Learning-Based Metaheuristic for Scheduling Unrelated Parallel Machines With Uncertain Setup Times

Chen-Yang Cheng,Chieh-Wen Chang,Shu-Fen Li,Pourya Pourhejazy,Kuo-Ching Ying

doi:10.1109/access.2020.2988274

Abstract

Setup time consists of all the activities that need to be completed before the production process takes place. The extant scheduling predominantly relies on simplistic methods, like the average value obtained from historical data, to estimate setup times. However, such methods are incapable of representing the real industry situation, especially when the setup time is subject to significant uncertainties. In this situation, the estimation error increases proportionally to the problem size. This study proposes a Random-Forest-based metaheuristic to minimize the makespan in an Unrelated Parallel Machines Scheduling Problem (UPMSP) with uncertain machine-dependent and job sequence-dependent setup times (MDJSDSTs). Taking the forging industry as an example, the numerical experiments show that the error percentage for the setup time estimation substantially decreases when the proposed approach is applied. This improvement is particularly significant when large-scale problems are sought. Overall, this study highlights the role of advanced analytics in bridging the gap between scheduling theory and practice.

Highlights

In a saturated market place, the strategic intention has been directed towards customer satisfaction, where product on-time delivery is a principal attribute
To effectively model the scheduling problems, different aspects need to be taken into consideration among which, time parameters uncertainties play a significant role
The extant scheduling problems rely on simplistic methods for the estimation of setup times, the inherent errors caused by inaccurate approaches may result in significant delays in the production plans

Summary

INTRODUCTION

In a saturated market place, the strategic intention has been directed towards customer satisfaction, where product on-time delivery is a principal attribute. The setup time estimation in practice is predominantly based on the operators’ experience, or, at best, simple average-based methods Such approaches are not effective in the production systems with general-purpose machinery, and the possible errors can be extensive when estimating the delivery due date for large-size orders. Sequence-dependent setup time is a prime example when the uncertainties prevail [6] because the estimation depends on the current job, and the job immediately preceding it, and the characteristics of the respective machine [2] This situation is prevalent in the Unrelated Parallel Machines Scheduling Problem (UPMSP). Advanced analytics can help address the uncertainties involved in the mathematical model parameters, time value estimations [12] Inspired by this idea, this study sought to integrate a learning-based method into scheduling optimization problems to help narrow the gap between research and practice.

LITERATURE REVIEW

HYBRID ARTIFICIAL BEE COLONY

NUMERICAL ANALYSIS

Findings

CONCLUSION