13.09. - 14.09.2023 Dresden
M. Sc. Jana Philips
An approach to optimization-based order release in engineer-to-order systems – model derivation and performance evaluation
Ralf Gössinger, Jana Philips
In engineer-to-order (ETO) systems, some of the incoming orders are characterized by a specification that is only completed at an unknown point in time after the order has been accepted. Therefore, in addition to the uncertainty about the order arrival, production planning is confronted with uncertainty about the order specification. This also affects the planning of order releases, which as a subtask of production planning is responsible for initializing the execution of orders that have already been accepted. Two sources of order-related uncertainty become relevant here, namely the period in which the order specification will be completed (specification time uncertainty) and the manufacturing process (capacity requirements uncertainty). In this situation, opposing ways of order release can be used to cope with this uncertainty. On the one hand, order release could be limited to orders with complete specifications. This would make release plans more robust, but waiting for the complete specification would also shorten the time between order release and delivery date. Consequently, it would be more expensive (e.g. overtime and backorder costs) to balance capacity supply and demand. On the other hand, order release could include orders with both complete and incomplete specifications. As a result, delayed completions of specifications could require a replanning of order releases and thus increase inventory holding costs (WIP, FGI). However, the processing of orders whose specification is completed by the expected date could start earlier, since the required capacities have already been reserved. Hence, it would be less costly to balance capacity supply and demand.
In this context, the research question arises as to whether it is possible to increase the performance of order release by considering the incompletely specified orders in addition to the completely specified ones. This gives rise to several sub-questions: What modifications need to be made to order release planning approaches developed for make-to-order (MTO) systems to account for the different types of uncertainties and information updates? Which formal changes need to be made to the optimization model? How can the performance of the proposed planning approach be measured? What are the (dis-)advantages of the proposed planning approach compared to an approach that ignores incompletely specified orders?
To answer these questions, an approach to order release planning in MTO systems is further developed in such a way that it considers completely and incompletely specified orders as well as information updates. Incomplete specifications are taken into account by dividing the orders into components (work packages), so that completely specified components can be finally released, whereas incompletely specified components are provisionally released. For incompletely specified components, more information is provided at a later, yet unknown point in time. These information updates, as well as those related to order arrival and fulfillment are incorporated into planning using a rolling horizon approach. As a result, the plan is continuously adapted to the actual situation.
For reasons of solvability, the planning problem is hierarchically decomposed into two sub-problems that aim to minimize the sum of inventory holding (WIP, FGI) and backorder costs as well as costs of providing additional capacity (e.g. for overtime). The first sub-problem is to determine the amount of additional capacity for those periods in which capacity requirements are expected to exceed standard capacity. The second sub-problem deals with planning the release periods of the individual order components, assuming that the planned additional capacity is available. Both sub-problems are modelled as MILP, in which the uncertainties concerning specification period and capacity requirements of a component are considered with chance-constrained estimations. Therefore, for each uncertainty type, a probability threshold needs to be set in advance.
In order to receive feedback on the implementation of the order release plan, the processing of orders released for the frozen period is (re-)scheduled in detail using priority rules (FIFO or SPT/SL). The orders are then executed (by simulation) taking into account the realizations of the random variables of specification period and capacity requirements. As soon as the end of the frozen period is reached, information on order fulfillment and capacity utilization is reported back to the related sub-problems.
An extensive full-factorial numerical study is conducted to measure the performance of the proposed approach in terms of costs, robustness and solution time. Considering robustness is very important to evaluate the performance of the proposed approach compared to a direct order release approach restricted to completely specified orders. Therefore, two dimensions of robustness are observed. While cost robustness is measured by the relative deviation of the actual from the planned cost, plan robustness is based on the number of changed component releases. To gain deeper insights into the influence of considering ETO-specific uncertainties, a special focus is placed on the influence of the probability thresholds to be set in advance. As these estimates also relate to capacity requirements, different levels of capacity requirements are considered. In addition, the number of incoming orders and the priority rules for detailed scheduling are varied systematically. Regression analyses shed light on the impact of these factors on the observations of both the proposed and direct order release approaches.