Railway Embankment Quality Control Based on Feature Extraction by Singular Value Decomposition and Bayesian Inference

We developed an estimation method for the stress-strain curve of soil using the observational information of embankment that can be obtained on site. In the proposed method, the stress-strain curve is estimated using Bayesian linear regression with observed information after summarizing the data from triaxial compression test results and grain size analysis results for various soil materials by singular value decomposition. This method makes it possible to predict the ground strength characteristics such as the maximum internal friction angle and the residual internal friction angle from the deformation characteristic information of the elastic region, such as the coefficient of subgrade reaction. Moreover, this method allows the prediction of the deformation performance of the soil over a wide range of strain levels. It has the potential to provide immediate confirmation of the performance of railway embankments during construction. In addition, this method can be used to evaluate the seismic resistance of existing embankments and design seismic reinforcement.

Keywords: Bayesian Inference, Singular value decomposition, Ground strength characteristics, Embankment.