Aircraft Fault Prediction Training Sets Using Synthetic Flight Series
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Flight sensor data for aircraft components is extremely unbalanced, with a high prevalence of nominal flights and rare fault events, making it challenging for machine learning models to accurately predict component faults.
Innovation Solution
A method is employed to generate a balanced training dataset by correlating flight sensor data with component fault data, using a remaining useful life (RUL) function to assign labels, and generating synthesized flight series through copying and modifying original flight series with noise and dropout to balance the dataset.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If flight sensor data is used directly for training machine learning models, then the model training process is simple, but the dataset is extremely unbalanced with rare fault events, leading to poor prediction accuracy
Solution Approach 1:
The patent creates synthetic flight series by copying existing flight data and applying transformations (noise addition, dropout, scaling) to generate additional training samples. This copying approach balances the dataset by increasing the representation of underrepresented fault classes while maintaining the characteristics of original flight data, thereby improving fault prediction accuracy.
Solution Approach 2:
The patent applies parameter changes to flight data by adding noise with different variances, applying dropout rates, and using scaling factors to transform original flight series into synthetic variants. These parameter modifications create diverse training samples that balance the dataset and improve the model's ability to predict faults accurately.
2Quantity of substance
If synthetic flight series are generated by copying and modifying original flight series, then the dataset balance is improved, but the data processing complexity increases
Solution Approach 1:
The patent segments the data processing into distinct stages: selecting original flight series, copying them, adding noise, applying dropout, and scaling. This segmentation of the synthesis process makes the complex task manageable and allows for systematic control of each transformation step to achieve dataset balance.
Solution Approach 2:
The patent introduces intermediate processing steps (noise addition, dropout application, scaling) as mediators between the original flight data and the final synthetic dataset. These intermediary transformations gradually modify the data to achieve balance while maintaining traceability and control over the processing complexity.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
The present disclosure provides a method of generating a balanced training dataset for a machine learning model in one aspect, the method including: receiving flight sensor data corresponding to a plurality of flights, and applying one or more criteria to the flight sensor data to generate a training dataset including a plurality of first instances corresponding to flights of the plurality of flights. The method further includes assigning, using component fault data, respective labels to the plurality of first instances, and generating, for groups of one or more labels of the respective labels, a respective plurality of flight series. Each flight series includes a respective sequence of second instances that is based on some of the plurality of first instances, and that concludes with a second instance that is assigned a label included in the group.