Failure Sign Diagnosis Using Time-Shifted Sensor Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing failure sign diagnosis technologies for industrial instruments fail to consider hidden relationships among sensors, leading to suboptimal sensor combination generation and diagnosis performance, especially when dealing with sensors from systems with different clocks and time lags.
Innovation Solution
A preprocessing condition is generated to increase correlation coefficients between sensor data, and a failure sign diagnosis model is updated based on these conditions, allowing for the automatic generation of sensor combinations and consideration of hidden relationships between sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If correlation models are built based on input sensor data themselves (PTL 1), then automatic generation of sensor combinations is achieved, but hidden relationships among sensors are not considered leading to suboptimal diagnosis performance
Solution Approach 1:
The patent applies preliminary action by pre-processing sensor data to generate processed sensor data that reveals hidden relationships before correlation analysis is performed. This includes generating time-series data with corrected time lags and calculating variation amounts, which prepares the data in advance to expose correlations that would otherwise remain hidden in raw sensor data.
Solution Approach 2:
The patent introduces processed sensor data as an intermediary between raw sensor data and correlation analysis. This intermediary layer includes time-corrected data and derived metrics (variation amounts) that mediate the relationship between sensors, allowing the system to detect hidden correlations that direct analysis of raw data would miss.
2Device complexity
If sensor data from systems with different clocks are processed directly, then data processing is simplified, but time lags cause false negative correlation judgments
Solution Approach 1:
The patent changes the time parameter of sensor data by generating processed sensor data with corrected time lags. It calculates appropriate time shifts for each sensor based on system characteristics and generates time-series data where the temporal relationships are aligned, transforming the time parameter to eliminate false negative correlations caused by clock differences.
3Productivity
If sensor combinations are generated without considering hidden relationships, then analysis is faster and simpler, but diagnostic performance deteriorates due to irrelevant sensor data
Solution Approach 1:
The patent performs preliminary processing to generate processed sensor data that highlights relevant relationships before correlation analysis. By pre-calculating variation amounts and time-corrected data, it prepares the dataset to automatically reveal meaningful sensor combinations, making the subsequent correlation analysis both faster and more accurate without requiring manual domain expertise.
Data Source
AI summary
Sensor combinations that are effective for failure sign diagnosis are automatically generated in consideration of hidden relationships between sensor data. The present invention includes: a time shift data generation unit 106 that processes sensor data acquired from an instrument and generates the processed sensor data; and a sensor combinations/pre-processing condition generation unit 112 that generates sensor combinations on the basis of correlation coefficients between the sensor data and the processed sensor data, and generates, as a preprocessing condition, a processing condition that increases a correlation coefficient between sensors for each sensor combination, in which the failure sign diagnosis of the instrument is performed on the basis of a change in the relationship between the sensor data included in the sensor combination and the processed sensor data.