Research On Deep-reinforcement Learning Of Process Planning For Energy-Efficient Flexible Machining

Manufacturing industry is widely distributed and consumes a significant portion of total energy consumption of the industry sector.Enhancing energy efficiency of the manufacturing industry becomes a research hotspot for manufacturing industrial practitioners to maintain the sustainable competitive advantages.Flexible machining system,developed in the background of contemporary manufacturing,is a typical system with computer numerical control machine tools or machining centers as mainstays.It operates closely in relation with machine tools,workpieces and cutting tools.During the machining process,it will consume a huge amount of electrical energy,but the energy efficiency is much low.Configurations of flexible machining system become much dynamic and changeable,more seriously,the impact effects of machining configuration parameters on energy efficiency come to be very complex.Therefore,how to reduce the energy consumption and improve the energy efficiency of the flexible machining system is an imperative research area worth of further investigation.With the support of the National Natural Science Foundation of China(No.51975075&51475059)and the National Key Research and Development Project(No.2017YFF0207903),this research investigates the impact factors of energy efficiency in flexible machining process,reveales the relation among machining configurations,process parameters and energy efficiency and achieves energy-efficiency process planning under various configurations.The detailed research points are elaborated as follows.Firstly,energy characteristics analysis of flexible machining system is explicitly analyzed with a comprehensive consideration of the influence of cutting tools,workpieces,cutting fluid and process parameters on energy consumption.Technical difficulties and challenges are given when conducting process planning for flexible machining system where machining configurations always frequently change.On the basis of that,a process optimization framework is proposed for energy efficiency improvement.Secondly,by using deep learning methods,such as conventional neural network,stacked auto-encoder and deep belief network,and the process and energy monitoring data collected on daily basis,the mapping relationship model between configuration parameters and energy efficiency are established.The comparative study is conducted particularly in terms of the sizes of the data,data collection duration,feature selection,algorithm performance,etc.Through the detailed multi-perspective analysis,the recommended energy efficiency modeling methods in different application scenarios are obtained.Thirdly,a parametric optimization model for flexible machining is developed with the consideration of machine flexibility,cutting tool flexibility and workpiece flexibility to control cutting speed,feed rate and cutting depth for improving energy efficiency under different machining configurations.The optimization problem is formulated as Markov Decision Process in which action space,state space and reward function are specifically designed.A reinforcement learning based parametric optimizer is designed with critic and actor networks for flexible machining under various machining conditions.To increase the generalbility of optimizer,meta-learning is used to realize mix training.Case study is carried out in machining practice to validate the proposed method compared with traditional evolutionary algorithms.The parametric optimization principles under various machining configurations are revealed.Fourthly,by considering the machining task flexibility and resource flexibility,a process route optimization model is established for energy saving.Based on the graph theory,process routing graph is developed which contains operation sequencing and allocation of machine tools and cutting tools.The key information of process route graph and candidate scheme is extracted by deep graph conventional network,based on which the task-parallel flexible process planning method is designed using gradient search reinforcement learning and multi-task training algorithm.Case studies are designed with a consideration of the changes of workpiece features and resources in flexible machining process to validate the proposed optimization model and algorithm.Finally,the support system of energy efficient process planning for flexible computer numerical control machining is elaborated including overall framework,energy efficiency online monitoring system,integration with manufacturing executive system and the framework of data acquiring.An application case conducted in a company is used to verify the effectiveness and practicality of the proposed method.