Supplier Qualification Using Linear Mixed Models for Batch Effects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for qualifying suppliers of manufactured parts do not adequately account for batch effects, leading to incorrect variance estimates and potential qualification of inadequate suppliers, which can result in increased risk and higher testing costs.
Innovation Solution
A method involving the use of a linear mixed model to account for batch effects by measuring quantifiable properties in samples, selecting an appropriate model, fitting it to empirical data, and computing acceptance parameters such as process capability indices or tolerance interval bounds to determine supplier qualification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If known techniques are used to qualify a process without accounting for batch effects, then the qualification process is simpler and faster, but incorrect variance estimates are produced leading to qualification of inadequate suppliers
Solution Approach 1:
The patent transforms the qualification approach by changing the statistical parameters used - specifically incorporating batch effect variables into the linear mixed model. This allows the model to account for between-batch variation while maintaining computational efficiency, thus improving reliability without sacrificing productivity
Solution Approach 2:
The patent introduces a linear mixed model as an intermediary statistical framework that mediates between the simplicity of traditional qualification methods and the accuracy needed to account for batch effects. This intermediary model enables proper variance estimation by separating within-batch and between-batch variation components
2Device complexity
If batch effects are not properly accounted for in variance estimation, then the qualification process remains simple, but inadequate suppliers may be qualified increasing risk
Solution Approach 1:
The patent segments the total variation into distinct components using the linear mixed model - specifically separating within-batch variation from between-batch variation. This segmentation allows each source of variation to be properly estimated and accounted for, preventing inadequate suppliers from being qualified while maintaining a manageable analytical framework
3Quantity of substance
If traditional qualification methods are used ignoring batch effects, then testing costs are lower, but false alarm rates increase leading to higher monitoring costs
Solution Approach 1:
The patent implements a feedback mechanism through the linear mixed model that continuously accounts for batch effects in the variance estimation. This feedback loop prevents false alarms by properly attributing variation to batch effects rather than treating them as process instability, thereby reducing unnecessary retesting and monitoring costs
Data Source
AI summary
Techniques for qualifying a candidate supplier are presented. Such techniques may include obtaining a part produced by a candidate supplier and measuring a quantifiable property in each of a plurality of samples to obtain an empirical data set. Such techniques may also include selecting, based on the empirical data set, and fitting to the empirical data set, an appropriate linear mixed model for the quantifiable property. Such techniques may further include computing an acceptance parameter from a mean and standard deviation obtained from the appropriate linear mixed model. The acceptance parameter may include a process capability index or a tolerance interval bound. Such techniques may further include determining that the candidate supplier qualifies based on comparing the acceptance parameter to a threshold.