Online optimizing hot forming parameters for alloy parts based on action-dependent heuristic dynamic programming

The microstructural evolution is complex and time-varying in the practice hot forming process of alloy parts. Therefore, accurately online optimizing forming parameters and controlling microstructural evolution are the urgent tasks. In this work, an action-dependent heuristic dynamic programming (ADHDP) is developed to online optimize processing parameters during hot deformation of alloys. The ADHDP is based on adaptive dynamic programming and only contains action/critic neural networks (ANN/CNN). ANN is utilized to determine the next control signals (processing parameters) of hot deformation, while CNN is used to obtain the approximation of cost-to-go function. The weights of these neural networks are online-updated by back-propagation algorithm. The control goals include the recrystallization volume fraction and average grain size. The proposed ADHDP method is successfully applied to online optimize processing parameters of GH4169 superalloy during hot deformation, and its effectiveness is verified by numerical simulations. According to the optimized processing parameters, the hot compressive tests of GH4169 superalloy are conducted to further verify the validity of the developed method. Furthermore, the microstructures, which are obtained by the proposed method, are more uniform and fine than those obtained by the traditional processing without online optimization.