Research On Critical Region Extraction Algorithm In Casting CAE Post-processing Based On Deep Learning

In the process of casting simulation analysis,the existing post-processing method overrely on the researchers' ability.Because researchers cannot analyze the simulation results accurately all the time,which may miss some details or even lead to wrong analytical conclusion.In this paper,the researcher's analysis in post-processing is tracked by deep learning methods,and thus extract the critical region for the same type casting process.And this ability will be adjusted dynamically by online machine learning method.This paper aims at providing researchers with the guidance for simulation analysis,and thus improving the analysis efficiency.Firstly,in order to deal with the three-dimensional characteristics of the simulation data,the extraction technology for three-dimensional voxel based on direct rendering is proposed.With the help of this technology,researchers can grasp the overall information of the simulation data and extract several voxel blocks from it,which forming the training data sets for deep learning.Secondly,the three-dimensional convolution neural network is used to divide the network training part of the critical region extraction algorithm into two stages: pre-training and online learning.The network model gained from the pre-training has the basic ability to identify the critical region.And the online learning network dynamically adjusts its parameters to meet the requirements of different simulation analysis researchers,which will also provide a carrier for keeping the researchers' experience accumulated form long-term analysis.Thirdly,inspired by the existing image detection methods,layer-by-layer detection algorithm based on the concept of space partition is proposed.This algorithm can reduce the number of regions that should be detected in the process of extracting the critical region without diminishing the accuracy,and therefor speed up the process.Finally,locomotive end cover,shell and the mounting bracket are taken as examples to verify the practicability of the critical region extraction algorithm based on deep learning.The critical region extraction algorithm can improve the casting simulation analysis workflow,provide the guidance for the analysis of the researchers,and thus improve their efficiency of the simulation analysis.At the same time,the algorithm also keep the experience of the researchers' long-term simulation analysis.