Aircraft Sensor Subset Selection via Genetic Programming
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning methods struggle to efficiently select relevant sensor subsets for predicting dependent variables, particularly in aircraft systems, due to the impracticality of using all sensors, redundancy, and the inefficiencies of manual feature extraction and multivariate approaches like random forests.
Innovation Solution
Implement genetic programming to iteratively analyze multivariate time series data, using multiple runs to automatically generate and rank sensor combinations based on fitness criteria, thereby selecting a subset of sensors that best predict the dependent variable.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all sensors are used in the machine learning model, then the prediction accuracy may be improved, but the time and computing resource requirements become impractical
Solution Approach 1:
The patent extracts and removes redundant and irrelevant sensors from the complete sensor set through automated feature generation. The system identifies and eliminates sensors that do not contribute meaningfully to prediction accuracy, retaining only the essential subset of sensors needed for effective machine learning model performance.
Solution Approach 2:
The patent segments the complete sensor set into distinct categories: relevant sensors, redundant sensors, and irrelevant sensors. This segmentation is achieved through automated analysis that evaluates sensor contributions to prediction accuracy, allowing the system to process only the relevant segment while discarding the rest.
2Loss of time
If manual feature extraction is performed to reduce sensor data, then the time and computing resources are reduced, but important temporal and interrelated behaviors in the raw data are hidden
Solution Approach 1:
The patent implements self-service through automated feature generation that performs sensor selection without requiring manual intervention. The system automatically identifies relevant sensors, evaluates their contributions to prediction accuracy, and generates the optimized sensor subset autonomously, eliminating the need for manual feature extraction while preserving important data behaviors.
3Productivity
If simple feature extraction such as straight statistics is applied, then the processing is simplified and fast, but important temporal and interrelated behaviors present in the raw data are hidden
Solution Approach 1:
The patent replaces simple statistical feature extraction with an automated machine learning-based sensor selection system. Instead of using basic statistics that lose temporal and interrelated information, the system employs automated feature generation that leverages machine learning algorithms to identify relevant sensors while preserving the full temporal and interrelated behaviors present in the raw sensor data.
4Measurement precision
If a multivariate approach with random forest is applied to use feature importance capability, then the sensor selection may be improved, but training on all raw multivariate sensor data is infeasible due to computational constraints
Solution Approach 1:
The patent performs preliminary action by automatically generating and selecting the relevant sensor subset before training the machine learning model. This preliminary sensor selection step reduces the dimensionality of the input data, making subsequent model training computationally feasible while maintaining the ability to accurately identify important sensors through the automated feature generation process.
Data Source
AI summary
A method is provided that includes accessing a multivariate time series of flight data for an aircraft, and iteratively performing runs of genetic programming on groups of the sensors. A population of computer programs is randomly generated from a selected group of the plurality of sensors, and primitive functions selected from a library of primitive functions. The population is iteratively transformed into new generations of the population, and includes sub-rankings of the group of sensors based on a quantitative fitness determined according to selected fitness criterion. A ranking of the group of sensors from the sub-rankings of the group of sensors is produced. An aggregate ranking of the plurality of sensors is produced from the ranking of the group of sensors over a plurality of iterations. And the subset of sensors is selected from the aggregate ranking of the plurality of sensors, and according to selected optimization criterion.