Duy Le

Duy Le, PhD

Machine Learning & Computational Physics Researcher

Melbourne, Australia

Deep Learning Modelling & Simulation Discrete Element Method Computer Vision Mechatronics
Download CV

Who I Am

About Me

I am a research engineer with a PhD in computational physics and deep learning, working at the intersection of physics-based simulation, scientific computing, and neural network surrogate models.

My recent work focuses on developing deep-learning surrogates for granular flow simulations, replacing expensive Discrete Element Method (DEM) solvers with fast, data-driven models while preserving physical fidelity.

I am particularly interested in research and engineering roles involving ML for physical systems, AI for engineering applications, and scientific computing.

PhD — Computational Physics & Deep Learning

Federation University, Australia. Thesis: Accelerated Surrogate Modelling of Granular Materials using Artificial Neural Networks.

70–120× Simulation Speedup

Deep-learning surrogates achieving major speedups over high-fidelity DEM while maintaining physically meaningful behaviour.

Large-Scale ML Pipelines

Built end-to-end data pipelines and training frameworks for 3D particle simulations using Python, PyTorch, and Slurm HPC.

First-Author Publications

Peer-reviewed work spanning machine learning, computer vision, robotics, and granular materials.

What I Study

Research Interests

ML for Physical & Engineering Systems
Surrogate Modelling for Numerical Simulations
Discrete Element Method & Granular Materials
Scientific & High-Performance Computing
3D Geometry & Dynamics Systems
Physics-Informed & Data-Driven Modelling

What I've Built

Selected Projects

// 01

Neural Surrogate for 3D Granular Flow

Deep Learning Discrete Element Method Surrogate Modelling Scientific ML
  • Developed a deep neural surrogate to approximate 3D granular flow in industrial systems (hoppers, rotating drums, mixers).
  • Designed 3D continuous convolutional architectures to model particle–particle and particle–boundary interactions from spatial data, enabling accurate learning of complex contact dynamics.
  • Achieved 70–120× speedup over high-fidelity DEM, preserving mixing dynamics, energy evolution, and collision interactions.
  • Built an end-to-end ML pipeline: DEM data generation, preprocessing, distributed HPC training, and physics-based evaluation.
  • Validated the model against DEM baselines using physically meaningful metrics (e.g., kinetic energy, mixing entropy, collision statistics), demonstrating strong generalisation across operating conditions.

Publications

A Neural Network Surrogate for Modelling Granular Flow Dynamics in Industrial Applications with Dynamic Boundary Conditions — Powder Technology, 2026 Machine Learning Accelerated Prediction of 3D Granular Flows in Hoppers — ICANN, 2024
Pseudo-LiDAR

// 02

Low-Cost Pseudo-LiDAR for 3D Object Detection

Computer Vision 3D Perception Pseudo-LiDAR Autonomous Driving
  • Proposed a low-cost pseudo-LiDAR pipeline reconstructing 3D structure from monocular imagery using SLIC-based superpixel segmentation.
  • Designed a region-aware depth refinement strategy to enforce spatial consistency and reduce noise in reconstructed geometry.
  • Converted image-derived depth maps into LiDAR-like point clouds, enabling standard 3D detection pipelines without expensive sensors.
  • Demonstrated competitive performance and cost efficiency compared to conventional image-based detection methods.

Publication

Simple linear iterative clustering based low-cost pseudo-LiDAR for 3D object detection in autonomous driving — Multimedia Tools and Applications, 2023
Haptic Exoskeleton

// 03

Haptic Hand Exoskeleton for Virtual Reality

Robotics Haptics Mechatronics Human-Machine Interaction
  • Designed a force-controllable hand exoskeleton delivering direct fingertip force feedback while preserving natural finger motion.
  • Developed a bio-inspired linkage mechanism based on human finger kinematics for accurate force transmission.
  • Integrated a series elastic actuator (SEA) system for stable, controllable, and compliant force feedback.
  • Demonstrated realistic haptic sensations in VR environments with a lightweight and low-cost design.

Publications

An Efficient Force-Feedback Hand Exoskeleton for Haptic Applications — International Journal of Intelligent Robotics and Applications, 2021 A Design of Haptic Hand Exoskeleton for Virtual Reality Applications — ASYU 2021 (IEEE)
Archaeological Toolkit

// 04

Automating the Archaeological Toolkit

Systems Engineering Mechatronics Prototyping Real-World Problem Solving
  • Developed a complete engineering solution for mechatronic microdrill sampling of inclusions within pottery sherds.
  • Translated stakeholder needs into quantifiable engineering requirements including performance targets and evaluation metrics.
  • Performed concept generation and trade-off analysis across feasibility, cost, performance, and implementation complexity.
  • Conducted testing and validation against defined metrics and delivered full technical documentation and presentations.

Project Website

View Full Project Website

// 05

DEM Data Tools & Physical Metrics

Scientific Computing Data Analysis Physical Validation
  • Developed Python utilities to process physics-based simulation outputs into ML-ready datasets.
  • Implemented computation of physically relevant metrics: mixing entropy, kinetic energy (translational & rotational), angular velocity fields, and collision-based metrics.
  • Used these tools to benchmark neural network surrogates against high-fidelity DEM baselines across multiple operating conditions.

My Research Output

Selected Publications

2026

A Neural Network Surrogate for Modelling Granular Flow Dynamics in Industrial Applications with Dynamic Boundary Conditions

Duy Le, Gary W. Delaney, Linh Nguyen, Truong Phung, David Howard, Gayan Kahandawa, Manzur Murshed

Powder Technology

2024

Machine Learning Accelerated Prediction of 3D Granular Flows in Hoppers

Duy Le, Linh Nguyen, Truong Phung, David Howard, Gayan Kahandawa, Manzur Murshed, Gary W. Delaney

33rd International Conference on Artificial Neural Networks (ICANN)

2023

Simple Linear Iterative Clustering Based Low-Cost Pseudo-LiDAR for 3D Object Detection in Autonomous Driving

Duy Le and Linh Nguyen

Multimedia Tools and Applications

2021

An Efficient Force-Feedback Hand Exoskeleton for Haptic Applications

Duy Le and Linh Nguyen

International Journal of Intelligent Robotics and Applications

2021

An Efficient Multi-Vehicle Routing Strategy for Goods Delivery Services

Duy Le, Ying Men, Yunkang Luo, Yixuan Zhou, Linh Nguyen

IEEE International Conference on Advanced Robotics and its Social Impacts (ARSO)

My Toolbox

Skills Overview

Programming Languages

Python C MATLAB Java

ML / AI Frameworks

PyTorch TensorFlow Scikit-learn Open3D

Data & Scientific Computing

NumPy Pandas SciPy Data Pipelines

Simulation & Modelling

Discrete Element Method Modelling & Simulation Optimisation

Tools & Infrastructure

Git Slurm Linux HPC Kubernetes

Visualisation & Writing

Matplotlib Seaborn ParaView LaTeX