Solutions Video Demos
× Expanded View

PROJECTS LOGBOOK

> Initializing Aerospace Research Data... [DONE]

UNDER REVIEW · CJOA-S-26-00957

PHI-Twin: Physics-AI
B787 Landing Gear Prognostics

A three-layer, edge-deployable predictive maintenance framework for the Boeing 787-8 main landing gear. Layer 1 deploys an ODE synthesis engine generating 18,000 physics-valid fault profiles at 300× the speed of Simscape. Layer 2 uses a zero-leakage CNN-BiLSTM with Temporal Self-Attention. Layer 3 integrates an offline Ollama 3.2 LLM translating neural outputs into ATA 32-compliant directives.

80.6%

Accuracy

±244h

RUL MAE

18k

ODE Samples

<1.5s

Edge Latency
ODE ENGINECNN-BiLSTM OLLAMA 3.2SHAP XAI PYTORCHCS-25.473
PAPER SUBMITTED · IEEE TAES-2026-0772

PHI‑CHAIN: Blockchain +
Physics‑Informed Security

A multi‑protocol security framework integrating a permissioned blockchain (Hyperledger Fabric) with a four‑layer Physics‑Informed Digital Twin (PIDT) to detect cyber‑attacks and physically impossible data injections. Covers ADS‑B, CPDLC, and ACARS telemetry at 10Hz. Every packet is signed with Ed25519, Merkle‑batched, and anchored on-chain.

10 Hz

Data Rate

36k

Packets/hr

1.8 s

Latency

100%

Detect Rate
HYPERLEDGERED25519 ADS‑B / CPDLCFLIGHTGEAR WSL2
PAPER SUBMITTED · RESS 2026

B787 Ignition Digital Twin
Deep Feature Spoofing

A closed‑loop PHM Digital Twin for the Boeing 787‑8 ignition system. Overcomes run‑to‑failure data scarcity with a physics‑informed MATLAB ODE solver generating 27,000 samples across 11 fault modes. A LightGBM regressor maps micro‑electrical degradation to RUL, deployed via a decoupled FastAPI microservice with a God-Mode GUI.

0.87

R² Score

39

MAE (Cyc)

27k

Samples

11

Fault Modes
LIGHTGBMFASTAPI MATLAB ODEFLIGHTGEAR PYTHON
PAPER SUBMITTED · AIAA JAIS 2026

AeroTwin: Physics-AI
Dornier 228 Landing Gear

A certification-compliant Cyber-Physical System fusing MATLAB Simscape Multibody physics with a dual-input LSTM network via PostgreSQL. The confidence-weighted fusion explicitly satisfies CS-23.473 hard-landing regulations, triggering mandatory ATA 32-00 inspections at high sink rates. Trained on 9,500 Dornier 228 landing cycles.

94.2%

Accuracy

350×

Cost Drop

47ms

Latency

9,500

Cycles
SIMSCAPEDUAL-INPUT LSTM PYTHON/UDPPOSTGRESQL PYVISTA 3DCS-23.473
DESIGN REGISTRY · DR. G. E. ABBE

STRATOS-14E
Ultra-Long-Range VIP Transport

A first-principles 14-seat VIP transport integrating aerodynamics, structures, propulsion, and systems architecture. Embeds a native PHI-SENSE HUMS grouping 1,027 concurrent structural strain sensors. A CNN-BiLSTM-Attention deep classifier delivers 47ms state inference at 98.5% accuracy. A permissioned PHI-CHAIN ledger secures tamper-proof lifecycle log tracking.

12.5k

km Range

M0.85

Cruise

36.2t

MTOW

FL510

Ceiling
PHI-SENSE HUMSCNN-BiLSTM PHI-CHAIN1,027 SENSORS DOC $4,870/hr
AFIT AEROSPACE · CO-DESIGNED WITH I. T. USMAN

PHI-DRONE
Digital Twin Reconnaissance Hub

A zero-cost open-source UAV simulation pipeline integrating ArduPilot SITL, FlightGear, and YOLOv8 for real-time object detection and telemetry fusion. Engineered a cross-platform (Windows/WSL2) architecture with screen-capture calibration and thread-safe MAVLink GPS logging. Validates the companion-computer pattern entirely in software.

15fps

Inference

Zero

HW Cost

WSL2

Bridge

Q1

Journal
PYTHONYOLOv8 OPENCVARDUPILOT SITL FLIGHTGEARMAVLINK
AIAA SCI TECH 2027 · STUDENT COMPETITION

CFD Spatiotemporal CVAE
Rotary Detonation Augmentation

A 4.3M-parameter generative Conditional Variational Autoencoder that augments scarce Rotary Detonation Engine CFD data. Built from 12 OpenFOAM v2406 reactive-flow simulations, the CVAE generates 72 physically consistent spatiotemporal fields. Documents a critical finding: standard literature velocity correlations overestimate true Chapman-Jouguet bounds by up to 4× at off-stoichiometric limits.

4.3M

Params

72

Fields

1.2M

Surrogate

CJ Error
OPENFOAM v2406CVAE CNN-BiLSTMPYTORCH REACTINGFOAMSURROGATE
ACTIVE RESEARCH · PROPULSION PHM

Gas Turbine Blade PHM
AGTF30 Predictive Maintenance

A laboratory-scale predictive maintenance system for High-Pressure and Low-Pressure turbine blade degradation. Uses the AGTF30 advanced gas turbine engine model in MATLAB with a 13-sensor continuous telemetry bus. A deep CNN-BiLSTM-Attention architecture classifies micro-fractures and erosion, connecting to a FlightGear UDP bridge for hardware-in-the-loop validation.

13

Sensors

99%

Classification

HP/LP

Stages

HIL

Validation
AGTF30 ENGINECNN-BiLSTM ATTENTIONMATLAB SIMSCAPE FLIGHTGEAR UDP

Autonomous Software & AI Infrastructures

While my core research resides in aerospace structures and digital twins, I actively deploy sovereign software architectures, real-time telemetry pipelines, and robust offline AI matrices.

Lifestone Engine v2.0

Offline-first speech-to-text processing node via native whisper-rs. Links an Aho-Corasick parser matching 311k text elements with downstream ONNX classifiers.

COLIG Full-Stack Node

Web-Scale Distributed Platform. Integrates high-throughput PocketBase transactions, chronological data models, and LiveKit Cloud infrastructure trees.

Penelope AI Agent v1.2

Sovereign Offline Execution Matrix. Drives multi-provider runtime fallback trees via deterministic ReAct execution loops with physical signature tracking.

Active & Future Initiatives

LIVE AI INFERENCE — NEURAL NETWORK SURROGATE

AI for Fluids — Airfoil Digital Twin

A neural network surrogate — trained on airfoil sweep data and served live from this page — predicts lift and drag coefficients in real time from geometry, angle of attack, and Reynolds number. Adjust the controls inside, then export the exact sweep you generate as a CSV.

Watch Demo
LIVE DIGITAL TWIN — PHI-ARC ENGINE PHM

Interactive Gas Turbine Digital Twin

Direct interface to the TPE331-25 lab-scale digital twin. Run live telemetry inputs through the CNN-BiLSTM-Attention architecture, visualize multi-stage degradation heatmaps, and generate compliant maintenance reports in real-time right here from the browser.

Watch Demo
IN DEVELOPMENT

Bearing Digital Twin

Vibration-signature-driven DT for rotating equipment. Combines FFT/envelope analysis with CNN-LSTM to predict spalling 50 cycles early.

CWRUFFTCNN-LSTM
CONCEPT PHASE

Structural Digital Twin

Full-aircraft SHM fusing FEA with Physics-Informed Neural Networks to track crack propagation and buckling risk compliant with EASA AMC 25.571.

ANSYSPINNLAMB WAVES
IN DEVELOPMENT

Ontology-Driven CAMP

Semantic reasoning brain encoding ATA iSpec 2200 as an OWL/RDF graph. Uses SPARQL to autonomously cross-reference ADs and MEL items.

OWL/RDFSPARQLLLM
OPERATIONAL

DevOps & Telemetry Stack

The backbone pipeline. FlightGear UDP streams through Telegraf to InfluxDB 3 and Postgres, visualized in Grafana, served via FastAPI.

INFLUXDB 3TELEGRAFGRAFANA

Technical Archives & Profiles

REP · 01
DATASETS

Kaggle Profile

Aerospace Datasets & Physics-AI

Repository of synthetic landing gear datasets, digital twin degradation profiles, and machine learning notebooks used in published research.

DatasetsNotebooksPhysics-AI
REP · 02
SPACES & MODELS

Hugging Face

Live Inference & Surrogates

Deployed generative models, CVAE spatiotemporal fields, and live interactive spaces for aerospace fluid and structural AI architectures.

SpacesModelsGradio
NET · 03
PROFESSIONAL LOG

LinkedIn

Career & Publications

Professional network, architectural design publications, aerospace portfolio updates, and progress on peer-reviewed journal submissions.

NetworkArticlesUpdates
REP · 04
OPEN SOURCE

GitHub Matrix

Source Code & Repositories

Complete codebase repository of aerospace research, digital twin simulation pipelines, and autonomous agent architectures.

Source CodeRepositoriesArchitecture