Sensor Group Analysis for Plant Abnormality Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In large-scale plant facilities, determining abnormalities is challenging due to the complexity of correlations between numerous sensors, as individual sensors' detection signals may not exceed thresholds but still indicate abnormalities, and vice versa, relying on experienced operators for monitoring correlations is inaccurate and limited.

Innovation Solution

An examining apparatus that acquires sensor data from targeted groups, learns analysis models using machine learning, and outputs alarms for abnormalities, with a user interface for feedback to improve model accuracy, excluding irrelevant data periods, and selecting appropriate models for examination.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If individual sensor detection signals are monitored against thresholds, then simple anomaly detection is possible, but accuracy deteriorates because individual signals may not exceed thresholds even when abnormalities exist

Engineering Contradiction:
Improveabnormality detection accuracyVSAvoidsensor correlation analysis complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the complex sensor data analysis into multiple components: individual sensor threshold monitoring, correlation relationship analysis between sensors, and integrated abnormality determination. This segmentation allows the system to handle complexity systematically while improving detection accuracy by considering both individual and relational aspects of sensor data.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary analysis layer that processes correlation relationships between sensor signals. This intermediary component analyzes how sensors relate to each other and mediates between simple threshold monitoring and complex abnormality determination, enabling accurate detection without requiring direct human analysis of all sensor correlations.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If correlation analysis between multiple sensors is performed manually by experienced operators, then abnormality detection accuracy can be improved, but productivity deteriorates due to time-consuming monitoring

Engineering Contradiction:
Improveabnormality detection accuracyVSAvoidexamination efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent implements self-service by enabling the system to automatically perform correlation analysis and abnormality determination without requiring continuous human intervention. The automated analysis engine processes sensor data, identifies correlations, and determines abnormalities independently, freeing operators from time-consuming manual monitoring while maintaining high detection accuracy.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical system of manual operator analysis with an automated computational system. Instead of relying on human operators to manually correlate sensor signals, the system uses algorithms to automatically analyze correlations and detect abnormalities, dramatically improving examination efficiency while preserving detection accuracy.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Adaptability or versatility

If all sensor data is used for model learning, then model comprehensiveness is improved, but learning accuracy deteriorates due to inclusion of irrelevant data periods

Engineering Contradiction:
Improvemodel coverageVSAvoidmodel learning accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent applies preliminary action by performing data filtering and period identification before the actual model learning process. The system pre-processes sensor data to identify and exclude irrelevant periods (such as maintenance periods, startup periods, or abnormal operating conditions) before training the machine learning model. This preliminary filtering ensures that only relevant data is used for learning, improving model accuracy while maintaining comprehensive coverage of normal operating conditions.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3489780B1Examining apparatus, examining method, program and recording medium
Publication Date: 2023.12.27 YOKOGAWA ELECTRIC CORP
  • EP3489780B1 patent drawingFigure 1
  • EP3489780B1 patent drawingFigure 2
  • EP3489780B1 patent drawingFigure 3

AI summary

To easily perform examination of at least one facility based on detection signals of a plurality of sensors installed in the facility. Provided are an examining apparatus, an examining method, a program and a recording medium, including: a group designation acquiring unit (120) to acquire designation of a targeted group including a plurality of targeted sensors (20) to be analyzed among a plurality of sensors (20) installed in at least one facility (10); a sensor data acquiring unit (110) to acquire sensor data from each targeted sensor included in the targeted group; a learning unit (160) to learn an analysis model by using the sensor data from each targeted sensor included in the targeted group; and an examining unit (190) to examine the facility (10) by using the learned analysis model.