Automatic Data Screening for ML Prognostic Surveillance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ML-based prognostic-surveillance techniques face challenges in matching appropriate datasets with specific techniques, dealing with signal quality issues such as missing values, low-resolution signals, and synchronization problems, which lead to false or missed alarms and increased storage costs.
Innovation Solution
An automatic data-screening framework and intelligent preprocessing pipeline that determine whether time-series signals are univariate or multivariate, perform necessary preprocessing operations like imputation, despiking, and synchronization, and select appropriate analysis techniques like MSET for multivariate signals, reducing false alarms and storage needs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional ML techniques are used to analyze all time-series data, then comprehensive anomaly detection is achieved, but storage costs and processing overhead increase significantly
Solution Approach 1:
The patent extracts and removes normal, expected patterns from time-series data during the training phase. The trained ML model learns to recognize typical correlations and deviations, allowing the system to discard storing all raw data while retaining only the essential anomaly detection capabilities. This extraction principle enables the system to achieve comprehensive anomaly detection without storing complete historical datasets.
Solution Approach 2:
The patent performs preliminary data processing and model training before actual surveillance operations. During this preliminary phase, the ML model learns from historical data to establish baseline correlations. This preliminary action allows the system to operate efficiently during surveillance by comparing new data against pre-learned patterns, reducing the need for extensive real-time storage and processing of all raw data.
2Adaptability or versatility
If data from multiple sensors with varying sampling rates is collected, then comprehensive monitoring coverage is achieved, but signal synchronization problems and false alarms increase
Solution Approach 1:
The patent implements a universal time-stamping mechanism that works across all sensor types and sampling rates. Each data point from any sensor is tagged with a standardized time reference, allowing the ML model to universally process and correlate data from heterogeneous sources. This universal time-referencing approach enables comprehensive multi-sensor monitoring while maintaining accurate temporal relationships, thereby preventing false alarms caused by synchronization issues.
3Measurement precision
If fine-grained observations are stored over long time horizons, then detailed anomaly analysis is enabled, but processing overhead costs greatly increase
Solution Approach 1:
The patent segments the time-series data analysis into distinct phases: training phase where the ML model learns from historical data, and surveillance phase where the model performs efficient real-time anomaly detection. During training, detailed fine-grained data is processed to build the model. During surveillance, the trained model processes only new incoming data, comparing it against learned patterns. This segmentation enables detailed anomaly analysis during training while keeping ongoing processing overhead minimal.
4Manufacturing precision
If users manually match datasets to ML techniques, then appropriate analysis methods can be selected, but system complexity and user expertise requirements increase
Solution Approach 1:
The patent implements an automatic dataset characterization and technique selection system. The system automatically analyzes incoming data characteristics (such as univariate vs. multivariate, presence of correlations) and selects the appropriate ML technique without user intervention. This self-service approach eliminates the need for users to manually match datasets to techniques, reducing system configuration complexity while maintaining accurate technique-dataset matching through automated data profiling and algorithm selection.
Data Source
AI summary
The disclosed embodiments relate to a system that automatically selects a prognostic-surveillance technique to analyze a set of time-series signals. During operation, the system receives the set of time-series signals obtained from sensors in a monitored system. Next, the system determines whether the set of time-series signals is univariate or multivariate. When the set of time-series signals is multivariate, the system determines if there exist cross-correlations among signals in the set of time-series signals. If so, the system performs subsequent prognostic-surveillance operations by analyzing the cross-correlations. Otherwise, if the set of time-series signals is univariate, the system performs subsequent prognostic-surveillance operations by analyzing serial correlations for the univariate time-series signal.