Sensor State Classification Using Mirrored Data Normalization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine learning models for plant control are highly dependent on the specific conditions and usage situations of sensors, making it difficult to apply models generated from one sensor's data to another sensor with different characteristics, leading to inefficiencies in model adaptation and increased time and costs for generating new models for each sensor.

Innovation Solution

An information processing system that includes a state data acquisition device, a first classification device for generating a determination line, a mirror image data generation device, and a second classification device to generate a determination model, which improves model versatility by canceling out obvious trends in sensor data and allowing for automatic adaptation across different sensors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a machine learning model is generated using measurement data from one sensor, then the model can accurately evaluate data from that specific sensor, but the model cannot be applied to other sensors with different usage situations or installation conditions

Engineering Contradiction:
Improvemodel accuracy for specific sensorVSAvoidmodel applicability across different sensors
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent extracts and removes sensor-specific characteristics (usage situation, installation conditions, deterioration state) from the measurement data through preprocessing operations. By separating these individual characteristics from the core evaluation targets, the model can focus on universal patterns while ignoring sensor-specific variations, enabling cross-sensor application without sacrificing accuracy.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent transforms measurement data by applying parameter changes that normalize sensor-specific variations. Through preprocessing steps such as normalization, standardization, or feature transformation, the data from different sensors with varying usage conditions are converted to a common parameter space, allowing a single model to accurately evaluate all sensors regardless of their individual characteristics.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If a new machine learning model is generated for each sensor to ensure accurate evaluation, then measurement precision is maintained, but time consumption and costs increase significantly

Engineering Contradiction:
Improveevaluation accuracyVSAvoidmodel generation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent creates a universal machine learning model that can evaluate multiple sensors simultaneously. By designing the model with multi-functionality to handle various sensor types and conditions through a single unified framework, the system eliminates the need to generate separate models for each sensor, thereby maintaining high evaluation accuracy while dramatically reducing the time and resources required for model generation.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent performs preliminary preprocessing of measurement data to remove sensor-specific characteristics before model training. By conducting this preprocessing action in advance, the model can be trained once on preprocessed data from multiple sensors, eliminating the need for repeated model generation for each sensor while ensuring accurate evaluation through the preliminary removal of confounding variables.

Inventive Principle:
Principle #10Preliminary action

3Adaptability or versatility

If artificial modification is applied to the generated model to make it applicable to other sensors, then model versatility is improved, but it becomes difficult to modify the model when it is hard to objectively digitize differences between sensor usage situations

Engineering Contradiction:
Improvemodel reusabilityVSAvoidmodel modification complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent replaces the manual, mechanical process of artificial model modification with an automated computational preprocessing system. By using algorithmic preprocessing operations to objectively transform and normalize sensor data, the system eliminates the need for difficult manual model adjustments, especially when sensor usage situation differences cannot be easily digitized. This substitution of automated processing for manual modification significantly reduces complexity while improving versatility.

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

Data Source

PatentEP3312695B1Information processing system, information processing method, information processing program, and recording medium
Publication Date: 2021.09.01 YOKOGAWA ELECTRIC CORP
  • EP3312695B1 patent drawingFigure 1
  • EP3312695B1 patent drawingFigure 2~3
  • EP3312695B1 patent drawingFigure 4A~4B

AI summary

An information processing system includes a state data acquisition device configured to acquire state data indicating an operation state of a plant, a first classification device configured to generate a determination line used to classify the state data, a mirror image data generation device configured to generate mirror image data obtained by mirroring the acquired state data using the generated determination line, and a second classification device configured to generate a determination model on the basis of the state data and the generated mirror image data.