Production Batch Quality Assessment Using Time-Warped Multi-Source Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining quality indicators in production batch runs face challenges due to varying time intervals and multi-variate data sources, leading to difficulties in accurately comparing time-series data and identifying quality categories, especially when patterns indicative of failures may be overlooked or detected too late.
Innovation Solution
A computer-implemented method that converts multi-variate time-series data into uni-variate data by multiplying source-specific data values with conversion factors and summing them at discrete time points, allowing for more efficient and accurate comparison using off-the-shelf software to determine similarity indices and identify characteristic patterns, thereby assessing quality categories in real-time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-variate time-series data is compared directly, then comprehensive quality assessment is achieved, but comparison accuracy deteriorates due to varying time intervals and data dimensions
Solution Approach 1:
The patent segments multi-variate time-series data into multiple uni-variate time-series, each representing a specific data source or parameter. This segmentation simplifies the comparison process by allowing individual uni-variate series to be analyzed separately using standard software, while still capturing comprehensive quality information through the combination of multiple segmented series.
Solution Approach 2:
The patent introduces an intermediary conversion process that transforms multi-variate time-series data into uni-variate representations. This intermediary step acts as a mediator between the complex multi-variate data and the simpler comparison operations, enabling accurate quality assessment through standard software while maintaining the essential characteristics of the original multi-source data.
2Reliability
If complex multi-variate time-series comparison is performed, then comprehensive pattern detection is achieved, but detection timing deteriorates (patterns detected too late)
Solution Approach 1:
By segmenting multi-variate time-series into uni-variate series, the patent enables parallel processing and earlier detection of patterns in individual data sources. This segmentation allows standard software to quickly analyze each uni-variate series independently, reducing overall detection time while maintaining reliable pattern identification through the combined analysis of multiple segmented series.
Solution Approach 2:
The patent applies partial action by initially focusing on converting and analyzing individual uni-variate time-series separately using standard software. This partial approach to the full multi-variate comparison problem enables faster initial pattern detection, with the option to perform more comprehensive multi-variate analysis if needed, thus reducing detection delay while maintaining adequate reliability.
3Ease of operation
If standard off-the-shelf software is used for time-series comparison, then ease of operation is improved, but measurement precision deteriorates due to inability to handle multi-variate data
Solution Approach 1:
The patent uses data conversion as an intermediary mechanism that bridges the gap between standard off-the-shelf software capabilities and the needs for accurate multi-variate time-series analysis. By converting multi-variate data into uni-variate representations, the patent enables the use of user-friendly standard software while maintaining measurement precision through the structured conversion process that preserves essential data characteristics.
Solution Approach 2:
The patent substitutes complex multi-variate data processing mechanics with simpler uni-variate processing mechanics that can be handled by standard software. This substitution replaces the need for specialized complex analysis tools with widely available standard software, improving ease of operation while maintaining adequate measurement precision through the conversion approach.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
To determine a quality indicator of production batch-run (220) of a production process (200), a computer (600) compares time-series with multi-source data from a reference batch-run (210) and time-series with multi-source data from the production batch-run (220). Before comparing, the computer converts multi-variate time-series to uni-variate time-series, by first multiplying data values of source-specific uni-variate time-series with source-specific factors from a conversion factor vector (610*) and second summing up the multiplied data values according to discrete time points. The source-specific factors of the conversion factor vector are obtained earlier by processing reference data, comprising the determination of characteristic portions of the time-series, converting, aligning by time-warping and evaluating displacement in time between characteristic portions before alignment and after alignment.