Diagnosis Model Relearning With Similarity-Based Feedback Correction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing diagnosis systems fail to correct incorrectly determined normal data and do not detect abnormalities similar to past ones, limiting the accuracy of equipment diagnosis.

Innovation Solution

A learning system that includes a learner to generate a diagnosis model, a diagnoser for diagnosis, a feedback processor for correcting incorrect results, and a similarity determiner to identify similar data for relearning, enhancing accuracy through user feedback and data correction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If feedback is only provided on abnormal diagnosis results, then the feedback process is simple, but incorrectly determined normal data cannot be corrected

Engineering Contradiction:

Improvefeedback process simplicityVSAvoiddiagnosis accuracy

Core Design Contradiction:

Ease of operationVSReliability

Solution Approach 1:

The system implements a feedback mechanism where users can provide corrections for both abnormal and normal diagnosis results. The feedback processor receives user inputs indicating whether diagnosis results are correct or incorrect, and uses this feedback to improve future diagnoses. This resolves the contradiction by enabling comprehensive feedback collection while maintaining operational simplicity through a standardized feedback interface.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

Instead of only providing feedback on abnormal results as in conventional systems, this invention inverts the approach by also providing feedback opportunities for normal results. The system outputs feedback for both abnormal and normal diagnosis results, allowing users to correct incorrectly determined normal data while maintaining the simple feedback input process.

Inventive Principle:

Principle #13The other way round (Inversion)

2Device complexity

If the system only detects abnormalities that completely match past results, then the diagnosis model is simple, but it cannot detect similar abnormalities

Engineering Contradiction:

Improvediagnosis model complexityVSAvoidabnormality detection accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The system performs preliminary actions by determining similarity between current diagnosis results and past results before final classification. The similarity determiner calculates similarity metrics between new diagnosis data and historical data, allowing the system to prepare for detecting similar abnormalities in advance. This enables the detection of similar but not identical abnormalities while maintaining reasonable model complexity.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system changes parameters by introducing similarity determination as an additional diagnostic parameter. Instead of relying solely on exact matching of abnormalities, the system evaluates similarity degrees between current and past diagnosis results. This parameter change allows the model to detect similar abnormalities with varying degrees of match, improving detection accuracy without excessively increasing model complexity.

Inventive Principle:

Principle #35Parameter changes

3Productivity

If relearning is performed using only incorrect diagnosis data, then the feedback process is efficient, but similar abnormalities are not addressed

Engineering Contradiction:

Improverelearning efficiencyVSAvoiddiagnosis model robustness

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The system performs preliminary similarity determination before relearning to identify data points that are similar to incorrect diagnosis results. The similarity determiner pre-processes the data by calculating similarity metrics, preparing relevant training data in advance. This allows relearning to focus on both the incorrect data and similar cases, improving model robustness while maintaining relearning efficiency through targeted data selection.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12475408B2Learning system, learning method, and recording medium

Publication Date: 2025.11.18 MITSUBISHI ELECTRIC CORP

AI summary

A learning system (1) generates a diagnosis model through machine learning using learning data, and diagnoses diagnosis target data based on the diagnosis model. A feedback processor (115) outputs incorrect diagnosis data when acquiring a user input indicating that a diagnosis result is incorrect, and corrects the incorrect diagnosis data based on the user input. A similarity determiner (113) determines similarity, to the incorrect diagnosis data output by the feedback processor (115) before being corrected, of the learning data or another diagnosis result of the diagnosis results yet to undergo the feedback process. The feedback processor (115) performs the feedback process on similarity data with at least a specific level of similarity. The learner (111) performs relearning using data including the incorrect diagnosis data corrected in the feedback process.