Failure Diagnosis Using Pre/Post-Maintenance Feature Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive abnormality diagnosis systems face challenges in effectively utilizing measurement data before and after maintenance to diagnose equipment failures, particularly in identifying optimal feature detection algorithms and diagnosis processing, and are often limited to specific devices, making them expensive and not suitable for mass production.
Innovation Solution
A failure diagnosis system that includes a sensor for acquiring data, pre- and post-maintenance data storage, a feature detection algorithm database, and units for detecting features and selecting optimal algorithms, allowing for the diagnosis of equipment failures by identifying differences in measurement data before and after maintenance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dedicated devices are used for predictive abnormality diagnosis, then diagnosis accuracy is improved, but device cost increases and mass production becomes difficult
Solution Approach 1:
The patent applies universality by designing a diagnosis system that can operate on general-purpose devices rather than requiring dedicated specialized hardware. The system uses standard computational components that can be manufactured and deployed across multiple platforms, enabling mass production while maintaining diagnostic capabilities through software-based processing of measurement data.
Solution Approach 2:
The patent employs copying by using general-purpose computing devices that can replicate diagnostic functions through software rather than requiring custom-hardware copies. The diagnosis logic is captured in algorithms and data structures that can be copied and deployed across standard devices, eliminating the need for expensive dedicated hardware replication.
2Measurement precision
If multiple feature detection algorithms are evaluated, then optimal algorithm selection is improved, but processing load and system complexity increase
Solution Approach 1:
The patent applies preliminary action by pre-storing multiple feature detection algorithms in a database before actual diagnosis operations. This allows the system to evaluate and select optimal algorithms in advance based on historical performance, rather than attempting all algorithms during real-time operation, thereby reducing processing load while maintaining detection accuracy.
Solution Approach 2:
The system applies self-service by automatically evaluating and selecting optimal feature detection algorithms based on stored performance data and current diagnostic needs. The algorithm selection process is automated through the controller, which independently determines the most suitable algorithm without requiring manual intervention or complex external configuration.
3Stability of the object's composition
If measurement data from maintenance periods is discarded, then model stability is improved, but useful diagnostic information is lost
Solution Approach 1:
The patent converts the previously harmful or problematic maintenance period data into beneficial diagnostic information. By recognizing that maintenance events create distinct patterns in measurement data, the system uses these periods to detect maintenance signs and assess equipment state changes, transforming what was discarded information into valuable diagnostic insights.
Solution Approach 2:
The system applies parameter changes by adjusting how maintenance period data is handled in the learning process. Instead of discarding this data or using fixed values, the system incorporates measurement data from maintenance periods with appropriate weighting or processing, allowing the normal model to adapt to state changes while maintaining stability through controlled parameter adjustments.
Data Source
AI summary
A failure diagnosis system flexibly responds to a change in a diagnosis target by using a difference in measurement data before and after maintenance in predictive failure diagnosis. A pre-maintenance data DB stores measurement data before maintenance, and a post-maintenance data DB stores measurement data after maintenance. A feature detection algorithm group DB is provided where a plurality of feature detection algorithms are stored. A first feature is detected based on the measurement data by using each of the plurality of feature detection algorithms read from the feature detection algorithm group DB. An algorithm search unit selects one of the plurality of algorithms based on the feature thus detected. A second feature is detected from the measurement data by using the feature detection algorithm, and a sign predictive of failure of diagnosis of target equipment is diagnosed using the detected second feature.