Material flow control in Remanufacturing Systems with random failures and variable processing times
Felix Paschko, Steffi Knorn, Abderrahim Krini, Markus Kemke
Abstract
Abstract Material flow control in remanufacturing is an important issue in the field of disassembly. This paper deals with the potential of autonomous material release decisions for remanufacturing systems to balance the uncertainties related to changing bottlenecks, to maximise throughput ( $$TH$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>TH</mml:mi> </mml:mrow> </mml:math> ) and to minimise work-in-process ( $$WIP$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>WIP</mml:mi> </mml:mrow> </mml:math> ). The goal is to achieve the highest possible throughput rate using real-time data while keeping costs to a minimum. Unlike traditional production systems, remanufacturing must consider and handle high uncertainties in the process. Up to now, classical methods such as CONWIP, Material Requirement Planning (MRP) and Kanban have been used for material flow control. However, these methods do not perform well in a system with high variation and uncertainties such as remanufacturing as they aim to find solutions for static environments. Crucial for optimal production in stochastic environments is finding the optimum pull or release rate which can vary over time in terms of maximising $$TH$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>TH</mml:mi> </mml:mrow> </mml:math> and minimising $$WIP$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>WIP</mml:mi> </mml:mrow> </mml:math> . We propose a deep reinforcement learning approach that acts on the environment and can adapt to changing conditions. This ensures that changing bottlenecks are taken care of and that there is a minimum $$WIP$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>WIP</mml:mi> </mml:mrow> </mml:math> in the system.