A machine learning approach to optimise the usage of recycled material in a remanufacturing environment

Remanufacturing has acquired importance in recent years because of the increasing environmental concerns of manufacturing processes that deplete the Earth's resources. Some examples of remanufactured products are automobile parts, furniture, photocopiers, and computer printers. In a remanufacturing setup, raw materials are drawn from two sources: (i) 'cores', which are obtained from recycled products, and (ii) 'non-recycled' or unused materials, which are produced from minerals freshly mined from the earth. An important decision for the manager is to select material optimally from these two sources. Using cores has environmental benefits, and because they are cheap, they reduce manufacturing costs. However, their use generally increases the production time, because of the additional pre-processing usually needed, which can negatively impact service levels. When the supply of finished products is running low, to satisfy service levels, it makes sense to use unused material. This research focuses on identifying an optimal strategy of switching between the two sources of material. A reinforcement learning algorithm is used to solve the switching problem. The switching algorithm produced encouraging results, showing up to 65% cost improvements over a policy that uses only unused materials.