Structure-Property Linkage of Packed Soil Particles

Team Members

  • Jie (Jessie) Cao - Civil and Environmental Engineering
  • Mahdi Roozbahani - Computational Science and Engineering

Motivation and Objective

Soil structure is the study object of interest for this project, which is related to our own research. Also, hydraulic conductivity, a key transport property of soils, is the property of interest.

Hydraulic conductivity describes the ease with which the fluid can move through pore spaces. To quantify this property, experimental tests are cost expensive and difficult to conduct. Additionally, due to the complex nature of soils, available empirical solutions have restricted applications. It cannot understand the impact of internal structure very well. Numerical simulations are very computational expensive.

As such, our project objective is to develop a fast, rigorous approach to quantify hydraulic conductivity of packed soil particles based on acquiring thorough microstructural information.

Challenges and Trials

What packed soil particles samples should we work on?

Our Initial idea was to only consider randomly packed mono-sized structures, but it cannot reflect the complex nature of soils and thus would lead to a very restricted application of the project results. It was necessary for us to consider soil particles with different sizes and different shapes to see how these parameters affect hydraulic conductivity.

Accordingly, we take four different packed soil structures into account: Mono-sized, binary-sized and multi-sized samples generated by Gravitational Sphere Packing Simulation (GSP) and one real soil sample reconstructed from 2D optical microscopic images. After obtaining the soil structures to be examined, we randomly sampled a total of 703 subvolumes as our fully datasets.

Packed Soil Particles Samples

How to obtain hydraulic conductivity data?

Lacking of experimental data for hydraulic conductivity, so we need to acquire the data using numerical method. At the beginning, we tried to use Lattice-Boltzmann method, a relatively new and advanced simulation technique for complex fluid systems. Details of this project description can be found in our first post Project Progress I post.

Based on this plan, we spent about one month struggling with the LBM simulation since neither of us had the experience in this simulation. We obtained the code from the website, modified it to meet our requirement and talked with other professor who is using LBM, but we had the problem on the unit conversion (lattice units to physical units). Details of the progresses for LBM simulation can be found in our posts Project Progress II and Unit Conversion from Lattice Units to Physical Units.

Then we tried FEM program COMSOL Multiphysics, which was suggested by one classmate, to calculate permeability. We spent one week on this trial, but still couldn’t acquire useful results. The problem was that when structure geometry was imported to COMSOL, the mesh creating errors always occurred.

At the end, with the help of Ahmet, we figured out a way to numerically calculate hydraulic conductivity using Finite Volume Method. It is on the basis of incompressible single-phase flow moving through porous media.

FVM Basis and Simulation for Real Subvolume

How to render and interpret the results?

We spent some time realizing better visualization and explanations of the obtained results. Based on what we learned in class and in related literatures, we further conduct multi-polynomial regression analyses combining cross validation to improve the structure-property linkage. These can be found in our posts right before the final presentation Two Point Statistics Computation and PCA Analysis and Structure-Property Linkages of Packed Soil Particles.

Visualization in the First Two Components

Multi-polynomial Regression Analysis

Project Contents

The workflow is shown as below.

  • 2-point spatial correlation is applied to quantitatively describe the soil structures.
  • PCA is conducted to obtain a reduced-order representation of structures.
  • Regression method combining leave-one-out cross validation is used to mine the desired structure-property correlation.

Method of Approach

Collaborations

Since we are in the same research group as well as work in the same office, we had a very good collaboration during the whole project process. Every Wednesday after the course, we had a meeting to discuss our project and figure out what we should do before the next Wednesday. Besides respective work assignment completed individually, every Friday we sat together to discuss and solve some issues.

To successfully reach the project goal, we need to express our thanks to Dr. Kalidindi, Dr. Tony Fast for great suggestions during the class presentations, and Ahmet Cecen and David Brough for joining our discussions and/or helping on some coding issues.

To be specific, the main responsibilities are distributed within the team as follows:

  • Mahdi: trial on LBM simulation and unit conversion, writing and debugging codes for GSP, PCA and regression analysis, preparing presentation
  • Jie: trial on LBM unit conversion and COMSOL simulation, running some parts of the codes for 2-point spatial statistics and PCA, updating posts and preparing presentation

Closure and Future Work

The most important conclusion that we achieved in this project was the main paramter affecting Hydraulic Conductivity. This parameter is pore structure. By comparing the spherical packed particles and real soil packed particles, we underestood a binary sized packed particles with the lowest value of porosity can capture a thorough information regarding Hydraulic Conductivity. Therefore, variation in hydraulic conductivity is directly related to how complex the pore structure is. As it was mentioned, this pore structure complexity can be highly captured by the binary packed particles which are spherical particles. Genreally, we could say hydraulic conductivity is independent of soil shape as long as we can produce high complex pore structures which can include many information of pore. In our case, binary packed particles could generate the comprehensive information about pore structure.

This course teaches us how to extract information from the raw datasets of structure from a data science perspective, building either a process-structure linkage or a structure-property linkage. This data-driven approach provides us a new and insightful idea to examine materials, like the soils that created by the nature.

This project is a very meaningful training session and renders us opportunity to practice the important theories we have learned from the course. We are thinking to write a research paper on the basis of our project.

References