Concepts to Improve the Quality of Production Plans using Machine Learning

Abstract

There are always deviations between production planning and subsequent execution. Furthermore, it has been found that the reliability of production plans and thus Planning Quality (PQ) can drop down to 25 % in the first three days after plan creation [1]. These deviations are caused by uncertainties, such as inaccurate or insufficient planning data (including data quality and availability); inappropriate planning and control systems; and unforeseeable events. Production planners therefore use buffers in the form of inventories or extended transitional periods to create possibilities for implementing corrective measures in production control. Buffers, however, lead to increased coordination and control efforts as well as to negative effects, particularly on the inventory, throughput time, and capacity utilisation. The potential for more accurate planning remains largely unexploited. The objective of this paper is to investigate the possibilities of increasing planning quality. Within a case study, the authors demonstrate how machine learning can be used to predict cycle times. Furthermore, the increased accuracy compared to the current method is shown. Based thereon, two approaches are presented, focusing on the reduction of gaps between the master data and predicted data used during the production planning process. Moreover, further research needs are identified.