Facility State Learning Model With Measurement Drift Removal

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional models used for diagnosing plant facilities are situation-dependent and struggle to adapt when feature drift occurs in measurement data over time, requiring frequent updates and increasing processing loads.

Innovation Solution

A learning apparatus and method that acquires and preprocesses measurement data to reduce drift, allowing for the learning and application of models across varying conditions, including feature quantity calculation and data conversion to remove drift, thereby minimizing the need for re-learning and reducing processing loads.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a model is educated by using learning data obtained in a specific situation, then the model can accurately diagnose facilities in that situation, but the model cannot be applied to diagnose facilities in other situations due to situation dependency

Engineering Contradiction:
Improvediagnosis accuracyVSAvoidmodel applicability
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent transforms the learning data by changing its parameters through drift removal processing. Specifically, it converts measurement data that contains drift (systematic changes over time) into drift-free data by applying mathematical transformations. This allows the model to learn from data in a standardized form that is not tied to specific temporal conditions, thereby improving both diagnosis accuracy and model applicability across different situations.

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If drift removal processing is performed on learning data, then the model can be applied across different situations, but processing time and computational load increase

Engineering Contradiction:
Improvemodel applicabilityVSAvoidprocessing time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent performs drift removal processing as a preliminary step before model learning, rather than during or after learning. By pre-processing the learning data to remove drift beforehand, the system enables the model to be trained once on standardized data and then applied repeatedly to different situations without requiring additional drift removal processing for each application, thus reducing overall processing time.

Inventive Principle:
Principle #10Preliminary action

3Reliability

If drift removal processing is performed on determination data, then consistent diagnosis results can be obtained across different situations, but processing load increases

Engineering Contradiction:
Improvediagnosis consistencyVSAvoidprocessing complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies the same drift removal processing method used for learning data to the determination data. By copying the preprocessing approach from the learning phase to the determination phase, the system ensures consistency in how data is handled, which improves diagnosis reliability while keeping the processing methodology simple and reusable.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3859455B1Learning apparatus, learning method, learning program, determination apparatus, determination method, determination program, and computer readable medium
Publication Date: 2022.10.05 YOKOGAWA ELECTRIC CORP
  • EP3859455B1 patent drawingFigure 1
  • EP3859455B1 patent drawingFigure 2
  • EP3859455B1 patent drawingFigure 3

AI summary

[Problem] In a model for determining a state of a facility, an influence of a drift of measurement data is reduced. [Solution to Problem] A learning apparatus is provided, which comprises: a learning data acquiring unit for acquiring learning data including measurement data obtained by measuring a facility and a state of the facility; a learning pre-processing unit for performing a pre-processing for reducing a drift of the measurement data in the learning data and outputting pre-processed learning data; and a learning processing unit for performing a processing for learning a model for determining the state of the facility from the pre-processed measurement data, by using the pre-processed learning data.