With an innovative system for the fully automated identification of pallets, the researchers in the »Identification Service based on Natural Features« development project add another component to the Silicon Economy. An intelligent algorithm identifies a pallet – among 500 million of the same kind – by its grain alone, making barcodes or RFID chips obsolete.

Identification processes for load carriers are still very time-consuming and personnel-intensive today as they are mainly based on barcodes. Silicon Economy researchers have now developed an alternative method for identifying pallets which speeds up identification processes considerably. Like all Silicon Economy developments, the special components are available as open source. Identifying shipping units in the warehouse or along the supply chain is one of the standard processes in logistics for which largely standardized solutions are already available today. However, this new Silicon Economy development helps companies to optimize their processes, because the new system is as sustainable as it is efficient.

The time spent on printing and attaching labels can be saved as well as the time for identifying pallets. By omitting non-value-added activities and costs, processes in the warehouse become significantly faster. Moreover, the number of incorrect deliveries due to illegible labels, which are subject to high levels of wear due to frequent transport, is reduced. 

Algorithm trained with a quarter of a million images

The whole process becomes possible by using artificial intelligence. »For the idea of identifying load carriers according to their external characteristics is actually not new«, says Julian Hinxlage, product owner in the project entitled »Identification Service based on Natural Features«, or »NaturIdent« for short. »But only now do we have the technological capabilities in machine learning and correspondingly powerful hardware with which we can enter a new dimension of object identification«. The starting point for this idea was that a pallet – like a human being – has an unmistakable fingerprint. The researchers generated this fingerprint using images of the six external pallet feet on the long sides, which vary in appearance depending on the wood grain and minor damage to the wood.

At the heart of the system is a pallet identification algorithm that is “fed” and trained with the existing images. 10 images per block were taken to train the algorithm. In normal operations, only one image per block is required. The algorithm with which »NaturIdent« is breaking new ground is the result of a research process lasting several months. Many different approaches must be tried out to increase identification rates. In the meantime, the scientists have produced the proof-of-concept which was started by creating 5,000 images themselves.  Currently, the re-identification rate is over 99 per cent. However, for a credible statement on the ultimate reliability of the algorithm, many more image data with heterogeneous environmental conditions, such as illumination, distance, and angle, are required.  

In the meantime, the data base contains more than a quarter of a million images – thanks to data generated in the industry. Here, the algorithm regularly receives new nourishment – i.e., the data set is being enlarged and expanded as part of companies’ day-to-day business. Companies also have older pallets on the road or those where pallet feet have been replaced. Moreover, lighting conditions when scanning the pallets can also be very different from those under test conditions. Two large companies – Robert Bosch GmbH in Karlsruhe and Arvato Systems GmbH in Gütersloh – have already shown interest in the process and are considering applying this technology across several locations and with delivery situations. For the time being, however, the focus will be on image data generation at the two sites, which will be used to continue training the algorithm.

Demonstrator illustrates reliability of the process

A new demonstrator now illustrates the process very vividly: This demonstrator, which is also to be shown at trade fairs, consists of two vertical pallet storage units – one for goods in and one for goods out –, two horizontal conveyor lines, one transport trolley and a camera system. Any pallet can be removed from the pallet storage units and pulled onto the conveyor line. There, the pallet is detected by a camera system, the signature is created quickly and reliably and ultimately identified at the recipient. 

The process works both within a company, e.g., in the warehouse, as well as across several company sites and along the entire supply chain. »The major advantages of the process for companies are its flexibility and low cost«, says Julian Hinxlage. »In this way, the camera system can simply be installed at the conveyor or at the truck gate. The images are captured incidentally as it were – without having to change any processes and without the pallet having to stop in front of the camera.« 

Besides increasing efficiency in the company, new applications and business models will also be possible on the basis of smart pallet identification. In addition to determining key figures for load carriers such as circulation speed, damage rate and pallet service life, the signature can be used as a serial number to combine objects with the load carrier. By marrying the pallet up to the goods, additional labels become unnecessary.

>> Click here for the »Identification Service based on Natural Features« project profile.