DNVGL.no

Unlocking 50 years of hidden knowledge using AI

Breaking down traditional boundaries between software and machine learning.

"DNV GL made some good digitalization choices many years ago. Now that large amounts of data have turned into gold in the IT sector and are the foundation for artificial intelligence (AI), we can really unlock this data treasure-trove," say Stian Barkbu and Elen Dybesland of DNV GL.

 DNV GL surveyor

They are in charge of the software and AI development environments in a company with more than 150 years of history. DNV GL has developed a highly skilled software house since the 1960s, and its new business unit, Digital Solutions, means it has also become one of Norway's largest IT environments.

The new unit is part of DNV GL's long digital history, but how has the road from large machines in the 1960s to AI in the cloud now been built?

Important decision in the 1990s

"In the 1990s, it was decided to create a separate production system - the Nauticus Production System - to support DNV GL's maritime classification service," says Dybesland.

Through this system, DNV GL has over time obtained large amounts of data from design approvals and ship surveys worldwide. Using steadily improved IT systems, the company forges links between its employees and customers in completely new ways.

"Now we can really reap the benefits from the data," explains Dybesland. 

From idea to AI system in two weeks

"We receive many thousands of plan drawings from hundreds of customers worldwide each year. Not surprisingly, all these customers have their own special ways of preparing their plan drawings. They use their own names, concepts, methods, location of the data in the plan drawings and other special characteristics. To put it briefly, different customers provide different plan drawings and we have to interpret them," says Kristian Ramsrud, the head of machine-learning activities in DNV GL Maritime. 

The old way could be to have one rule for each customer in order to handle the various specifications, but that is not scalable. Ramsrud gives us an example of how this can now be done quickly and efficiently.

"Preparing customer-specific decoding rules is not tenable and will require a massive amount of maintenance work in the future. We put machine learning to work interpreting the files submitted by customers. After two weeks, the system was so well trained that, with a few software adaptations to existing business applications, we could put machine-learning-based filing and revision identification into production," says Ramsrud.

"We've also established a set-up that involves us delivering software changes several times each week and constantly improving the monitoring of the system. This means we can continuously and efficiently improve both the machine-learning algorithms and software applications," adds Dybesland.

It is this difference that puts DNV GL far ahead in the digitalization race.

"If we are to be the leader in what we do, we must digitalize and constantly assess whether machine learning or AI can be a possibility. Our clear focus on benefits means that both large and small AI projects are considered candidates for AI and machine-learning projects," says Ramsrud. 

People are still the most important

One of DNV GL's advantages is the good collaboration between the software developers, experts in machine learning and AI, and the professional experts in various fields.

"Today, we're completely dependent on the surveyors for gathering data. They are the ones that go onto thousands of ships and conduct surveys," interjects Stian Barkbu.

"They are now helped by drones and the ships' own sensors, but in the end it is human beings who are the most important factor in all the surveys we do worldwide. Everything they do is entered into our systems and in that way they help to increase the value of our data every single day," he says.

The production system has large volumes of data from thousands of projects, drawings, results, surveys and conclusions going back several decades.

"This rich data history gives us a considerable advantage when we start to use machine learning and AI. The bottleneck for many of those working on machine learning is to make the data ready for analysis. Our data have been gathered by employees and customers for several decades. At the same time, we have ensured that the data are structured so they can be used for analysis," says Ramsrud.

Has started to automate the processes

Where some companies prepare a script for handling the data from a sensor and call this digitalization, DNV GL has set far higher standards for what it is to include in the digitalization concept.

"Over the past three years, something entirely different has happened in the field of digitalization. Knowledge can now be unlocked from data and connected using good programming skills, innovation from an open source code and our professional knowledge and implementation ability. This combination is where the magic arises," says an enthusiastic Barkbu.

The technology is there to benefit people 

DNV GL wants to give its surveyors tools that help them to plan better and make fewer unnecessary journeys in order to make their working life as efficient as possible.

"We also use machine learning to calculate whether a planned survey can be conducted while a ship is in port. That means our employees don't have to return to the vessel because there wasn't enough time," explains Barkbu.

"Several machine-learning algorithms have been put into production. Our own helpdesk routes incoming inquiries based on recognition of the inquiry's wording. Another algorithm assesses whether a plan drawing that is being dealt with has been dealt with before, and thus simplifies the employee's work. Right now, DNV GL is looking into voice-control of the reports.

"Many of our employees are in locations where they don't want to pull out a PC and write the report in Word. So we envisage that they can dictate the report or take a picture for an automatic analysis while they are inside a ship's tank.

"For us, it's important that if a machine-learning solution works, it must be put into production as quickly as possible. In the big picture, we're confident we have achieved a good start to exciting developments that will lead to many opportunities for those who want to influence an established global industry," conclude Dybesland, Barkbu and Ramsrud.

Elen Stian Kristian
Kristian Ramsrud, Stian Barkbu and Elen Dybesland of DNV GL.
DNV GL surveyor
DNV GL surveyor at work