Aircraft Fault Prediction Classifier With Temporal Label Reassignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing aircraft fault prediction methods, particularly model-based techniques, are resource-intensive and require extensive knowledge of the expected operating state, making them time-consuming and inefficient, especially when the aircraft undergoes maintenance or configuration changes, leading to outdated models and increased false positive identifications of fault precursors.
Innovation Solution
A method that uses sequences of latent feature state values as feature vectors, labeled based on temporal proximity to fault occurrence, with probabilities determining correct labels and reassignment of labels to reduce false positives, enabling a data-driven approach for fault prediction without extensive human labeling and resource-intensive physics-based models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If model-based fault prediction methods are used, then fault prediction capability is improved, but system resource consumption and time requirements increase significantly
Solution Approach 1:
The patent replaces physics-based model-based fault prediction methods with a data-driven machine learning approach. Instead of using complex physics models that require significant computational resources and domain expertise, the system uses sensor data directly trained with neural networks or other ML algorithms, substituting the mechanical/mathematical modeling process with statistical learning from actual operational data.
Solution Approach 2:
The patent creates simplified representations of fault conditions by generating synthetic fault data that copies the structure and characteristics of actual fault scenarios. This allows the training dataset to include fault conditions without requiring physical access to or replication of actual fault states, reducing the need for resource-intensive model development while maintaining prediction capability.
2Reliability
If model-based fault prediction is implemented, then fault detection accuracy is improved, but the system becomes less adaptable when aircraft undergo maintenance or configuration changes
Solution Approach 1:
The patent implements a dynamic fault prediction system where the model automatically adapts to configuration changes through continuous learning from operational data. When aircraft undergo maintenance or configuration changes, the system collects new sensor data under the updated configuration and retrains the machine learning model, allowing it to dynamically adjust to new operating conditions without requiring manual model updates or expertise in the specific aircraft configuration.
Solution Approach 2:
The system performs self-updating by automatically collecting operational data, identifying configuration changes through data analysis, and retraining its own prediction models without external intervention. This self-service capability allows the fault prediction system to maintain high accuracy across different aircraft configurations and after maintenance events without requiring manual recalibration or expert involvement.
3Measurement precision
If extensive human labeling of fault data is performed, then training data quality is improved, but time consumption and human resource requirements increase
Solution Approach 1:
The patent generates synthetic fault data that copies the essential characteristics of actual fault conditions through simulation and data augmentation techniques. This synthetic data serves as high-quality training data without requiring manual labeling of actual fault events, which are rare and require extensive expert time to identify and annotate accurately.
Solution Approach 2:
The system performs automatic labeling of training data through unsupervised learning techniques and anomaly detection algorithms that identify fault conditions without human intervention. The machine learning models automatically learn to distinguish normal from faulty operations by analyzing patterns in the sensor data, eliminating the need for time-consuming manual labeling while maintaining high data quality.
Data Source
AI summary
A method includes receiving input data including a plurality of feature vectors and labeling each feature vector based on a temporal proximity of the feature vector to occurrence of a fault. Feature vectors that are within a threshold temporal proximity to the occurrence of the fault are labeled with a first label value and other feature vectors are labeled with a second label value. The method includes determining, for each feature vector of a subset, a probability that the label associated with the feature vector is correct. The subset includes feature vectors having labels that indicate the first label value. The method includes reassigning labels of one or more feature vectors of the subset having a probability that fails to satisfy a probability threshold and, after reassigning the labels, training an aircraft fault prediction classifier using supervised training data including the plurality of feature vectors and the labels.