Event Prediction Model Input Selection for Emerging Asset Failures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing event prediction models for asset failures rely heavily on subject matter expert input, which limits the selection of input data variables to those known by the expert, often missing relevant variables and failing to account for new failure modes, leading to inaccurate and untimely predictions.
Innovation Solution
A data-driven approach is employed by an asset data analytics platform to select input data variables for event prediction models, involving initial variable selection, classification, scoring, and transformation based on historical data analysis, to identify the most predictive variables and improve model accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If subject matter expert input is used to select input data variables, then the model development process is simpler and more controllable, but the selection of input variables is limited to those known by the expert, missing relevant variables and new failure modes
Solution Approach 1:
The system performs automatic variable selection using algorithms that analyze historical data and event information independently, without requiring manual expert intervention for each variable selection decision. The processor automatically evaluates potential input variables based on their relationship to target events, enabling the system to discover relevant variables including new failure modes that experts may not have identified.
Solution Approach 2:
The manual expert judgment process is replaced with automated computational algorithms. Instead of relying on human experts to identify and select variables, the system uses processors to automatically analyze data relationships, calculate variable importance metrics, and select optimal input variables through computational methods.
2Measurement precision
If more input data variables are included in the event prediction model, then the prediction accuracy may improve, but the computational cost and complexity increase
Solution Approach 1:
The system extracts and selects only the most relevant input variables from a larger pool of potential variables. By identifying and removing unnecessary or redundant variables through automated analysis, the model maintains high prediction accuracy while reducing computational complexity and processing requirements.
Solution Approach 2:
The system applies a balanced approach by selecting a subset of variables that provides sufficient prediction accuracy without including all possible variables. This partial selection approach avoids the excessive computational burden of using every available variable while still capturing the essential patterns needed for accurate predictions.
3Reliability
If manual variable selection by experts is used, then the model development time is reduced for known failure modes, but the model fails to account for new failure modes and emerging patterns
Solution Approach 1:
The system performs preliminary analysis of historical data and event information to pre-identify potential input variables and their relationships before model deployment. This preliminary computational action enables the model to be pre-configured with relevant variables for known failure modes while maintaining the flexibility to adapt to new patterns as they emerge in the data.
Solution Approach 2:
The system uses feedback from historical event data and prediction outcomes to continuously refine variable selection. By analyzing actual event occurrences and their associated data patterns, the system learns to identify relevant variables for both known and emerging failure modes, improving model reliability while maintaining adaptability to new situations.
Data Source
AI summary
Disclosed is a process for creating an event prediction model that employs a data-driven approach for selecting the model's input data variables, which, in one embodiment, involves selecting initial data variables, obtaining a respective set of historical data values for each respective initial data variable, determining a respective difference metric that indicates the extent to which each initial data variable tends to be predictive of an event occurrence, filtering the initial data variables, applying one or more transformations to at least two initial data variables, obtaining a respective set of historical data values for each respective transformed data variable, determining a respective difference metric that indicates the extent to which each transformed data variable tends to be predictive of an event occurrence, filtering the transformed data variables, and using the filtered, transformed data variables as a basis for selecting the input variables of the event prediction model.