Modeling crack growth of aluminum alloy under variable-amplitude loading using dynamic neural network

摘要

In this article, a new dynamic neural network is developed to model the crack growth process of aluminum alloy under variable-amplitude loading condition. The proposed neural model is incorporated with the feedback loops at the output layer. This model can thoroughly address the dynamics as well as the relationship between the crack growth and variable-amplitude cyclic stress excitations of aluminum alloy. Therefore, the model's capability of capturing the dynamics of crack growth is enhanced due to this feedback loops. Based on extreme learning machine (ELM) algorithm, the input weights are uniformly randomly selected in a proper range, then the output weights and the feedback parameters can be globally optimized by utilizing batch learning type of least square technique. The simulation is conducted based on the experimental data of the 7075-T6 aluminum alloy to demonstrate the capability of the proposed neural model. The prediction performance of crack length under variable-amplitude loading is confirmed very well.