Sequential Pattern Mining for Aeronautical Fault Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex aeronautical systems generate numerous fault messages, leading to false alarms and unnecessary maintenance actions, which result in costly downtimes and reduced flight efficiency due to incomplete knowledge of system models and dependencies.
Innovation Solution
A method involving on-board and ground processes that filter fault messages using data mining techniques, specifically sequential pattern analysis, to build models that distinguish real from parasitic failures, utilizing a knowledge model that updates and refines patterns based on expert feedback and system dependencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If model-based diagnostic methods are used to identify fault sources, then diagnostic precision is improved, but the method becomes inapplicable when system models are incomplete or partially known
Solution Approach 1:
The patent introduces sequential pattern mining as an intermediary method between incomplete system models and fault diagnosis. Instead of directly applying model-based reasoning to systems with unknown behaviors, the system mines sequential patterns from historical alarm data to infer causal relationships. This intermediary approach allows diagnostic capabilities to function effectively even when complete system models are unavailable, bridging the gap between limited knowledge and accurate fault identification
Solution Approach 2:
The system performs self-service by automatically learning system behavior patterns from historical data without requiring complete a priori knowledge. Through sequential pattern mining, the system autonomously discovers causal relationships and alarm sequences that occur in practice, enabling it to diagnose faults in systems whose full operational logic is not formally specified or documented
2Measurement precision
If all fault messages are processed and analyzed, then diagnostic completeness is improved, but unnecessary maintenance actions increase due to false alarms
Solution Approach 1:
The patent extracts and separates true faults from false alarms by analyzing sequential patterns in alarm messages. Instead of treating all alarm messages equally, the system identifies patterns characteristic of genuine failures versus those caused by upstream failures or system dependencies. This extraction process isolates the subset of alarms requiring actual maintenance attention, eliminating unnecessary interventions while maintaining complete fault detection
Solution Approach 2:
The system uses feedback from historical alarm data and maintenance outcomes to refine its understanding of true versus false alarms. By continuously learning from past incidents and their resolutions, the system improves its ability to distinguish genuine faults from parasitic messages, reducing false maintenance actions over time while maintaining comprehensive fault detection capabilities
3Reliability
If comprehensive system models with complete dependency knowledge are created, then false alarm reduction is improved, but system complexity and model construction difficulty increase
Solution Approach 1:
The patent performs preliminary action by automatically mining sequential patterns from historical data before diagnostic operations begin. Instead of requiring manual construction of complex dependency models, the system pre-processes historical alarm sequences to identify causal patterns and relationships. This preliminary pattern extraction creates a usable diagnostic framework without requiring complete a priori knowledge of system dependencies, reducing both model construction complexity and false alarms
Solution Approach 2:
The patent replaces the mechanical process of manual model construction with an automated data mining approach. Instead of engineers manually mapping system dependencies and creating comprehensive models, the system uses sequential pattern mining algorithms to automatically discover relationships from operational data. This substitution eliminates the complexity of manual model building while achieving comparable or superior false alarm reduction through empirically derived patterns
Data Source
Figure 1
Figure 2~3
AI summary
The invention relates to a method for processing a sequence of fault messages occurring in an apparatus including numerous systems, that makes it possible to discriminate fault messages most likely originating from a real system fault and fault messages without any real fault of the associated system (no fault found), wherein said method combines: an "air" method (100) executed in real time for acquiring fault message data defining sequences stored in a fault database and for immediate diagnosis assistance using a diagnosis tool; and a "ground" method (200) for the subsequent analysis of fault sequence data and for improving the diagnosis tool.