Hatem Ghorbel, who is originally from Tunisia, obtained his PhD from the EPFL where he became familiar with language theory in the theoretical computer science laboratory, specialising in the study of the logical and mathematical foundations of computer science. He has been a professor at the HE-Arc in Neuchâtel since 2004, where he is responsible for the "Data Analysis" group. He is a regular swimmer and really appreciates being close to nature and the lake.

Meet a researcher who knows how to make the link between data and the industrial world.

What are the main skills of your group at He-Arc?

We work on three types of interconnected activities based on applied artificial intelligence. These are data analysis, machine learning and optimisation.

Can you explain what data analysis means?

When we talk about data analysis, on the one hand we have textual data, i.e. digital documents (texts, PowerPoint, pdf, etc.) and on the other hand, digital data that come from the web or from various technological devices (internet, social networks, etc.).

Our main goal is to discover and extract knowledge from these documents by applying classification techniques and linguistic models.

We work on keyword extraction to try to filter the content and keep what we are interested in. When using a search engine, the challenge is to be able to give relevant computer results to a query. Understanding the intention and context of the text is still the most complicated part of the “translation” process.

We do a lot of work in the medical field, analysing thousands of scientific publications to identify, for example, new plausible associations between a drug molecule and a protein in the body. This is equivalent to several scientists brainstorming together. I'll give you a very simple example from everyday life: paracetamol is used for pain, and pain is often associated with fever, so maybe paracetamol could be used against fever.

For the moment, this is a research project, which we have already published. The project was initiated by Novartis, but the collaboration was not successful. We hope to find a collaboration with a company in the canton. The biomedical field can nowadays be done on the basis of knowledge from texts, internal documents, technical reports, etc. It is not easy today to do this.

It is not easy today to make this model generic across different industries. Indeed, when you go from MedTech to construction for example, the vocabulary associated with the description of a problem is not the same. However, the latest Transfer Learning technologies offer interesting avenues of exploration through the language models of the BERT family.

What is the sector of application of Machine Learning and your group’s contribution in this activity?

Machine Learning is also applied to the industrial sector. The aim is to analyse large volumes of data in order to provide decision support elements for optimising processes and anticipating unplanned defects. Today, thanks to the many sensors installed on the machine (pressure, temperature, power consumption, etc.), it is possible to correlate a signal and an event, provided there is an algorithm that makes the link. It is the real effort of the machine that is taken into account and the part is changed only when necessary, which saves maintenance costs.

The algorithms we develop are based on analyses, mathematically of time series data, which are process measurements of a phenomenon through operating time. These algorithms allow us to find automatic solutions to quality control and abnormality detection. Sometimes this abnormal behaviour is not directly perceived by a human. We are able to perform quality control more efficiently and quickly and we have experience in the field of machine tools.

We have also taken part in several projects, for example in nuclear power stations to detect abnormalities, in wastewater treatment plants or incineration plants, etc.

How do companies contact you?

There is strong cohesion between industry and the academy, which has been strengthened in recent years through the regional economic fabric.

The engineering school is known by the companies that contact us directly or through the canton's economic department, which acts as a link between us.

Optimisation is the latest core activity of your group, what does this mean in concrete terms?

There can be difficulties in optimising a process. I'll take the example of a robot that picks up parts from a rolling tray and has to deposit them in another place. Imagine that there are conditions for picking up a part, because not all the positions of this part can be taken. So after analysis, we know that we have to vibrate the tray to be able to put the part in a pickable position. So we have to find the right vibration characteristic of the part and the tray that will maximise the number of pickable parts on the belt so that the robot doesn't stop. The robot has a camera that looks for the part in the pickable position. If the image does not conform, it does not pick it up. Basically, it's a mechanical problem and our aim is to find the right parameter to be able to optimise the process. First of all, we have to understand the operation and the process for each area. For each project, we go to the site to understand the mechanism, the machine, the mechanical aspect, which allows us to take into account the necessary parameters.

Is data collection within the company a sensitive issue?

Yes it is. The loss of data due to attacks remains a major issue. The data is on internal servers, because companies lack confidence in the Cloud.

Before collecting the data, we create scenarios with the companies to ensure that the data collected will be of good quality and will answer the question being asked.

We are still at a stage where we do not mix data between companies. We need to know the process of each company in order to interpret the data.

Are your partners cantonal companies?

We have many partners in the canton, especially in the machine tool sector. We work with Mikron, Tornos, the Providence Hospital, NOMAD, Johnson & Johnson, etc. We are part of the internal project Mill (MicroLean Lab). This is a very large project that brings together more than thirty companies. A microfactory that is capable of hosting apps, manufacturing, assembly and control technologies. This microfactory can be configured as required according to what needs to be produced. The path opened up by the Micro5 makes machining more autonomous with less human intervention.

What are your current challenges?

The main challenge for my group is to continue developing Innosuisse projects in the field of AI. These are projects that give us opportunities for innovation. We are in a field where, every day, new algorithms appear and to make innovation, we need a concrete applied case and serious partners. We have carried out several projects with partners in the region, which has allowed us to gain experience and maturity to be able to quickly solve problems in an innovative way, especially in the fields of machine tools and MedTech. It's a real challenge to create value by making sense of data.

What do you think of our ecosystem?

We receive a lot of invitations to collaborate from the canton, which is a very good thing.

I think we need to strengthen the networking that already exists. I think that the first step is difficult for companies that want to innovate, so why not consider helping them to carry out their first innovation project, which would boost our ecosystem even more.

‍

‍