Approaches for the Prediction of Lead Times in an Engineer to Order Environment—A Systematic Review

Abstract

The interest of manufacturing companies in a sufficient prediction of lead times is continuously increasing -especially in engineer to order environments with typically a large number of individual parts and complex production processes. A multitude of approaches have been proposed in the literature for predicting lead times considering different data and methods or algorithms from operations research (OR) and machine learning (ML). In order to provide guidance at setting up prediction models and developing new approaches, a systematic review of the available approaches for predicting lead times is presented in this paper. Forty-two publications were analyzed and synthetized: Based on a developed framework considering the used data class (e.g. product data or system status), the data origin (master data or real data) and the used method and algorithm from OR and ML, the publications are classified. Based on the classification, a descriptive analysis is performed to identify common approaches in the existing literature as well as implications for further research. One result is, that mostly order data and the status of the production system are used for predicting lead times whereas material data are used seldom. Additionally, ML approaches primarily use artificial neural networks and regression models for predicting lead times, while OR approaches use mainly combinatorial optimization or heuristics. Furthermore, with increasing model complexity the use of real data decreased. Thus, we identified as an implication for further research to set up a complex data model considering material data, which uses real data as data origin.

Machine Learning-Based Prediction of Missing Components for Assembly – a Case Study at an Engineer-to-Order Manufacturer

“…ANN and RF have already proven successful in including material data in the prediction model. When looking at the data origin, the authors of [24] also identified that the use of real data strongly decreases with an increasing number of considered data classes. Thus, with increasing complexity of the prediction model they identified a lack of models using real data.…”

“…A systematic literature review conducted by BURGGRÄF ET AL. [24] has analyzed existing approaches focusing on the prediction of lead times in the research fields of ML and OR and classified them according to the three criteria data class, data origin and used method/algorithm. Looking at the data class, the authors identified that the majority of publications examined use order data and information about the system status of the production system (see, for example, [25,26]).…”

“…According to BURGGRÄF ET AL. [24], there is a lack of prediction models for machine and plant manufacturers that use the primarily used data classes and material data for the prediction of lead times. ANN and RF have already proven successful in including material data in the prediction model.…”

“…Here we compared the performance of a linear regression (LR), a DT-regressor, a RF-regressor, an AdaBoost-regressor utilizing an DT-regressor as base estimator, a GB-regressor and an ANN, since they are established approaches for regression which have already been successfully applied in lead times prediction [24]. For the ANN, specifically, a singlehidden-layer feedforward net with a rectified linear unit as activation function utilizing a Keras regressor was applied.…”

“…The data model on all four levels of detail consists of the four data classes order data, machine data, material data and system status and thus follows the recommendation of the authors in [24]. For the data acquisition, we followed the procedure of FAYYAD ET AL.…”