PROJECTS LOGBOOK

> Initializing Aerospace Research Data... [DONE]

UNDER REVIEW · CJA · CJOA-S-26-00957

AeroTwin: 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 in 7.9 minutes at 300× the speed of Simscape Multibody. Layer 2 uses a zero-leakage CNN-BiLSTM with Temporal Self-Attention achieving zero false negatives on catastrophic faults. Layer 3 integrates an offline Ollama 3.2 LLM translating neural outputs into ATA 32-compliant directives in under 1.5 seconds — fully air-gapped.

0%

Accuracy

±244h

RUL MAE

18k

ODE Samples

<1.5s

Edge Latency
ODE ENGINE CNN-BiLSTM OLLAMA 3.2 SHAP XAI PYTORCH FOCAL LOSS CS-25.473
AEROTWIN · 3-LAYER PIPELINE · B787-8 LIVE INFERENCE
[L1] ODE synthesising...
[L2] Class: Normal | Conf: 96.1%
[SH] SHAP r=0.781 p<0.0001
[L3] ATA 32 · No action required
PAPER SUBMITTED · IEEE TAES-2026-0772

PHI‑CHAIN: Blockchain +
Physics‑Informed Avionics Security

A multi‑protocol security framework integrating a permissioned blockchain (Hyperledger Fabric) with a four‑layer Physics‑Informed Digital Twin (PIDT) to detect both cyber‑attacks and physically impossible data injections. Every packet signed with Ed25519, Merkle‑batched, and anchored on-chain with a semantic integrity verdict.

  • Multi‑protocol coverage – ADS‑B, ADS‑C, CPDLC, ACARS, engine telemetry @10 Hz
  • Cryptographic layer – Ed25519 signing, Merkle batching, Hyperledger Fabric via WSL bridge
  • 100% attack detection – spoofed positions, impossible engine states, fuel fraud, ADS‑B injection

10 Hz

Data Rate

36k

Packets/hr

1.8 s

Batch Latency

100%

Attack Detect
HYPERLEDGER FABRIC ED25519 ADS‑B CPDLC FLIGHTGEAR WSL2
PHI‑CHAIN · LIVE BLOCKCHAIN STREAM · B787-8
[NET] ADS‑B Lat:9.072 Lon:7.491 Alt:FL350
[SIG] Ed25519 SIGNED · Batch #0041
[CHN] Block #1204 · root: a3f8…c91d
[DT] VERIFIED · No anomaly
PAPER SUBMITTED · RESS 2026

B787 Ignition Digital Twin
PHM & Deep Feature Spoofing

A closed‑loop PHM Digital Twin for the Boeing 787‑8 ignition system (ATA 74). Overcomes run‑to‑failure data scarcity with a physics‑informed MATLAB ODE solver generating 27,000 samples across 11 fault modes. A LightGBM regressor (R² 0.87, MAE 39 cycles) maps micro‑electrical degradation to RUL, deployed via a decoupled FastAPI microservice with "Deep Feature Spoofing" God-Mode GUI.

0.87

R² Score

39

MAE (cycles)

27k

Samples

11

Fault Modes
LIGHTGBMFASTAPI MATLAB ODEFLIGHTGEAR PYTHONUDP
B787 ATA-74 · IGNITION PHM · DEEP FEATURE SPOOF
[ENG] Spark: 3.85J | EGT: 682°F
[RUL] 421 cycles · HEALTHY
[FLT] Injecting: None
[AMM] ATA 74-00-00 · Monitor
PAPER SUBMITTED · AIAA JAIS 2026

AeroTwin: Physics-AI
Dornier 228 Landing Gear DT

A certification-compliant Cyber-Physical System fusing MATLAB Simscape Multibody physics with a dual-input LSTM deep learning network via PostgreSQL. The confidence-weighted fusion architecture explicitly satisfies CS-23.473 hard-landing regulations — triggering mandatory ATA 32-00 inspections at sink rate ≥ 3.05 m/s or acceleration ≥ 1.8g. Trained on 9,500 Dornier 228 landing cycles across 11 ATA-32 fault classes.

0%

AI Accuracy

0×

Cost Reduction

47ms

Latency

9,500

Cycles
SIMSCAPEDUAL-INPUT LSTM PYTHON/UDPPOSTGRESQL PYVISTA 3DCS-23.473
AEROTWIN V16 · DO228 LIVE SIM · ABUJA→LAGOS
[SYS] AeroTwin V16 · Run 001
[DAT] Sink: 2.81 m/s | G: 1.22g
[AI] Class 00: Normal | Conf: 96.2%
[ATA] Source: AI · Ref: N/A

Active & Future Initiatives

LIVE PROOF OF CONCEPT

AI for Fluids (Neural CFD)

Revolutionizing computational fluid dynamics by replacing traditional Navier-Stokes solvers with Physics-Informed Neural Operators (PINNs). The interactive engine below uses Aerosandbox/XFoil for real-time demonstration; final deployment will transition to ANSYS Fluent for industrial-grade predictions.

ACTIVE DEVELOPMENT · ENERGY SECTOR

NNPC Pipeline
Health Monitoring System

A real-time, IoT-driven pipeline integrity monitoring platform designed for the Nigerian National Petroleum Corporation (NNPC) crude oil network. The system fuses distributed fibre-optic and MEMS pressure sensor arrays with an LSTM-Autoencoder anomaly engine to detect leaks, corrosion hotspots, third-party interference, and wax deposition — hours before catastrophic failure. A GIS-integrated dashboard maps every fault event to its GPS coordinate on the Trans-Niger Pipeline corridor in real time.

  • Leak detection — sub-1 bar pressure anomaly detection along 614 km corridor with GPS fault localisation <50 m accuracy
  • Corrosion RUL prediction — physics-informed electrochemical ODE model feeding BiLSTM wall-thickness degradation estimator
  • Bunkering/tampering alerts — acoustic emission signature classifier differentiates equipment noise from unauthorised tapping

614km

Pipeline

<50m

Fault GPS

IoT

Edge Nodes

LSTM

Anomaly AI
LSTM-AUTOENCODER NODE-RED IOT INFLUXDB GRAFANA GIS MQTT FASTAPI PYTORCH
NNPC TRANS-NIGER PIPELINE · LIVE SENSOR STREAM
[SEN] Node-14 · Pressure: 42.1 bar
[AI] Anomaly: NONE · Corrosion RUL: 8.2yr
[GPS] 5.412°N, 6.438°E · Segment OK
[SYS] All 22 nodes nominal
IN DEVELOPMENT · CRANFIELD COLLABORATION

AI-Powered MRO
Inventory Forecasting Platform

A joint initiative with Muhammed Fikri (PhD Researcher, Cranfield University) — building an end-to-end AI platform that transforms MRO parts management from reactive stockpiling to predictive, zero-AOG inventory control. The system ingests flight cycles, component MTBF curves, supplier lead times, and live maintenance records to produce a time-series demand forecast for every line-replaceable unit, triggering purchase orders autonomously before a stock-out can ground an aircraft.

  • Demand forecasting — Transformer-LSTM hybrid trained on 5-year consumption records; 14-day rolling horizon with 91% MAPE accuracy
  • Supplier chain optimisation — multi-echelon inventory model balancing safety stock, EOQ, and AOG risk scores per part number
  • CMMS integration — REST API bridge to AMOS / SAP PM for real-time part consumption feedback and auto-replenishment orders
  • Explainability dashboard — SHAP-driven "why this order?" interface giving engineers full visibility into every AI recommendation

91%

MAPE Acc.

14-day

Forecast

LRU

Part Tracking

0-AOG

Target
TRANSFORMER-LSTM PYTORCH PROPHET FASTAPI POSTGRESQL SHAP XAI REACT DASHBOARD SAP-PM API

🎓 Joint research with Muhammed Fikri, PhD Researcher · Cranfield University, UK

MRO AI · INVENTORY FORECAST · LIVE SIMULATION
[PRD] Part: B787-LG-SEAL-032 | Stock: 14
[AI] Forecast: 22 units / 14 days
[ROP] Reorder Point hit in 6 days
[PO] Auto-PO: 40 units · Lead: 8d
IN DEVELOPMENT

Rotating Machinery & Bearing Digital Twin

A vibration-signature-driven Digital Twin architecture for rotating equipment — aircraft engines, APU shafts, and gearboxes. Combines FFT/envelope analysis with a CNN-LSTM fault classifier trained on CWRU bearing datasets to predict spall, pitting, and cage fracture up to 50 duty cycles before detectable wear. Integrates MATLAB Simulink multi-body dynamics with a real-time Python inference layer for continuous RUL tracking under variable load and speed conditions.

CWRU DATASETFFT/ENVELOPE CNN-LSTMSIMULINK SCIPY SIGNALPYTORCHINFLUXDB
CONCEPT PHASE

Airframe Structural Digital Twin

A full-aircraft structural health monitoring Digital Twin combining Finite Element Analysis (ANSYS Mechanical) with Physics-Informed Neural Networks to track fatigue crack propagation, composite delamination, and buckling risk across the primary load-bearing structure. The FEM mesh is updated continuously from a distributed Structural Health Monitoring (SHM) network of piezoelectric Lamb-wave sensors and strain gauges, enabling real-time stress mapping and remaining fatigue life prediction compliant with EASA AMC 25.571.

ANSYS MECHANICALPINN SHM SENSORSLAMB WAVES FEM MESHPYTORCH EASA AMC 25.571PYTHON FEniCS
IN DEVELOPMENT

Ontology-Driven Computerized Aircraft Maintenance Program

A semantic "reasoning brain" for aircraft maintenance, encoding the full ATA iSpec 2200 chapter structure as an OWL/RDF knowledge graph. The system uses SPARQL inference to autonomously cross-reference airworthiness directives, task card dependencies, and MEL items — eliminating manual cross-referencing errors. A natural language interface (LLM + SPARQL backend) lets engineers query the maintenance program in plain English: "What deferred defects affect my dispatch today?"

OWL/RDFSPARQL PROTÉGÉNEO4J LLM INTERFACEATA iSpec 2200 FASTAPIREACT
OPERATIONAL

Real-Time Telemetry & DevOps Stack

The backbone powering every project in this logbook — a production-grade data engineering pipeline from simulation to insight. FlightGear telemetry arrives via UDP, is parsed in Python, streamed through Telegraf to InfluxDB 3 Core (time-series) and PostgreSQL (warehouse), visualised live in Grafana, and served via FastAPI microservices. MATLAB ODE engines feed PyTorch training pipelines; models are versioned in MLflow and deployed as REST endpoints. The full stack runs on a single Intel i7-12700H workstation — proving that enterprise-grade aerospace AI needs no cloud.

FLIGHTGEAR UDPPYTHON TELEGRAFINFLUXDB 3 POSTGRESQLGRAFANA FASTAPIPYTORCH MATLAB ODEMLFLOW HYPERLEDGEROLLAMA

Interactive Tools & Concepts

Digital Twin Monitor

Real-time structural health monitoring dashboard with live telemetry visualization and predictive analytics.

MRO Interval Calculator

Predictive maintenance planning tool for calculating optimized aircraft inspection intervals based on usage data.

Safety Analytics

Aviation safety monitoring interface for tracking incident trends and airworthiness compliance metrics.

Technical Archives

UDP + FlightGear Integration

Real-time telemetry bridge connecting FlightGear to cloud databases — UDP parsing, data validation, streaming to InfluxDB.

View Repository

Telemetry Data Pipeline

Complete ETL pipeline from simulation to visualization. Python UDP listeners, Telegraf, PostgreSQL/InfluxDB dual-database architecture.

Explore Code

Structural Beam Analysis

FEA of aircraft wing spars using Ansys Workbench. Optimized material thickness for weight reduction while maintaining FAA safety factors.

Project Completed

Grafana Monitoring Dashboards

Custom aviation dashboards — real-time flight parameters, system health indicators, predictive maintenance alerts.

View Screenshots

Airworthiness Documentation

Digital compilation of NCAA regulatory procedures — Technical Logs, CoA renewals, MEL protocols.

Read Experience

LSTM Fault Detection Model

Deep learning model for predicting landing gear failures from time-series sensor data. 94.3% accuracy with 50-cycle early warning.

View Model