Off-Duty-Cycle-Robust ML for Anomaly Detection in Assets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning prognostics for IoT-connected assets face challenges in detecting anomalies during intermittent and variable off-duty cycles, where assets are intermittently turned on and off at random times, leading to difficulties in data aggregation and false or missed alerts.
Innovation Solution
The development of Off-Duty-Cycle-Robust (ODCR) ML systems that utilize the Multivariate State Estimation Technique (MSET) to monitor assets with random downtimes, automatically detecting and handling transitions between on-duty and off-duty cycles, and accommodating variable off-duty durations without external curation, ensuring accurate anomaly detection without false or missed alerts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional ML prognostics are applied to assets with intermittent off-duty cycles, then anomaly detection capability is reduced, but system complexity increases due to need for external curation and handling of random downtimes
Solution Approach 1:
The ODCR ML system automatically detects and handles off-duty cycles without requiring external curation or manual intervention. The system self-adjusts to random downtimes by detecting transitions between on-duty and off-duty states and automatically switching between appropriate ML models, eliminating the need for complex external management while maintaining high anomaly detection accuracy
Solution Approach 2:
The system dynamically switches between different ML models based on the asset's operational state. When the asset is in on-duty mode, one ML model is applied; when in off-duty mode, a different ML model is used. This dynamic adaptation allows the system to maintain high measurement precision across varying operational conditions without requiring complex static architecture
2Reliability
If ML models are trained on time series data with random downtimes, then false alerts increase, but data aggregation becomes more difficult
Solution Approach 1:
The time series data is segmented into distinct on-duty and off-duty portions based on detected transitions. Each segment is then processed by the appropriate ML model trained for that specific operational state. This segmentation approach prevents mixing of data from different operational modes, thereby reducing false alerts while maintaining straightforward data aggregation through automatic transition detection
Solution Approach 2:
The system introduces an intermediary mechanism that detects transitions between on-duty and off-duty states and automatically routes data to the appropriate ML model. This intermediary layer simplifies data aggregation by handling the complexity of random downtimes automatically, preventing false alerts caused by inappropriate model application while requiring no manual data curation
3Adaptability or versatility
If off-duty cycles are handled with fixed intervals, then adaptability to random downtimes decreases, but system complexity increases with variable duration handling
Solution Approach 1:
The system automatically adapts to random off-duty cycles by detecting transitions and self-managing the switching between ML models. This self-service capability provides high adaptability to variable duration downtimes without requiring complex external scheduling or manual configuration, as the system handles all adaptations autonomously based on real-time asset state
Data Source
AI summary
Systems, methods, and other embodiments associated with off-duty-cycle-robust machine learning for anomaly detection in assets with random downtimes are described. In one embodiment, a method includes inferring ranges of asset downtime from spikes in a numerical derivative of a time series signal for an asset; extracting an asset downtime signal from the time series signal based on the inferred ranges of asset downtime; determining that the asset downtime signal carries telemetry based on the variance of the asset downtime signal; training a first machine learning model for the asset downtime signal; detecting a first spike in the numerical derivative of the time signal that indicates a transition to asset downtime; and in response to detection of the first spike, monitoring the time series signal for anomalous activity with the trained first machine learning model.