French software start-up Cartesiam manufactures AI modules that can be integrated into any connected object and give it AI functions. We met with Joël Rubino, Cartesiam CEO and co-founder.
DirectIndustry magazine: Cartesiam wants to put AI in every connected object. How would you manage to do that?
Joël Rubino: Our software works inside special components called microcontrollers. They are like micro computers. Around 15 billion microcontrollers are sold every year and you can find them everywhere. For example, there are between 100 and 500 microcontrollers in a car which control everything in it. Microcontrollers can also be found in all the connected objects of the IoT. We built an AI module that runs directly into the microcontroller. And since all connected objects have microcontrollers, you can put AI everywhere, in all objects.
DirectIndustry magazine: What was at the origin of this project?
Joël Rubino: Our company was launched in 2016 and was the result of an analysis. First, all objects were individual. Then, they were connected to the internet. Today we want objects to be intelligent. All these objects produce data and in order to analyze it, it has to be sent to the cloud. But there are more and more objects on earth. Experts expect that by 2025 there will be 100 billions connected objects (from cars to remote controls) and it will not be possible, even with 5G, to manage all this data in the cloud. We need a way to analyze the data before the cloud. We have to analyze it where and when the signal becomes data. We have to analyze the data as close as possible to where it is created, because we won’t have the infrastructure to manage the data of billions of objects. Microcontroller is the technology available at the edge of networks. This is why we focused on it.
DirectIndustry magazine: Who has implemented your technology and what forms does it take concretely?
Joël Rubino: French company Eolan was the first to integrate our technology into a vibration sensor and it became a predictive maintenance solution called Bob Assistant. When you put Bob Assistant on a machine it will listen to its vibrations and learn the normal vibrations with machine learning. Within a week, Bob defines a model that reflects its complete understanding of the machine. Bob then switches to analysis mode, where it can determine if the machine is vibrating normally as it has learned or if there are deviations. If this is the case, Bob connects to the network and sends this information to alert that there is a problem with the machine. While traditional detection systems send all information and data to the cloud (which requires infrastructure, security and bandwidth at the server level), our technology allows us to do the analysis directly in the sensor, which is quicker.
DirectIndustry magazine: Does your technology have great potential for predictive maintenance?
Joël Rubino: Predictive maintenance and the ability of the objects to self-diagnose themselves was the use case from the start. Total, SNCF, Veolia and EDF are some of the companies that are equipped with Bob Assistant for onboard predictive maintenance. We are also working with Naval Groupe who bought our Nano Edge Studio technology to make onboard sensors for all their equipment for predictive maintenance.
DirectIndustry magazine: How does Nano Edge Studio work?
Joël Rubino: NanoEdge Studio is a software environment that our customers install on their PC which allows them to create a program that is specific to their use case (vibration analysis, magnetic analysis etc.). In just one hour, they can have a program that integrates machine learning functions that they can install in their object easily. And all this is done without sending any information to the cloud. We offer the ability to work locally.
DirectIndustry magazine: Apart from predictive maintenance, what are other applications of your technology?
Joël Rubino: Our technology can be used as an anti-fraud system as it allows you to digitally seal a signal. One of our clients manufactures gas pumps, and therefore handles large volumes of liquids. What they found is that when they install pumps, they see alterations in the signal. They realized that their pumps were charging for 10 liters and actually delivering only 9 liters. So they decided to install our technology to avoid any fraud. Now, when they install gas pumps, they integrate a sensor in the microcontroller that is in the pump that delivers the gas so that when the pump charges 10 liters, it actually delivers 10 liters. A microcontroller in the pump has recorded the normal signal of the pump when it is running. If a pump operator slightly alters this signal, it will be instantly detected and an alert will be sent.
DirectIndustry magazine: So your t.echnology is participating in the democratization of the use of AI?
Joël Rubino: Our technology helps embedded developers integrate AI without being AI specialists.There are too many barriers to AI integration due to the lack of data scientists and the lack of data sets. Our technology doesn’t need data scientists and infrastructure, which cost time and money. We save years of development time and it costs less.