Power Electronics RUL Prediction With Dynamic Feature Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting the remaining useful life (RUL) of power electronic devices are inaccurate due to their reliance on inflexible mechanisms like predefined degradation rates or static mathematical models, which fail to consider individual device nuances and specific failure modes, leading to unreliable failure predictions.
Innovation Solution
A system that tracks the operation of power electronic devices to model their activity, using dynamic programming to select features and implement anomaly and RUL models, including a partial-least squares with cumulative sum algorithm and a self-organizing map with interacting multiple models, to predict anomalies and RUL based on historical data and failure mechanisms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If predefined degradation rates or static mathematical models are used to estimate remaining useful life, then the estimation process is simplified and can be implemented broadly, but the accuracy of failure predictions deteriorates due to inability to consider individual device characteristics
Solution Approach 1:
The patent transforms static mathematical models into dynamic machine learning models that continuously learn from device operation data. The system adapts to individual device characteristics by training models on historical data, enabling the prediction system to evolve and improve accuracy over time while maintaining broad applicability across different device instances.
Solution Approach 2:
The patent changes the parameters from fixed predefined degradation rates to dynamic, data-driven parameters extracted from actual device operation. By using machine learning models that process operational data, the system transforms static parameters into adaptive parameters that reflect individual device behavior patterns, thereby improving prediction accuracy without sacrificing implementability.
2Adaptability or versatility
If generalized approaches are used to predict remaining life for classes of devices, then the solution can be applied broadly without device-specific customization, but the precision of predictions deteriorates due to lack of specificity to individual device instances
Solution Approach 1:
The patent implements self-service through automated machine learning model training that uses each device's own operational data to create customized prediction models. The system automatically collects data, trains models, and updates predictions without requiring manual device-specific configuration, thereby achieving both broad applicability and individualized precision through autonomous adaptation.
Solution Approach 2:
The patent applies preliminary action by training machine learning models on historical device data before actual failure prediction is needed. The system pre-processes and learns from past operational patterns, creating ready-to-use prediction models that can immediately provide accurate predictions when deployed, combining broad applicability with device-specific precision.
3Device complexity
If static mathematical models with fixed degradation rates are used, then the model structure remains simple and computationally efficient, but the ability to capture actual device behavior trends deteriorates
Solution Approach 1:
The patent substitutes mechanical mathematical models with machine learning-based computational models. This replacement transitions from rigid, formula-based degradation rate calculations to flexible, data-driven predictions that can capture complex non-linear device behavior patterns, significantly improving reliability while managing complexity through automated model training and selection.
Data Source
AI summary
Systems, methods, and other embodiments described herein relate to predicting anomalous operation of a device and an associated remaining useful life (RUL). In one embodiment, a method includes acquiring usage information about operation of an electronic device. The method includes selecting at least one feature from the usage information according to an optimization. The method includes determining whether the least one feature indicates a presence of an anomaly in the operation of the electronic device according to an anomaly model. The method includes, responsive to detecting the anomaly, determining a remaining useful life (RUL) for the electronic device according to a RUL model. The method includes providing the RUL.