Digital Twin of a Jet Engine Subsystem

Build a living simulation that mirrors real engine behavior

Advanced Digital Twins 8–12 weeks
Last reviewed: March 2026

Overview

Digital twins are virtual replicas of physical systems that update in real-time with sensor data. In aerospace, GE Aviation monitors 44,000+ jet engines with digital twins, predicting failures before they happen and optimizing performance across entire fleets.

In this project, you'll build a digital twin of a jet engine compressor stage. You'll create the physics model in MATLAB/Simulink (thermodynamic cycle, compressor map, performance equations), generate synthetic sensor data, and build a Python-based anomaly detection system that flags when the "real" engine deviates from expected behavior.

This is a semester-scale project that combines thermodynamics, control systems, signal processing, and machine learning — exactly the multidisciplinary skill set that propulsion companies are hiring for.

What You'll Learn

  • Model a jet engine compressor stage using thermodynamic cycle analysis
  • Build a Simulink model with compressor maps, inlet conditions, and performance outputs
  • Generate realistic synthetic sensor data with noise, drift, and fault signatures
  • Implement anomaly detection using statistical methods and ML (isolation forests, autoencoders)
  • Create a dashboard that compares model predictions with "measured" data in real-time
  • Understand the digital twin concept and its role in modern aerospace operations

Step-by-Step Guide

1

Model the Compressor Thermodynamics

Start with a single-stage axial compressor model. Implement the isentropic relations: total temperature ratio and total pressure ratio as functions of mass flow rate and shaft speed. Use a compressor performance map (available in open literature) to relate operating point to efficiency and pressure ratio.

Validate your model against published data: at design-point conditions, your predicted efficiency, pressure ratio, and temperature rise should match textbook values within 2–3%.

2

Build the Simulink Model

Create a Simulink block diagram with: inlet conditions (ambient temperature, pressure), compressor block (your thermodynamic model), shaft dynamics (spool speed response), and sensor outputs (temperature, pressure, shaft speed, vibration).

Add realistic elements: sensor noise (Gaussian), sample rate (10–100 Hz), and time delays. This makes your synthetic data look like real engine telemetry.

3

Simulate Normal Operations

Run the model through a realistic operating profile: ground idle → takeoff power → climb → cruise → descent → landing. Log all sensor outputs to create a "flight" of data. Run 50+ simulated flights with varying ambient conditions (hot day, cold day, high altitude) to build a baseline dataset.

4

Inject Fault Signatures

Add degradation modes to the model:

  • Compressor fouling: gradual decrease in efficiency (0.5% per 100 flights)
  • Blade erosion: reduction in pressure ratio with increased tip clearance
  • Bearing wear: increasing vibration amplitude at specific frequencies
  • Sensor drift: gradual offset in temperature sensor reading

Generate datasets with each fault mode at varying severity levels.

5

Build the Anomaly Detection System

In Python, implement a detection pipeline: feature extraction (calculate deltas between model prediction and "measured" data), baseline modeling (learn normal variation from healthy data), and anomaly scoring (flag deviations).

Try multiple approaches: statistical thresholds (simple but interpretable), isolation forests (good for high-dimensional data), and autoencoders (learn to reconstruct normal patterns, flag reconstruction errors).

6

Create a Monitoring Dashboard

Build a simple dashboard (using Plotly/Dash or Streamlit) that shows: current engine parameters, model predictions, residuals (difference between predicted and measured), and anomaly alerts. This is the operator-facing side of a digital twin system.

7

Evaluate Detection Performance

Test your anomaly detection on the faulted datasets. Calculate detection rate (fraction of faults caught), false alarm rate, and detection lead time (how early can you flag a developing fault). Create ROC curves comparing your different detection methods.

Document the results in a format suitable for a conference paper or thesis chapter.

Go Further

Advance your digital twin capabilities:

  • Full engine model — extend from a single compressor stage to a complete turbofan cycle (fan, compressor, combustor, turbine, nozzle)
  • Try Ansys Twin Builder — recreate your model using the industry-standard digital twin platform
  • Physics-informed detection — use PINNs to create a hybrid physics-ML anomaly detector
  • Fleet-level analysis — simulate 100 engines and detect fleet-wide trends vs. individual engine issues