Design of Experiments Model Sampling Invalid Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current design of experiments (DOE) methods are less effective in identifying faults and problematic operational regimes of devices, as they rely on existing products and provide less-than-optimal data explanations, lacking in specificity and sensitivity.
Innovation Solution
The proposed method involves obtaining data from similar devices, thinning it to a minimum relevant subset using spatial voting, generating synthetic data, and employing gene expression programming or complete polynomial models to identify model boundaries and improve predictive accuracy for failure prediction and remaining useful life (RUL) analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If design of experiments (DOE) methods are used to identify faults and problematic operational regimes, then device analysis can be performed, but the methods provide less-than-optimal data explanations with reduced specificity and sensitivity
Solution Approach 1:
The patent transforms operational data from multiple devices by changing its representation parameters through spatial voting algorithms, converting raw operational data into a transformed space where patterns of device failures and operational regimes become more distinguishable and analyzable
Solution Approach 2:
The patent creates synthetic copies of operational data through spatial voting, generating representative data points that capture the essential characteristics of device behavior without requiring access to the actual physical devices, thereby improving data explanation quality
2Quantity of substance
If data from multiple devices is collected and analyzed using traditional methods, then more data is available, but the data volume increases without proportional improvement in model explanation quality
Solution Approach 1:
The patent extracts only the most relevant features and patterns from large volumes of operational data through spatial voting, identifying and isolating the critical information needed for accurate device failure prediction while discarding redundant data
Solution Approach 2:
The patent segments the operational data space into distinct regions using spatial voting, dividing the continuous data space into discrete cells that represent different operational regimes and failure modes, making the data more manageable and interpretable
3Reliability
If models are generated to explain device behavior, then predictive capability is achieved, but existing models fail to provide perfect explanation in certain operational regions
Solution Approach 1:
The patent employs dynamic model generation where the modeling approach adapts based on the operational region being analyzed, switching between different modeling techniques (spatial voting for certain regions, gene expression programming for others) to optimize predictive accuracy for each specific operational regime
Solution Approach 2:
The patent introduces spatial voting as an intermediary transformation layer between raw operational data and the final predictive models, creating a bridging representation that simplifies the relationship between input data and model outputs, thereby improving explanation quality without excessive model complexity
Data Source
AI summary
Presented herein are alternatives to design of experiments. A method can include sampling a model that explains a measurement corpus of measurement data to generate a sampled model, identifying an invalid region of the sampled model, determining whether a device will operate within the identified invalid region, if the device will operate within the identified invalid region, causing further measurement data to be captured in the identified invalid region, and generating a new model, based only on the further measurement data, to explain device operation within the identified invalid region that augments the sampled model to explain the device behavior.