Operation Data Segmentation for Clean Anomaly Detection Training

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine training models struggle to accurately detect anomaly values from industrial system operation data, especially when trained on data that includes both normal and anomaly periods, leading to inefficiencies in maintaining industrial systems.

Innovation Solution

A training data generation apparatus that acquires operation data, divides it into temporary training and test data, detects anomaly values, and generates training data by excluding anomaly periods, ensuring the final data set includes only normal data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If operation data including both normal and anomaly periods is used as training data, then the machine training model can be trained with sufficient data volume, but the anomaly detection accuracy deteriorates

Engineering Contradiction:
Improvedata volumeVSAvoidanomaly detection accuracy
Core Design Contradiction:
Quantity of substanceVSMeasurement precision

Solution Approach 1:

The operation data is segmented into normal periods and anomaly periods through temporal division. The detection unit identifies anomaly periods by comparing operation data across different time windows, then the generation unit excludes these anomaly periods from training data. This segmentation allows sufficient data volume from normal periods while ensuring anomaly detection accuracy by removing contaminated anomaly segments.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If anomaly periods are excluded from training data to improve detection accuracy, then the anomaly detection accuracy improves, but the data processing complexity increases

Engineering Contradiction:
Improveanomaly detection accuracyVSAvoiddata processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary anomaly period detection and exclusion before the training data is actually needed. The detection unit pre-identifies anomaly periods by analyzing operation data patterns, and the generation unit pre-excludes these periods to create clean training data. This preliminary action ensures high anomaly detection accuracy while managing processing complexity by performing exclusions in advance rather than during real-time detection.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If manual inspection is used to identify anomaly periods, then the training data quality improves, but the time consumption increases

Engineering Contradiction:
Improvetraining data qualityVSAvoidtime consumption
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs self-service anomaly period identification through automated detection algorithms. The detection unit automatically identifies anomaly periods by comparing operation data patterns across different time windows without requiring manual inspection. This self-service approach maintains high training data quality while dramatically reducing time consumption compared to manual methods.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4614268A1Training data generation apparatus, method, and program
Publication Date: 2025.09.10 KK TOSHIBA
  • EP4614268A1 patent drawingFigure 1
  • EP4614268A1 patent drawingFigure 2
  • EP4614268A1 patent drawingFigure 3

AI summary

According to one embodiment, a training data generation apparatus (1) includes an acquisition unit (11), a division unit (12), a detection unit (13), and a generation unit (14). The acquisition unit (11) acquires operation data related to an operation state of a device in a predetermined period. The division unit (12) divides the operation data into at least temporary training data and temporary test data. The detection unit (13) detects an anomaly value from the temporary test data based on the temporary training data. The generation unit (14) generates training data from the operation data by excluding an anomaly period in which the anomaly value is detected from the predetermined period.