A new low-carbon design method based on multi-agent interactive reinforcement learning

Zi-li Wang,Shu-you Zhang,Le-miao Qiu
DOI: https://doi.org/10.1177/0954406218762944
2019-01-01
Abstract:With the development of manufacturing industry, considerable attention has been paid to the issue of environmental problems caused by the manufacturing process. One of the most important methods to solve this kind of problem is the low-carbon design of the products before manufacturing process. Therefore, a new part structure low-carbon design method is proposed in this paper. At first, the layered structure model based on part features is constructed in order to divide the product into part features, and as a result, the carbon footprint is calculated for each part feature. Then, the part features are clustered into several classes looking forward to find the higher carbon emissions part features as initial features for low-carbon design. After that, the multi-agent interactive reinforcement learning method is performed on the topological path of multiple initial parts using the Q-learning algorithm in Markov environment, so as to meet the design requirements of low-carbon design. Finally, a 6400 kN injection molding machine moving template design was taken as an example to verify the effectiveness of the proposed low-carbon design method.
What problem does this paper attempt to address?