Fraunhofer-Gesellschaft
Web special Fraunhofer magazine 2.2025
We hear practically daily about just how vulnerable global trade is. Now, companies are looking to AI to protect themselves. And demand is rising.
Trump’s threats of tariffs, terror attacks on container ships in the Red Sea, wars: Day after day, global trade is under threat. Supply chains are fragile. Eduardo Colangelo and his team at the Fraunhofer Institute for Manufacturing Engineering and Automation IPA in Stuttgart are studying how companies can harness AI to detect disruptions quickly, preventing production stop-pages. Colangelo, an industrial engineer who has been developing solutions for industry at Fraunhofer IPA for 12 years, has seen rising demand in recent months in particular. “We’re seeing accelerating and multiplying crises that are pushing companies to ramp up their investments in resilience,” he explains.
Building resilience takes careful, strategic advance work, Colangelo points out. Analyzing individual production condi-tions is an especially important factor. Which raw materials are needed? How can they be obtained? What suppliers and modes of transportation are there? Are there alternatives? This information is fed into AI models that the Fraunhofer IPA team developed as part of the PAIRS research project. These models detect disruptions along supply routes while at the same time keeping track of critical trends in commodities markets. To do this, they draw on sources including up-to-date satellite images from Copernicus, the European Union’s earth observation program, and reports from news portals for their analyses.
Resilient supply chains thanks to AI
The AI measures factors such as the den-sity of ship traffic. If cargo ships are lined up outside a port or in a passage, there is probably an issue somewhere. The port might have had to close, or a container ship might have had an accident and is keeping others from proceeding, as the Ever Given did in the Suez Canal in 2021. If the threat poses a risk to the company, the AI issues a warning. Colangelo’s colleague Theresa-Franziska Hinrichsen notes: “But the decision whether to take action rests with the human.” If the human operator concludes that intervention is not needed, that information also flows back into the AI model. “This way, our AI is always learning and get-ting better and better at assessing how relevant certain events are to that specific organization,” Colangelo comments.
Right now, the research team is also working on simulation models based on individual company data. Of course, there are many conceivable scenarios that pose a risk to production, Colangelo says, but there is no need to prepare for every single one. “That would only increase costs unnecessarily. We only simulate the ones that are likely and critical. Resilience is always a balance between impact and likelihood of occurrence.”
Learning from crises using AI models in disaster preparedness and response
The AI models from Fraunhofer IPA could strengthen not only industrial resilience but also disaster preparedness and response. In collaboration with the German Federal Agency for Technical Relief, the team of researchers plans to use scenario analyses to identify high risks so that targeted investment decisions can be made. “But to train the AI models, we need the relevant data in the first place. And unlike in industry, that data is hardly available in a structured form,” Hinrichsen explains. With that in mind, the researchers’ first step was to develop a tool that makes it easier to gather relevant data in the event of a disaster. The tool has various uses, including simple monitoring of stock levels at emergency response sites and smart planning for needed materials. “There isn’t much time and effort involved, but the benefits are huge,” says Hinrichsen. In addition to information on response efforts, other types of data, such as mete-orological data, information on the availability and range of communication net-works, infrastructure data and historical data about disasters, can be highly valuable in planning relief and preparedness mea-sures. This information can also help to answer important questions. How did the weather conditions and the speed at which river water levels rose influence a flood disaster? How did a crisis situation change when certain factors interacted? “The data is a treasure trove just waiting to be explored. Once that happens, AI will be really helpful for disaster preparedness and response,” Hinrichsen says. And we will be able to learn from past crises.