Bachelor and Master Theses

Description

Understanding the complex interaction between a wheel loader’s bucket and a gravel pile during loading is critical for optimizing energy use, reducing component wear, and improving operational efficiency. Traditional empirical models often fall short in capturing the nonlinear and high-fidelity dynamics of soil-tool interaction, especially under varying conditions.

This thesis aims to combine physics-based simulation (AGX Dynamics) with machine learning techniques to develop a robust, hybrid model that accurately captures the interaction between the bucket and granular material during the loading (digging) process. The model will serve both predictive and control purposes, enabling more realistic digital twins and automated operation strategies within the Tested-SOS project.

Research Aim
The goal is to develop, train, and validate a high-fidelity model of bucket–gravel interaction using a hybrid approach that fuses physical simulation data from AGX Dynamics with machine learning-based approximation methods.

Research Questions (RQs)

• RQ1: How can AGX Dynamics be used to simulate realistic material interaction during the bucket loading process?
• RQ2: What machine learning models are suitable for approximating the complex contact forces, energy consumption, and material flow?
• RQ3: How can the hybrid model be validated and generalized across different pile geometries, gravel types, and digging strategies?

Expected Outcomes

• A validated AGX Dynamics setup for simulating bucket–gravel interaction with tunable parameters (material, geometry, etc.)
• A trained machine learning model (e.g., neural networks, GPR, or ensemble models) that approximates physical interaction patterns based on simulation and/or sensor data
• A hybrid model that can be integrated into digital twin environments for real-time or near-real-time analysis
• Insights into operational parameters that affect loading efficiency and material flow, with potential recommendations for automation strategies

Education in robotics, industrial systems, software engineering, applied AI or computer science, with knowledge of multibody dynamics, contact mechanics, or granular material modeling. Familiarity with physics-based simulation tools (e.g., AGX Dynamics, Simscape, or Modelica) and programming in Python or C++. Experience with machine learning methods (e.g., neural networks, regression, or GPR) for modeling physical systems, and interest in digital twins, off-road machinery, and hybrid physics–ML approaches.

The thesis will be carried out as part of the project Tested Site Optimization Solutions (TESTED-SOS[i]), a Vinnova-FFI project (2024-03678), which involves industrial collaboration with equipment manufacturers and site operators. The student will have access to AGX Dynamics simulation tools and existing datasets from quarry site operations.

Company Supervisors

Albin Nilsson (Volvo CE)

Abdulkarim Habbab (Volvo CE)