Executive Summary : | Surface wave-based methods like spectral analysis of surface wave (SASW) and multi-channel analysis of surface waves (MASW) have become popular tools for site characterization in the last two decades. However, these methods face challenges such as generating a multi-modal dispersion curve, assessing the accuracy of the MASW test, and using theoretical curves. Additionally, numerical simulation methods like FEM/FDM are not preferred in surface wave analysis due to their time-consuming nature. The inversion problem is known for its inherently ill-posed, non-linear nature, making it difficult to find a unique solution. Traditional inversion methods only perform satisfactorily when a-priori information is available, making the MASW method less reliable, especially for irregularly dispersive soil profiles. This study aims to develop a limited channel surface wave test methodology that uses half to one-third of the number of geophones required by the traditional MASW method while generating a high-resolution multi-modal dispersion image. The proposed method also introduces a novel semi-analytical elastic wave filed modelling technique to simulate surface wave propagation through layered half-space, taking under 2 seconds to simulate. A convolution neural network architecture with over 25 million parameters is employed to predict depth and shear wave velocity of different subsurfaces without a-priori information. This new method reduces equipment cost by 50%-75%, requires 99% less CPU time, and accurately predicts both depth and shear wave velocity of soil layers without any a-priori information. |