At the Institute for Aircraft Production Technology (IFPT) an AI algorithm is being developed that optimally configures the packing density of air cargo.

Do you remember Tetris – a computer game in which you had to build colorful geometric shapes that fit perfectly on top of each other as they fell? Container ships and cargo aircraft are loaded according to the same principle. The packing unit for aircraft is called a ULD (Unit Load Device). These are pallets or containers measuring approximately two by three meters that are used to store baggage, cargo or mail on and in wide-body aircraft. Felix Geiger from the Institute for Aircraft Production Technology (IFPT) has set out to optimize this packing structure. The more tightly the ULDs can be packed and the less space remains between them, the better the capacity of an aircraft can be utilized: “This is practical because then more goods can be transported and the aircraft have to fly less frequently and can save kerosene,” says Felix Geiger. “Suitable software and fast data processing speeds with the help of machine learning help us in this.”

In 2023, almost 58 million tons of air cargo were transported by aircraft worldwide. But it could be even more if aircraft were packed more intelligently and no space on board was wasted. An aircraft can take around 100 tons of baggage on board. Last September alone, almost 40,000 planes took off from Frankfurt Airport to distribute their commercial cargo worldwide. That adds up to a lot.

The freight items have to fit

All airports follow the same principle when loading freight: the incoming freight is scanned for hazardous substances and stored in a packing station. There, for example, it is also checked that nothing protrudes over the edge of the ULD. “Ideally, the load is adapted to the oval shape of the aircraft,” explains Geiger. This task is performed by the loadmaster, who directs the individual pieces of cargo to their location on the ULD. The danger here is not only that the area is not well utilized, but also that the ideal center of gravity must be found.“This often causes delays because the cargo plans are inaccurate and then have to be spontaneously rescheduled.”

In the field of aviation logistics, manual processes are still used today to assemble pallets for cargo aircraft. Employees have to find the best possible loading configuration under time pressure and changing cargo conditions. The parameters to be taken into account are the tightest possible packing, weight distribution, avoiding damage, and processing all cargo items on time. Time is a crucial factor, which is why optimization potential cannot be fully exploited by individuals.

Generating training data for the AI

“We have developed an AI algorithm to avoid the disadvantages mentioned above and to make better calculations with existing information,” says Geiger, an engineer and logistics specialist. The algorithm is able to recognize patterns and react to them. The AI leaves nothing to chance: even minor factors such as the size and weight of packaging film or transport nets are taken into account, and in the end, the positions of all ULDs are known.

Data processing plays a special role here: on one hand, machine learning is used to identify a denser packing structure under the given constraints of the aircraft used and the total cargo. On the other hand, the IFPT has encountered a prominent problem in AI research: data availability. To this end, real environments are to be simulated in order to generate training data for state-of-the-art AI applications, thus creating possibilities for generating data that would otherwise be difficult to obtain. These are to be used to further improve the evaluation by stochastic methods in areas where analytical methods in the processing of sensor values reach their limits.

So far, Felix Geiger and his team have only tested the algorithm virtually on the computer. However, there is already an idea for a packaging station where the software can prove itself at the end. A random generator will provide the packages and it will show whether all of them can be optimally placed on the ULD. An even more far-reaching idea already exists: “In a follow-up project, we would like to use our software to optimally pack an entire aircraft,” hopes IFPT scientist Geiger.

More information:

Find out more about the project eCargo on the website of the Institute for Aircraft Production Technology