Mahalanobis Distance Evaluation via Data Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing method for abnormality cause diagnosis using Mahalanobis distance requires inverse matrix calculation, leading to increased calculation time, especially with a large number of features.
Innovation Solution
An evaluation device that groups unit data into multiple groups, calculates per-group unit spaces, and computes Mahalanobis distances for signal data within each group, then combines these distances to reduce the overall calculation burden.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the Mahalanobis distance is calculated using the conventional method, then the evaluation accuracy is maintained, but the calculation time increases significantly due to inverse matrix calculation
Solution Approach 1:
The patent divides the unit data collection into multiple groups (first group, second group, etc.) and calculates the Mahalanobis distance separately for each group using their respective unit spaces. This segmentation approach reduces the computational complexity by avoiding a single large-scale inverse matrix calculation, thereby decreasing calculation time while maintaining evaluation accuracy through the combined results from all groups.
2Measurement precision
If the number of data items is increased to improve evaluation accuracy, then the measurement precision improves, but the calculation time increases due to increased computational burden
Solution Approach 1:
The patent segments the unit data into multiple groups, each with its own unit space and Mahalanobis distance calculation. This allows the system to handle a large total number of data items while maintaining computational efficiency, as each group's contribution is calculated independently and then combined, rather than processing all data items as a single large dataset.
Solution Approach 2:
The patent introduces a new dimension of organization by grouping data items into multiple categories (first group, second group, etc.), each processed through separate unit space calculations. This dimensional reorganization allows the system to scale with the number of data items while controlling computational complexity through the structured combination of group-level results.
Data Source
AI summary
An evaluation device according to the present disclosure groups unit data in a unit data collection into two or more groups each including at least one of a plurality of data items, calculates, as a per-group unit space, a unit space in each of two or more per-group unit data collections which are unit data collections obtained by grouping the unit data into the two or more groups, calculates, using the per-group unit space calculated, a first Mahalanobis distance of a corresponding per-group unit data collection and a second Mahalanobis distance of a signal data collection in a corresponding data item, calculates a first linear combination of a plurality of first Mahalanobis distances calculated and a second linear combination of a plurality of second Mahalanobis distances calculated, compares the first and second linear combinations calculated, and outputs the comparison result as a first comparison result.