Time-Series Feature Selection for Malfunction Detection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Selecting appropriate feature values from various types calculable from time-series data for predictive maintenance is challenging due to the multitude of combinations possible using statistics and ranges, making it difficult to determine which ones to use for effective malfunction detection.
Innovation Solution
An information processing system that includes an obtaining module for time-series data, a decision module to determine different types of feature values based on a combination of a range-defining function and a statistic-defining function, and an assessment module to evaluate these feature values for accuracy of separation, allowing for the selection of more appropriate feature values and generation of a malfunction detection algorithm.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple types of feature values are calculated from time-series data using various combinations of statistics and ranges, then the comprehensiveness of malfunction detection improves, but the difficulty of selecting appropriate feature values increases
Solution Approach 1:
The system automatically evaluates and selects appropriate feature values without requiring manual expert intervention. The assessment module autonomously calculates accuracy of separation for multiple feature value types and identifies the most suitable ones, allowing the system to serve itself in the feature selection process rather than relying on external expertise.
Solution Approach 2:
The patent replaces the manual mechanical process of feature value selection with an automated computational system. Instead of experts manually evaluating and selecting feature values, the system uses the assessment module to automatically calculate separation accuracy and select optimal feature values, substituting human judgment with algorithmic evaluation.
2Ease of operation
If manual selection of feature values is performed based on expert knowledge, then the selection process is simple, but the reliability of malfunction detection decreases due to subjectivity and inconsistency
Solution Approach 1:
The assessment module provides objective feedback by calculating the accuracy of separation for each feature value type. This quantitative feedback mechanism allows the system to evaluate feature values based on their actual performance in separating normal from malfunctioning data, rather than relying on subjective expert judgment, thereby improving reliability while maintaining ease of operation.
Solution Approach 2:
The system changes the parameter used for feature value selection from subjective expert judgment to objective accuracy of separation metrics. By evaluating feature values based on their ability to separate normal and malfunctioning data quantitatively, the system transforms the selection criterion from a soft, subjective parameter to a hard, objective parameter, improving reliability.
3Measurement precision
If comprehensive assessment of all feature value types is performed, then the accuracy of malfunction detection improves, but the processing time and computational resources increase
Solution Approach 1:
The system performs comprehensive assessment of multiple feature value types to ensure accurate selection, evaluating all candidate feature values for accuracy of separation. This excessive action ensures that no potentially suitable feature value is missed, achieving high measurement precision in feature value selection despite the increased time and computational resources required.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An information processing system (1) includes: an obtaining module (230) configured to obtain time-series data from a control device (100); a decision module (232) configured to decide a plurality of different types of feature values (50) based on combination of a first function (30) that defines a range used for feature value calculation in a target piece of the time-series data and a second function (40) that defines a statistic used as feature value; and an assessment module configured to assess the time-series data for each of the plurality of different types of feature values.