Time-Series Data Grouping for Faster Abnormality Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing time-series data processing methods face challenges in efficiently analyzing large volumes of data from semiconductor manufacturing devices, as they require extensive time to detect abnormal data, and lack standardized methods for determining data abnormality and comparing different types of data.
Innovation Solution
A time-series data processing method that groups similar data, normalizes it, calculates an abnormality degree based on differences between data points, and displays a ranking to facilitate user analysis, including link information for easy access to detailed data rankings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If time-series data is measured with high frequency over a long time, then measurement precision is improved, but processing time increases and productivity deteriorates
Solution Approach 1:
The patent divides the enormous time-series data into multiple groups based on similarity of change patterns. By segmenting the data processing task into group-level operations, the system reduces the computational burden from analyzing every individual data point to analyzing representative groups, thereby maintaining measurement precision while significantly improving processing speed.
Solution Approach 2:
The patent introduces an intermediate processing layer that calculates representative values (such as average values or typical patterns) for groups of time-series data. This intermediary representation serves as a mediator between the raw high-frequency data and the final analysis, enabling efficient processing while preserving the essential characteristics needed for abnormality detection.
2Reliability
If multiple pieces of time-series data are mutually compared to detect abnormality, then abnormality detection accuracy is improved, but processing time increases
Solution Approach 1:
The patent merges multiple time-series data into groups based on their similarity in change patterns. By combining data with similar characteristics into single groups, the system maintains the ability to detect abnormalities through comparison while reducing the number of individual comparisons needed, thus improving efficiency without sacrificing detection accuracy.
Solution Approach 2:
The patent creates a universal grouping mechanism that can handle different types of time-series data (temperature, flow rate, etc.) using the same clustering approach. This multi-functional grouping system enables efficient abnormality detection across various data types simultaneously, reducing overall processing time while maintaining comprehensive detection accuracy.
3Adaptability or versatility
If different kinds of time-series data are compared, then analysis comprehensiveness is improved, but difficulty of comparison increases
Solution Approach 1:
The patent transforms different kinds of time-series data into a common parameter space by grouping them based on similarity of change patterns rather than their original physical units or meanings. This parameter transformation enables comprehensive comparison across different data types (temperature, flow rate, pressure, etc.) while simplifying the comparison process through unified clustering criteria.
Data Source
AI summary
A group generating section generates, from a plurality of pieces of time-series data, a plurality of groups each made up of a plurality of pieces of time-series data that change in a similar manner. A normalization section linearly transforms, for each group, data included in the time-series data in the group so that a maximum value and a minimum value of median values included in median-value time-series data are transformed to 1 and 0, respectively. An abnormality degree calculating section obtains, for each group, an average value of differences between pieces of data of the same time for every combination of two pieces of time-series data in the group, to take a maximum value of the obtained average values as an abnormality degree of the group. A ranking generating section generates a group ranking based on the abnormality degrees of the groups. The group rankings are displayed.