Implementing Machine Learning in Earthquake Engineering

The use of machine learning across many fields has seen a rise in recent years, from life and physical sciences to finance and athletics. Within the physical sciences, it is just starting to see some implementation in the field of Earthquake Engineering. The objective of this paper is to implement machine learning to Earthquake Engineering data to create more literature in the field. In particular, this project aims to implement predictive models to properly capture the residual displacement of a structure caused by an earthquake using acceleration data. Current methods, which involve double integration of the acceleration data with a combination of baseline correction and filtering, do not do a good job at capturing residual displacements. For this reason, machine learning was investigated as a possible alternative to numerical integration. The results showed that Feedforward and Recurrent Neural Networks are not able to pick up the residual displacement. In addition, it was found that ground displacement was an important feature to get reasonable results. More research needs to be done on this topic before discarding neural networks as a possible solution for obtaining residual displacements from acceleration data from an earthquake.