ML Training Vector Selection via Local Extrema
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ML models for prognostic-surveillance operations face challenges in reducing false alarms due to noise in time-series sensor data, which leads to 'clipping' of estimates when encountering noise values outside the training range, resulting in unnecessary asset downtime and service requests.
Innovation Solution
The system selects training vectors based on local minimum and maximum values from non-overlapping time windows of sensor data, ensuring these values fall into a training set, while using a validation set to tune model parameters and reduce errors, thereby maximizing the operational range and minimizing false alarms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training data includes all possible noise variations, then prediction accuracy improves, but data collection becomes infeasible due to infinite noise combinations
Solution Approach 1:
The patent performs preliminary identification of local extrema (maximum and minimum values) in time-series data before training the ML model. By pre-selecting these critical data points that represent the full operational range, the system prepares a comprehensive yet manageable training dataset that covers all possible noise variations without requiring infinite data collection
Solution Approach 2:
The patent extracts only the essential features (local extrema values) from the raw time-series data. Instead of using all data points including redundant information, the system identifies and extracts only the maximum and minimum values within specified time windows, creating a condensed training dataset that maintains prediction accuracy while reducing data volume
2Reliability
If training data covers full operational range, then model reliability improves, but false alarms increase due to noise clipping
Solution Approach 1:
The patent performs preliminary identification of local extrema (maximum and minimum values) in time-series data before training the ML model. By pre-selecting these critical data points that represent the full operational range, the system prepares a comprehensive yet manageable training dataset that covers all possible noise variations without requiring infinite data collection
Solution Approach 2:
The patent extracts only the essential features (local extrema values) from the raw time-series data. Instead of using all data points including redundant information, the system identifies and extracts only the maximum and minimum values within specified time windows, creating a condensed training dataset that maintains prediction accuracy while reducing data volume
3Stability of the object's composition
If ML model estimates are clipped to training range, then prediction stability improves, but false alarms increase due to noise values outside training envelope
Solution Approach 1:
The patent performs preliminary identification of local extrema (maximum and minimum values) in time-series data before training the ML model. By pre-selecting these critical data points that represent the full operational range, the system prepares a comprehensive yet manageable training dataset that covers all possible noise variations without requiring infinite data collection
Solution Approach 2:
The patent extracts only the essential features (local extrema values) from the raw time-series data. Instead of using all data points including redundant information, the system identifies and extracts only the maximum and minimum values within specified time windows, creating a condensed training dataset that maintains prediction accuracy while reducing data volume
Data Source
AI summary
A system is provided for training an inferential model based on selected training vectors. During operation, the system receives training data comprising observations for a set of time-series signals gathered from sensors in a monitored system during normal fault-free operation. Next, the system divides the observations into N subgroups comprising non-overlapping time windows of observations. The system then selects observations with a local minimum value and a local maximum value for all signals from each subgroup to be training vectors for the inferential model. Finally, the system trains the inferential model using the selected training vectors. Note that by selecting observations with local minimum and maximum values to be training vectors, the system maximizes an operational range for the training vectors, which reduces clipping in estimates subsequently produced by the inferential model and thereby reduces false alarms.