Prescriptive cost analysis in manufacturing systems

Abstract

In many industries there is a high probability that production lines might produce extreme quantities, far away from what would be expected and/or desired considering the production and market conditions. This impacts considerably in product costs and respective margins. Thus, to support planning and decision-making in manufacturing systems, new analytical approaches must emerge to enable the transformation and representation of the available data. Namely, mathematical modeling can be used to improve cost analysis and optimization. This paper presents and discusses the use of prediction models, based on supervised machine learning algorithms and, particularly, linear regression in the context of manufacturing systems. Multiple linear regression models can be trained for different time periods allowing a better control of costs. They can be also used as decision-making tools for subsequent periods. Data from September 2020 to June 2021 of a first-tier supplier of the automotive industry was used considering two different training periods, in two groups of production lines of a specific product. Results of R2 and MAE validate and show the relevance of the proposed models. The accuracy of these models depends on the artificial intelligence techniques and the training periods.