:
- it leads to a better data and resource management and to a better control of data quality (which is of paramount importance in AI)
- it allows a flexible distribution of processing to the machines and devices
- it opens a way to evolve our devices from automation to autonomy
Those devices can get more and more autonomous as they learn from the central analysis of the overall dataset generated by all of them.
The example we started to tackle shortly is the usage of AI in facility management.
We built a POC for a project on a large infrastructure monitoring use case proposed by the GSMA, Orange and Sierra Wireless. We built a battery-powered IoT device POC based on the mangoH Red board. It is connected to the Live Objects data and device management platform from Orange which we use to stream real-time data from the board to our analytical and AI processing components.
We measure light level, temperature, pressure and humidity levels in tunnels and capture IoT infrastructure information like battery and connectivity levels as well.
Through an appropriate user interface we provide the maintenance team with analytical and realtime information about what is happening within the tunnel. The information about the evolution over time of the light level makes it possible to plan maintenance more efficiently and even getting planning proposals from the AI. The AI is also already able to recognise in realtime what is happening in the tunnel, like some vehicle crossing, through the analysis of patterns in the measured conditions. With that, it can make the difference between real issues and some normal variations within the tunnels.
Why did LTE-M make this possible?
- It makes data transfer available in a cost-effective way with a connectivity infrastructure which is already available.
- It simplifies deployment of battery-based devices through low-energy requirements, also in environments where maintenance cannot be done frequently
- It makes remote steering of devices possible to regulate the frequency of data acquisition depending on use case needs, which helps us increase the battery life
As a summary, AI and the IoT won’t solve all the world’s problems, but they will change the way we approach them as we saw here. AI gives a voice to IoT. Together, they give us a way to a much richer and more natural interaction with data, analytics and AI for everyone. And this is what we aim for.