Mechanical Intelligence Research Group · London South Bank University
Researching the intersection of machine learning, bio-inspired structural design and immersive AR/VR — building computational models and interactive tools for engineering and manufacturing.
Ashkan Aghamoali is an AI Researcher and XR Developer with the Mechanical Intelligence Research Group at London South Bank University (LSBU). His research combines machine learning with bio-inspired structural engineering.
He develops Python-based automation around CAE / Abaqus simulations, trains machine-learning models to predict structural behaviour from large simulation datasets, and applies optimization and sensitivity analysis to inform design decisions. In parallel, he builds Extended Reality (AR/VR) applications for visualising simulations, finite-element analysis and parametric models.
He holds an MSc in Digital Electronic Systems and a BSc in Electrical & Electronics Engineering from the University of Guilan, and is a gold-medal-winning member of the International Federation of Inventors' Associations (IFIA).
Peer-reviewed and in-progress research in bio-inspired design, sustainable manufacturing and computer vision.
AR/VR applications developed at LSBU for visualising machine-learning predictions, simulations and parametric models. Recordings auto-play below.
An immersive walkthrough presenting a bio-inspired structure and its simulation results, anchored in the user's environment.
An augmented-reality interface for collaborative-robot visualisation, linking digital simulation with physical automation.
Interactive visualisation of a biomimetic structure in augmented reality, with live controls to morph its geometry and explore design variations.
An AR panel that takes design parameters through an in-scene keypad and returns neural-network predictions of structural deformation, shown as live metrics and charts.
Real-time tuning of a lattice structure's parameters — density, radius, height, thickness — through an immersive AR control panel.
An AR scene visualising a robotic joint in real time alongside sensor/data overlays — bridging simulation data and physical motion.
Open-source work spanning robotics & machine learning, augmented reality, computer vision and web development — part of 85 public repositories.
An iOS-first platform that turns an iPhone into an industrial scanning tool. It captures room geometry live with ARKit + RoomPlan, detects and labels equipment on-device with Core ML / Vision, reads nameplates with OCR (fusing labels such as Pump P-101), and visualises scan coverage in real time. Scans export as JSON plus glTF/GLB models, sync to a Node.js + PostgreSQL backend, and are inspectable from a browser dashboard — while an edge gateway binds each scanned asset to real devices over MQTT and other industrial protocols with auditable, confirmation-gated control. Measured data is kept strictly separate from AI-inferred data throughout. Built milestone-by-milestone, each one independently runnable and verifiable.
An emergency-aware adaptive lighting system for a 6-lane smart motorway, built simulation-first across the full edge-to-cloud stack. A Python control backend holds the highway at a 30% eco glow and brightens only the luminaires ahead of each detected vehicle — escalating to a 100% safety corridor for ambulances, police and fire trucks — while modelling energy use, CO₂, luminaire health, remaining useful life and Re-X lifecycle decisions with digital product passports. Telemetry streams over MQTT to a Streamlit operations dashboard, a five-node STM32 edge-controller board network (RS485 / CAN) simulated in Wokwi, and a Unity 6 night-highway digital twin — cutting lighting energy by up to 70% versus an always-on baseline.
A simulation-based, hardware-ready digital twin for robotic recovery (Re-X) of end-of-life electric motors from Dubai's district-cooling infrastructure. A UR5e cobot with a Robotiq 2F-85 gripper sorts eleven separable motor components across reuse / repair / replace / recycle stations in Gazebo; a Python decision engine scores each part's health and risk from a Digital Product Passport and sensor stream, MoveIt 2 plans the 17-step pick-and-place workflow through ros2_control, and a Streamlit dashboard plus a Unity visual twin track every decision.
An operational digital twin for end-of-life lithium-ion battery recovery and robotic sorting. A Unity control twin streams each battery's image to a FastAPI backend that classifies the cell type, looks up NASA-derived state-of-health and scores risk; a decision engine routes every battery to reuse / remanufacture / recycle / quarantine under selectable policy modes, an ABB CRB-style cobot sorts it on the line, and a Streamlit dashboard tracks KPIs and digital product passports. Built on a real RecyBat24 image subset and NASA PCoE battery-aging data.
A digital twin of a desert solar farm that inspects every panel with computer vision, predicts energy loss with machine learning, and dispatches a cleaning robot to the highest-value panels under a realistic water budget. A soiling-regression CNN and a ResNet-18 fault classifier feed a gradient-boosted power-loss model; economic triage and 2-opt routing plan the cleaning run, exported to a ROS 2 Nav2 FollowWaypoints goal with a Gazebo physics twin — all served through a FastAPI inference service and a Streamlit dashboard.
A simulation-only inspection pipeline for circular manufacturing. A PyTorch model detects surface defects and localizes them with Grad-CAM, an ML grader sorts each part into reuse / repair / recycle with a confidence score, and a digital twin keeps a live record per part — all wired into a ROS 2 (Humble) node graph with a Gazebo inspection cell and RViz, plus a Streamlit dashboard.
A real-time, environment-aware AR experience using Niantic Lightship ARDK and Unity. Classifies real-world surfaces — sky, ground, trees, buildings — from the camera feed and drives AR behaviour with custom shaders.
A browser-based AR face filter built with Unity and the Needle Engine — real-time face detection and tracking with customizable effects, running cross-platform with no app installation.
A responsive React + TypeScript application for tracking cryptocurrency markets, demonstrating modern front-end development and live API integration.
A serverless Telegram bot on Cloudflare Workers that posts live foreign-exchange, gold and tether prices to a channel every 15 minutes with a clean Persian (RTL) layout. Each price is tracked against the 23:00 baseline held in Cloudflare KV, a cron trigger drives the schedule, and a nightly summary stores the next day's baseline — and if the data source fails it skips the cycle and alerts the admin rather than posting anything broken.
Develops bio-inspired structures within a multidisciplinary team; built a Python automation system integrated with CAE software for parametric modeling; analysed 1,000+ Abaqus simulation instances; implemented machine-learning models of structural behaviour and optimization for design tuning; and developed XR software (iOS / Android) presenting products, simulations and research.
Developed a high-performance object-detection system for camouflaged targets in video; a hybrid HSV + Local Binary Patterns approach improved accuracy by 40%; custom CUDA kernels achieved an 8× speedup enabling real-time 4K analysis at 30 fps; built an OpenGL visualisation module with dynamic thresholding.
Professional certificates and specializations in machine learning, deep learning, XR and digital manufacturing.
Unity badges link to verified Credly credentials. All course certificates link directly to verified credentials.
Open to research collaborations, XR projects, and roles bridging AI, bio-inspired design and immersive technology.