Optimizing pallet packing decreases the resource use and cost of transporting goods. A classical optimization approach is bin packing. It assumes knowledge of item dimensions, is computationally demanding, and has many practical complications.
We introduce an alternative approach that relies only on historical data of packed pallets and item identities. For a new order to be packed, the method first produces pallet templates from the data. Given the templates, a discrete optimization problem is solved to find the optimal item-to-pallet assignments.
In addition, we designed a web-based user interface for logistics experts to examine and fine-tune the result. The user interaction is captured in the background to enrich the pallet template data, teaching the method better solutions.
The application is deployed at a furniture maker, integrated into their enterprise resource planning software. Its maintenance is easy.