Assessment of Effect of Strain Amplitude and Strain Ratio on Energy Dissipation Using Machine Learning

Abstract

In this study Artificial Neural Networks (ANN) as a machine learning technique is used to predict and assess the effect of strain amplitude and strain ratio on energy dissipated in steel reinforcing bars in reinforced concrete members. The steel reinforcement bars were experimentally tested and were subjected to variable strain amplitudes beyond yield. The developed machine learning model, which is based on Back-Propagation ANN, accurately predicted the experimentally measured dissipated energy. The developed model is then used to deeply assess the effect of a range of strain amplitudes and strain ratios in the amount of energy dissipated at the first cycle, in an average of selected number of cycles and in all cycles, all at different levels of low-cycle fatigue loading of the reinforcement bars. It is concluded that the developed machine learning model can accurately predict the hysteresis energy dissipated in steel bars subjected to low-cycle fatigue load and more importantly it is a viable machine learning tool for deep assessment of the tested specimens with several parameter values that were not covered by the experimental program, but within the domain bounded by the maximum and minimum values of the training data. Based on the prediction and the deep assessment results, several conclusions were drawn.