Second International Workshop on Symbolic-Neural Learning (SNL-2018)

July 5-6, 2018
Nagoya Congress Center (Nagoya, Japan)

Prediction Forging Dies via Generative Adversarial Networks for Pairs in Sequences

Hayato Futase (TTI), Yuta Hayashida (TTI),Tomoki Tsujimura (TTI), Makoto Miwa (TTI), and Yutaka Sasaki (TTI)

Abstract:

Computer-aided design of forging dies in die forging processes has been demanded in the field of manufacturing. Simulation-based analysis is accurate, but it is time-consuming and infeasible to search for ideal designs of dies in practice. We instead treat the designing task as a task of computer vision. We propose a novel method to predict intermediate shapes or shape images, which correspond to forging dies, from target shapes using a novel generative adversarial network (GAN) architecture for a pair in sequential data called shares a latent vector for GANs among the shapes in a die forging process, and it controls the time-series changes of shapes in die forging processes using labels that represent the degrees of progress. Our method consists of two modules: an encoder and a decoder. The encoder estimates a latent vector of a process from the target shape of the process and its label. The decoder is based on and it generates intermediate shapes from the latent vector and the labels. In the experiments, our method is applied to a data set of sequential changes of shapes in die forging processes. The experimental results show that achieves about 30% reduction in the volume variance compared to a conditional GAN that does not consider sequential changes. The generated shapes also show the possibility of predicting the intermediate shapes with our method.