Despite the crisis, Fieldbox.ai observes the growth of AI in the industrial sphere

Founded between 2011 and 2012 within the Telecom ParisTech incubator, the IDMOG startup moved to Bordeaux in 2014. It specialized in the development of software solutions for oil and gas players. The company has since changed its name – Fieldbox.ai – and grown from 6 to 80 people. Fieldbox.ai is also installed in Paris and Singapore.

“One of my two associates worked on an industrial site. During the day, he processed data by hand, and in the evening, he developed tools to automate processing,” recalls Aymeric Préveral-Etcheverry, co-founder and CEO of Fieldbox.ai. “He realized that this problem he had, that is to say the need to bring intelligence to the processing of industrial data, was found in many companies”, he continues. . “From the start, we had this desire to develop tools that allow manufacturers to improve their efficiency by making better use of their data”.

Since then, Fieldbox.ai has “expanded the scope” of its activity. “We cover different sectors of industrial activity: logistics, supply chain, manufacturing industry, energy, smart city”, explains Aymeric Préveral-Etcheverry. The company’s customers are major groups such as Suez, Total, the Aéroport de Paris group, RATP, SNCF and medium-sized companies. In total, the company has 50 customers and operates its solutions installed on 150 industrial sites.

“For example, for a client, we are going to respond to a problem of reducing energy consumption on a machine, detecting the breakage of a compressor, switching on pumps at the right time to reduce energy consumption” , illustrates the CEO.

A “triple expertise” combining artificial intelligence and industrial knowledge

But that doesn’t actually say what Fieldbox.ai does. “We have triple expertise. First, we understand business needs,” says Aymeric Préveral-Etcheverry. “Secondly, we have data scientists capable of meeting the needs expressed by developing the appropriate algorithm. Third, our development and DevOps expertise allows us to design the application, deploy it and maintain it in operational condition”.

In this context, Fieldbox R&D develops software bricks, Python libraries, pre-trained models, or even templates. “We are trying not to be just a consulting or data processing firm, but to have a more global technological ambition”, states the CEO. “We contribute to the PyTorch project, we publish research papers and we share them at conferences.”

In this sense, the company does not develop metamodels, unlike Meta or OpenAI. “We try to be pragmatic, to find the right model for the use case,” says the manager. “When we need to use deep learning algorithms, we look at what is available in the open source ecosystem and we can perform transfer learning”.

As a reminder, transfer learning aims to apply the knowledge of a model designed to perform a task – for example, the recognition of objects in an image – to another task.

To establish certain use cases, manufacturers sometimes have little data. “In this case, we use standard machine learning models,” says the CEO. Fieldbox.ai is also interested in “domain adaptation”, a sub-technique of transfer learning.

The company defends additional expertise, appreciated by industrial engineers. Moreover, the Bosch equipment manufacturer has made it its leitmotif. “Systematically, in our teams, we have data scientists and engineers to cross artificial learning models with real physical models”.

According to Aymeric Préveral-Etcheverry, a “buzzword” qualifies this practice: “physics informed AI”. “There are announcements with great fanfare in the market, but sometimes you just need to recode a physics ruler into a model and it does the job,” he said.

More and more varied use cases…

On a daily basis, Fieldbox responds to three main families of projects. “We handle everything related to equipment maintenance. The first thing is to optimize maintenance, or even make it predictive if possible,” he says.

The CEO seems to consider that the expression predictive maintenance is partly overused. “True predictive maintenance relies on supervised learning, with a very clean set of data. In theory, this is the ideal special case,” notes the CEO. “Before that, we can do unsupervised learning, to do weak signal detection on clusters of data where the equipment is performing abnormally.”

Aymeric Préveral-Etcheverry cites the case of his client Valorem, an operator of photovoltaic panel fields. “With Valorem, we collect all the signals from the photovoltaic panels, but also from the electrical equipment that surrounds them. Our algorithm makes it possible to identify which panels are the most faulty in order to optimize maintenance and therefore the productivity of the site,” he explains.

The second type of projects concerns the optimization of operations, which the CEO sums up by the use of recommendation algorithms. “It can be recommending the temperature of an oven, the speed of a pump, a trajectory for a vehicle”.

“For example, we are co-developing with the RATP a model to help operators decide which train should go to which station at which time”, he illustrates. “There are obviously schedules, but depending on the hazards – delays, objects on the track – the operator must react very quickly to redirect a train in a station”.

The algorithmic model in question takes into account all the parameters that an operator checks to make his decision. “That is to say, how long has the driver already worked? Does the train have to go for maintenance?, etc. “.

The model must “very quickly” perform recommendation scenarios for the operator “to reduce his cognitive load”.

“We work with many customers to optimize their energy bills. It’s a big topic for them right now.”

Aymeric Preveral-EtcheverryCEO, Fieldbox.ai

Moreover, this notion of optimizing operations is gaining popularity among Fieldbox.ai customers, due to the current energy crisis. “We work with many customers to optimize their energy bills. It’s a big subject for them at the moment, ”said the CEO.

The third theme concerns computer vision. “On industrial sites, there is a large fleet of cameras already installed to monitor the facilities. This can be used for devious purposes to automate operations,” he says.

This is what Fieldbox.ai has implemented with Suez. The water specialist inspects the small pipes in the network using cameras. The images were only viewed by human operators. “I let you imagine the interest of watching this type of video in order to be able to indicate that at kilometer 12 there is a root or that at kilometer 45 there is a collapse”, declares Aymeric Préveral-Etcheverry. “It’s something that we can completely automate with a computer vision model.”

Fieldbox.ai has developed an algorithm capable of detecting objects. It is embedded in an application that lets the operator make the decision for “borderline cases” that the model would have difficulty recognizing. Suez would have reduced the viewing time by 80%, improved the quality of service as well as its intervention times.

… In production

“POC purgatory has been a reality.”

Aymeric Preveral-EtcheverryCEO, Fieldbox.ai

The leader of the company assures that these are not just POCs or experiments. “POC purgatory was a reality,” he admits. “Today, I think the situation has changed dramatically. Businesses and operational staff have understood the possible gains. The operational staff want it now, the stewardship has to follow: we have to find the means to deploy”.

Aymeric Préveral-Etcheverry mentions the fact that around 70% of data processing and AI projects fail among manufacturers. “This is precisely where we position our triple expertise,” he boasts. “We have to support the businesses, have functional algorithms and master the deployments”.

A rise in the power of the cloud among manufacturers

In addition to a greater number of use cases, the CEO observes the rise of the cloud among his industrial customers.

“Over the past two years, our customers have started to catch up on cloud adoption, especially when it comes to industrial topics,” he notes. “Before the pandemic, we often did on-premises, virtualization-based deployments. During Covid, manufacturers had time to improve their IT infrastructures. More and more often, we deploy projects agnostically, often on Azure, but also GCP, AWS or OVHcloud”.

More or less mature data lakes are used to collect industrial data. “We orchestrate the projects in Kubernetes clusters, in the customer’s cloud account.”

Edge and hybrid deployments are starting to democratize. “What we did historically, which is to deploy our on-premises applications on VMware, we are seeing it come back through solutions like Azure Edge. We use the K3S project a lot, but we follow the customer’s choice of infrastructure”.

In addition, the issue of the sovereign cloud, and more generally of the protection of industrial data, is increasingly understood by Fieldbox.ai customers, assures Aymeric Préveral-Etcheverry.

“Overall, manufacturers have aligned the technological bricks, now they are looking to obtain returns on investment”.

Aymeric Preveral-Etcheverry

“A few years ago we made our customers aware of localization and data protection issues. We had rather a role of prescriber. Today, the client himself is fully aware of the issues, and makes requests in this regard. So I think I think the offer [de cloud souverain] happily unfolds. This corresponds to a real market segment that will be increasingly important,” he anticipates.

“Well framed” projects would require three to four months of development, when more complex subjects require about a year of gestation, according to the manager. This does not scare companies: Fieldbox.ai forecasts a 25% growth in turnover for the year 2022. The startup plans to recruit around forty people next year.

“It’s about responding to market demand: there are different topics related to energy consumption, optimization of logistics, etc.,” observes the manager. “Overall, manufacturers have aligned the technological bricks, now they are looking to obtain returns on investment”.

In this respect, software projects are perceived as effective solutions compared to the purchase of new equipment often

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *