Machine Learning-Based Low Cycle Fatigue Techniques for Reinforcing Bars

Presented By: Islam Mantawy, Rowan University

Description: This presentation will include machine learning techniques to predict the vulnerability of reinforcing bars to fracture in concrete rocking structures obtained from experimental data from large scale shake table testing from large scale testing of 2-span bridge specimen conducted as structural health monitoring using recorded time-series data recorded from structures during extreme events using convolution neural networks (CNNs). Often times, training CNNs with time-series data into frequency domain or converting into histograms. Even though the reported nature of the time-series data and damage accumulation due to strain cycles. ML-SHNM approach was developed by converting the input strain data (time-series_ into input images using Markov Transition Field (MTF). Then, the encoded images were used to train and test CNN models. Convolutional neural network models trained on strain data performed with 100% accuracy during the training phase and more than 97.8% for the testing phase. This approach enables future research to develop robust machine learning based low cycle fatigue techniques for any structural component and material type which will help overcome the drawbacks of existing low cycle fatigue models.