Non-parametric ground motion model for displacement response spectra and Fling for Himalayan region using machine learning

Displacement response spectra (DRS) are crucial for seismic design as earthquake damage correlates more with displacements than forces. Previous efforts to develop attenuation relations for DRS have been largely approximate. Permanent displacement or Fling poses significant design, repair and rehabilitation challenges. Consideration of DRS and Fling in seismic design and performance assessment necessitates its accurate estimation. This paper presents the two Artificial Neural Network (ANN)-based non-parametric Ground Motion Models (GMMs). The first model predicts DRS for horizontal and vertical spectral ordinates. The second model focuses on predicting the Fling step in fault-parallel, fault-normal and vertical components. Both the models are developed for the Himalayan region. Given the limited availability of recorded data, ground motion recorded in tectonically similar regions is also utilized to develop DRS GMM. The sparsely recorded Fling data in the Himalayan region is supplemented by additional Fling values simulated using a physics-based approach, alongside data recorded from tectonically similar regions. The simulated Fling values are validated against recorded Fling data. The performance of developed GMMs is compared with existing GMMs and seismic codes which demonstrated its satisfactory performance. The correlation coefficient for ordinates of DRS and Fling are reported to be greater than 0.86 and 0.80, respectively.