Frequency Domain Correlation for Sensor Error Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information processing systems face challenges in generating significant correlation models from time-series data due to variations in data sampling periods, and frequency analysis methods struggle with error detection unless sensors with similar changing trends are selected.
Innovation Solution
The system converts time-series data into frequency data and uses correlation models to analyze the correlation between sensors, allowing for error detection regardless of changing trends, by generating correlation models from frequency data and comparing their strengths to determine if correlations have broken down.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If correlation analysis is performed on time-series data obtained at predetermined sampling rates, then error detection capability is improved, but correlation coefficients change only due to differences in data obtaining period making it difficult to generate significant models
Solution Approach 1:
The patent transforms the analysis from time domain to frequency domain by converting time-series data into frequency data through Fourier transform. This parameter transformation allows correlation analysis to be performed on frequency characteristics rather than temporal sequences, eliminating the problem where correlation coefficients change only due to sampling period differences. The frequency domain representation captures essential correlation patterns that are invariant to sampling rate variations.
2Reliability
If frequency analysis is performed using correlation coefficients, then error detection is possible, but correlation coefficients have signs that vary depending on changing direction making satisfactory analysis difficult unless sensors with same trend are selected
Solution Approach 1:
The patent inverts the conventional approach by instead of analyzing correlation in the time domain where sign variations occur, it performs correlation analysis in the frequency domain. The frequency domain representation eliminates the sign variation problem because frequency components represent magnitude and phase information independently of the direction of change in the time domain. This allows universal application across sensors with different changing trends.
3Adaptability or versatility
If correlation models are generated from frequency data, then error detection works regardless of changing trends, but additional frequency conversion processing is required
Solution Approach 1:
The patent substitutes the mechanical/time-based correlation analysis with a frequency-based approach. By using Fourier transform to convert time-series data into frequency data, the system replaces complex time-domain signal processing with more robust frequency-domain correlation analysis. This substitution simplifies the overall processing by making the correlation analysis invariant to time-domain variations such as sampling rates and changing trends.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An information processing system, an information processing method, and a program capable of satisfactory data analysis are provided. The information processing system includes a frequency conversion unit 205 which converts a plurality of pieces of time-series data obtained through detection carried out by a plurality of sensing units 201 into pieces of frequency data 208, respectively, a model construction unit 211 which generates a correlation model 213 using pieces of the frequency data 208 for at least two sensing units 201 from among the plurality of sensing units 201 and calculates a correlation strength 214 of the correlation model 213, and an error detection unit which determines whether an error has occurred based on the correlation strength 214.