Sheet metal assortment optimization with k-Means

This paper feals with inventory optimization in the procurement of sheet metal for small series and contract manufacturers. An AI-based algorithm was developed and tested for a real-world application. The underlying use case is a variant of the cutting stock problem. The objective is to optimize the number of metal sheet variants to be source to reduce inventory and handling costs. Therefore, demand for metal sheets with different dimensions are combined so that large steets are sourced from which the sheets required for production are cut, Here we have a trade-off between the savings in inventory management versus the additional costs due to scrapping of cut losses. Decision models for this application case which take all these factors into account have not been developed so far. This paper presents a first approach by modelling parts of the problem as a linear integer problem which is heuristically solved with a k-Means clustering algorithm. With this model different scenarios can be simulated for making better inventory decisions. This approach was implemented in a software application based on Power BI and the Python library scikit-learn and tested in practice by a company. The test showed considerable savings potential for the use case and provided important stimluli for further research, especially regarding further modelling approaches ans search algorithms.