Waveform Frequency Filtering for Accurate Factory Diagnosis Models

Solution Overview

Problem

The existing diagnosis system for factory automation apparatuses faces challenges in maintaining high accuracy of the normal model of waveforms due to the use of non-hierarchical cluster analysis methods, leading to laborious processing and variability in analysis results.

Innovation Solution

A data analysis device that includes a data collector and a data organizer, which removes waveforms with appearance frequencies lower than a reference frequency, generating highly-accurate learning data by focusing on waveforms with higher appearance frequencies.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If non-hierarchical cluster analysis method is used to generate normal model of waveform, then the model can be created, but the accuracy of the normal model cannot be maintained due to variability in analysis results

Engineering Contradiction:

Improveaccuracy of normal modelVSAvoidconsistency of analysis result

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The patent extracts and removes waveforms with low appearance frequencies from the dataset before generating the normal model. This extraction of problematic data points eliminates the source of variability that causes inconsistent analysis results, thereby improving both the accuracy and reliability of the normal model.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces a frequency-based filtering parameter (appearance frequency threshold) to change the composition of the dataset. By adjusting this parameter to remove low-frequency waveforms, the analysis becomes more consistent and reliable, directly addressing the variability issue in non-hierarchical cluster analysis.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If non-hierarchical cluster analysis method is used to generate normal model of waveform, then the model can be created, but the processing becomes laborious

Engineering Contradiction:

Improveaccuracy of normal modelVSAvoidprocessing time for model generation

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent extracts and removes waveforms with low appearance frequencies from the dataset before generating the normal model. This extraction of problematic data points eliminates the source of variability that causes inconsistent analysis results, thereby improving both the accuracy and reliability of the normal model.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces a frequency-based filtering parameter (appearance frequency threshold) to change the composition of the dataset. By adjusting this parameter to remove low-frequency waveforms, the analysis becomes more consistent and reliable, directly addressing the variability issue in non-hierarchical cluster analysis.

Inventive Principle:

Principle #35Parameter changes

3Adaptability or versatility

If all waveforms are included in learning data generation, then comprehensive coverage is achieved, but the learning data accuracy is reduced due to inclusion of rare or abnormal waveforms

Engineering Contradiction:

Improvecoverage of waveform typesVSAvoidaccuracy of learning data

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent extracts and removes waveforms with low appearance frequencies from the dataset before generating the normal model. This extraction of problematic data points eliminates the source of variability that causes inconsistent analysis results, thereby improving both the accuracy and reliability of the normal model.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces a frequency-based filtering parameter (appearance frequency threshold) to change the composition of the dataset. By adjusting this parameter to remove low-frequency waveforms, the analysis becomes more consistent and reliable, directly addressing the variability issue in non-hierarchical cluster analysis.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS11353860B2Data analysis device, system, method, and recording medium storing program

Publication Date: 2022.06.07 MITSUBISHI ELECTRIC CORP

AI summary

A data analysis device (1) according to the present disclosure includes a data collector (131) and a data organizer (135). The data collector (131) collects an input signal that is input from an apparatus connected to the data analysis device (1). The data organizer (135) removes, from among waveforms included in the input signal collected by the data collector (131), a waveform that has an appearance frequency lower than a reference frequency. The data organizer (135) generates learning data by collecting, from among the waveforms included in the input signal, waveforms that have an appearance frequency higher than the reference frequency.