Multi-Level Fault Isolation for Aircraft Engine Diagnostics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current fault isolation systems in aircraft engines face challenges in distinguishing between similar fault conditions due to limited sensor data and noise, with existing classification algorithms like physics-based models, empirical neural networks, and knowledge-based systems having limitations such as non-linear measurement issues, overtraining, and lack of adaptability to new systems.
Innovation Solution
A tailored fault isolation system that selects and combines different classification methods based on performance analysis of sensor data to differentiate between potential faults, using multiple isolators at each node to narrow down fault conditions, and storing the system for focused maintenance efforts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If physics-based models are used for fault classification, then the system can handle linear relationships between measurements, but it cannot account for non-linear measurements such as vibration or oil related sensors
Solution Approach 1:
The fault classification system is divided into multiple specialized classification schemes, each optimized for specific types of measurements. Physics-based models handle linear relationships, while neural networks handle non-linear relationships. This segmentation allows each scheme to excel at its designated task, improving overall adaptability without sacrificing reliability.
Solution Approach 2:
The system implements a universal fault classification framework that can accommodate multiple types of classification schemes (physics-based models, neural networks, knowledge-based systems) within a single integrated system. This multi-functionality enables the system to handle both linear and non-linear measurements effectively.
2Adaptability or versatility
If empirical neural networks are used for fault classification, then the system can handle non-linear measurements, but it is prone to overtraining and requires large amounts of data
Solution Approach 1:
The system segments the classification task by using neural networks only for specific non-linear measurement types where they provide superior performance, rather than applying them universally. This reduces the overall data requirements by limiting neural network usage to cases where their non-linear handling capability is essential.
Solution Approach 2:
The system dynamically selects and adjusts classification parameters based on the specific fault condition and available data. When data is limited, the system may switch from complex neural networks to simpler physics-based models or knowledge-based systems, effectively changing the classification parameters to match data availability.
3Reliability
If knowledge-based systems are used for fault classification, then the system can utilize experience-based knowledge, but it is not useful for newer systems without historical data
Solution Approach 1:
The system performs preliminary classification using physics-based models that do not require historical data, and only invokes knowledge-based systems when sufficient historical data is available. This preliminary action ensures that new systems can be classified reliably from the start using physical principles, while gradually building up knowledge-based capabilities as data accumulates.
Solution Approach 2:
The system dynamically adapts its classification approach based on the maturity and data availability of the specific system being analyzed. For newer systems with limited historical data, it relies more on physics-based models and engineering knowledge, while for mature systems with extensive data, it increasingly utilizes knowledge-based systems trained on accumulated operational experience.
4Device complexity
If a single classification algorithm is used for all fault conditions, then the system is simpler to implement, but it cannot effectively differentiate between similar fault conditions with limited sensor data
Solution Approach 1:
The system segments the classification process into multiple specialized algorithms, each optimized for specific fault types or measurement characteristics. This segmentation improves measurement precision by matching the right algorithm to the right fault condition, while the modular structure keeps implementation complexity manageable through clear division of responsibilities.
Solution Approach 2:
Different classification algorithms are applied to different fault conditions based on their specific characteristics. Each algorithm has local quality optimized for its designated fault type, allowing high precision differentiation of similar fault conditions while maintaining overall system manageability through specialized, focused implementations.
Data Source
AI summary
A fault isolation method and system includes tailored fault isolators for each grouping of similar data to differentiate between one or several potential faults within any group.