Adaptive Prognostic Surveillance for Asset Aging Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional ML-based prognostic-surveillance systems struggle to adapt to age-related changes in monitored assets, leading to nuisance alarms due to normal aging phenomena, which can result in false alerts and reduced sensitivity to actual degradation modes.
Innovation Solution
A system that automatically updates its ML model by calculating a reward/cost metric to swap the trained inferential model with age-specific models, trained to account for aging phenomena, using a restructurable adaptive controller to minimize false alarms and maintain high sensitivity throughout the asset's lifetime.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a conventional ML-based prognostic-surveillance system uses a fixed trained inferential model to monitor assets, then the system can detect incipient anomalies effectively during normal operation, but the system generates nuisance alarms when assets experience normal aging phenomena that change signal correlations
Solution Approach 1:
The system dynamically adapts the inferential model based on the asset's age. Instead of using a fixed model, the system automatically updates the model parameters and structure as the asset ages, allowing the surveillance system to adapt to changing signal correlations while maintaining anomaly detection accuracy. The adaptation is triggered when aging patterns are detected, transforming the static ML system into a dynamic one that evolves with the asset lifecycle.
Solution Approach 2:
The system changes the parameters of the inferential model based on asset age. When normal aging phenomena are detected through monitoring, the system automatically adjusts the model parameters to account for age-related changes in signal correlations. This parameter adaptation allows the system to distinguish between normal aging variations and actual anomalies, reducing nuisance alarms while maintaining detection sensitivity.
2Reliability
If the ML model is frequently updated to account for aging phenomena, then the system maintains high sensitivity to detect new degradation modes, but the system complexity and computational resources increase
Solution Approach 1:
The system implements a feedback mechanism where monitoring data from the aging asset is continuously fed back to the adaptation module. This feedback loop automatically triggers model updates only when aging patterns are detected, rather than continuously updating. The feedback-driven approach maintains high sensitivity to degradation modes while avoiding unnecessary updates that would increase system complexity and computational burden.
Solution Approach 2:
The prognostic-surveillance system performs self-adaptation through automated model updating without requiring external intervention. The system autonomously detects aging patterns, triggers appropriate model updates, and maintains optimal surveillance performance. This self-service capability reduces the need for complex manual model management while maintaining high sensitivity to degradation modes throughout the asset lifecycle.
3Object-generated harmful factors
If the system distinguishes between normal aging and actual degradation, then false alarms are reduced, but the difficulty of detecting and measuring aging patterns increases
Solution Approach 1:
The system performs preliminary action by pre-training inferential models for different asset age stages and pre-identifying normal aging patterns. Before actual aging occurs, the system prepares multiple models corresponding to different lifecycle stages. When the asset ages, the system switches to the appropriate pre-trained model, enabling it to distinguish normal aging from actual degradation without requiring complex real-time analysis of aging patterns.
Solution Approach 2:
The system introduces an intermediary adaptation module that mediates between the monitoring system and the inferential model. This intermediary automatically detects aging patterns in the monitoring data and triggers appropriate model selections or updates. By introducing this intermediary layer, the system simplifies the detection of aging patterns while effectively reducing false alarms, as the intermediary handles the complexity of pattern recognition and model adaptation.
Data Source
AI summary
The disclosed embodiments relate to a system that automatically adapts a prognostic-surveillance system to account for aging phenomena in a monitored system. During operation, the prognostic-surveillance system is operated in a surveillance mode, wherein a trained inferential model is used to analyze time-series signals from the monitored system to detect incipient anomalies. During the surveillance mode, the system periodically calculates a reward/cost metric associated with updating the trained inferential model. When the reward/cost metric exceeds a threshold, the system swaps the trained inferential model with an updated inferential model, which is trained to account for aging phenomena in the monitored system.