Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : November 02, 2019 - November 04, 2019
The load hysteresis loop is observed when fatigue loading is applied. Stress-strain hysteresis is an important characteristic as a material parameter that greatly affects analysis accuracy in extremely low cycle fatigue evaluation. The kinematic hardening rule and the isotropic hardening rule for expressing the hysteresis loop deviate greatly from the experimental values in the range of the elastic deformation part and the maximum and minimum loads. In addition, there are many unmeasurable parameters in these models. For this reason, we try to obtain approximate functions and models of load hysteresis loop using machine learning, which has been spreading rapidly in recent years. We will examine whether it can be replaced with an elastoplastic constitutive equation. The objective of this study is to express the elasto-plastic calculation with higher accuracy than the previous constitutive equation. At the present stage, function fitting is possible but a physical model is not obtained. Therefore, a physical model such as kinematic hardening rule, or isotropic hardening rule should be modeled into the neural network configuration. It is necessary for hysteresis representation to configure Autoencoder(AE) and Long Short-Term Memory(LSTM). AE learn a representation (encoding) for a set of data, typically for dimensionality reduction, by training the network to ignore signal noise. LSTM learns time sequential data by adding output to its input. Using these neural network architecture. The objective of this work is establishment for expression of the load hysteresis loop.
