Augmented Reliability Models Using ML Feature Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional theoretical reliability models fail to accurately predict the reliability of products after high-volume manufacturing and field use, as they do not account for practical test and in-use data, leading to potential failures and underperformance.
Innovation Solution
An augmented reliability performance model is generated by developing a machine learning model based on manufacturing and testing data, identifying critical features, and incorporating these features into a design reliability model to improve prediction accuracy and product reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If theoretical reliability models are used during product design phases, then the design process can proceed with reliability predictions, but the prediction accuracy does not match the practical reliability of manufactured products
Solution Approach 1:
The patent implements feedback by using field failure data and reliability testing data from actual product usage to update and refine theoretical reliability models. This creates a closed-loop system where practical data flows back into the design phase to improve future predictions, resolving the contradiction between theoretical predictions and practical reliability.
Solution Approach 2:
The patent applies preliminary action by collecting and analyzing reliability data from field usage and testing before finalizing the reliability model for design applications. This ensures that the model incorporates real-world performance characteristics upfront, improving prediction accuracy from the outset rather than relying solely on theoretical assumptions.
2Reliability
If manufacturing data from high volume production is used to build reliability models, then the models become more practical, but the data sets are too small and do not include enough reliability failures for proper modeling
Solution Approach 1:
The patent merges multiple data sources including field failure data, reliability testing data, and manufacturing data into a comprehensive dataset. This combination increases the quantity and diversity of failure data available for modeling, enabling more accurate practical reliability models even when individual data sources are limited.
Solution Approach 2:
The patent creates a multi-functional reliability modeling system that can handle various types of data (field failures, testing data, manufacturing variations) and apply them across different product design scenarios. This universal approach maximizes the utility of limited data by extracting reliable patterns applicable to broader contexts.
3Ease of manufacture
If conventional theoretical reliability models are used, then the design process is straightforward, but the models do not account for process manufacturing variations and field conditions
Solution Approach 1:
The patent transforms the reliability modeling approach by changing key parameters from fixed theoretical values to dynamic values that incorporate manufacturing variations and field conditions. This allows the model to account for real-world variability while maintaining computational efficiency and integration into the design process.
Data Source
AI summary
A method for generating a reliability performance model includes developing a reliability prediction machine learning model for predicting reliability performance of a product based on data obtained from manufacturing and testing of the product, and obtaining feature names for the reliability prediction machine learning model and their predictive power values. The feature names may correspond to features from the data obtained from manufacturing and testing of the product. The method may further include extracting a set of feature names corresponding to features having highest predictive power values from the feature names, and generating a reliability performance model using one or more model parameters derived from the set of feature names.