Aircraft State Prediction Using Learned Flight Data Models

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for predicting the future state of an aircraft are limited by their reliance on generic, static models that fail to account for individual aircraft variability and behavioral dynamics, leading to suboptimal performance optimizations, particularly in converging towards local minima and being non-robust to real-world behavioral variations.

Innovation Solution

The implementation of machine learning techniques, including unsupervised and supervised learning methods, to determine the future state of an aircraft by applying models learned from flight data, using parameters such as SEP, FF, and N1, allowing for end-to-end performance calculation and trajectory prediction, independent of manufacturer-provided models, and enabling continuous learning from flight data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If model-based approaches incorporating aircraft physics equations are used, then the prediction is based on theoretical models, but the method is not robust to behavior variability of individual aircraft compared to average modeled aircraft

Engineering Contradiction:
Improverobustness to behavior variabilityVSAvoidindividual aircraft characteristics
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system continuously collects flight data from multiple aircraft and uses machine learning models to learn actual behavior patterns. This feedback loop allows the system to adapt to individual aircraft characteristics while maintaining robustness through aggregation of data from the fleet, resolving the contradiction between theoretical model reliability and individual variability adaptation.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The machine learning models automatically learn and adapt to each aircraft's specific behavior patterns without requiring manual modeling or adjustment. The system self-adjusts to individual aircraft characteristics by processing actual flight data, eliminating the need for aircraft-specific theoretical model calibration while maintaining robustness through data-driven learning.

Inventive Principle:
Principle #25Self-service

2Productivity

If generic static models are used for predicting aircraft future state, then the implementation is simple, but the optimization performance is suboptimal and converges to local minima

Engineering Contradiction:
Improveoptimization performanceVSAvoidmodel complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system transitions from static generic models to dynamic machine learning models that continuously adapt to changing flight conditions and individual aircraft characteristics. This dynamic approach enables the system to escape local minima by learning from actual flight data and adjusting predictions in real-time, improving optimization performance while managing complexity through efficient model architectures.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The machine learning models learn optimal parameters from actual flight data rather than relying on fixed theoretical parameters. By continuously adjusting model parameters based on observed aircraft behavior, the system achieves superior optimization performance while the modular architecture manages computational complexity through parameter efficiency.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If end-to-end machine learning is applied for performance calculation and trajectory prediction, then accurate real-time prediction is achieved, but computational requirements increase

Engineering Contradiction:
Improveprediction accuracyVSAvoidcomputational power
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The system performs preliminary training of machine learning models offline using historical flight data, creating pre-trained models that can be deployed for real-time predictions. This preliminary action separates the computationally intensive training phase from the operational phase, achieving high prediction accuracy in real-time with reduced computational power requirements during actual flight operations.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3671392B1Automatic learning in avionics
Publication Date: 2024.10.30 THALES SA
  • EP3671392B1 patent drawingFigure 1~2
  • EP3671392B1 patent drawingFigure 3
  • EP3671392B1 patent drawingFigure 4

AI summary

The document concerns systems and methods for managing the flight of an aircraft, including the steps of receiving data (200) from aircraft flight recordings; said data including data from sensors and/or data from the onboard avionics; determining the aircraft state at point N (220) from the received data (200); determining the aircraft state at point N+1 (240) from the aircraft state at point N (220), by applying a model learned by machine learning (292). Developments describe the use of flight parameters SEP, FF, and NI; unsupervised, offline, and/or online machine learning, according to a variety of algorithms and neural networks. Software aspects are described.