Robot ML Model Permission Control and Operation Traceability

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing technologies do not adequately manage machine learning model training data, risking unauthorized copying and use, which can lead to loss of business opportunities.

Innovation Solution

A machine learning model operation management system comprising a model building server, robot controller, management server, and operation server, where the robot controller inquires permission to use the trained model and operation information is stored without sensitive data, ensuring secure management and reducing construction and running costs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If training data and learned model data are stored and managed without permission control, then the system is simple and easy to operate, but security is compromised and unauthorized copying/use can occur

Engineering Contradiction:
ImprovesecurityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system divides the management function into separate components: the robot controller stores and manages trained models locally, while the operation server handles operation information. This segmentation allows security controls to be applied specifically to model access without complicating the entire system.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The robot controller performs preliminary permission verification by inquiring the operation server before executing a trained model. This preliminary action prevents unauthorized use before it occurs, ensuring security without requiring complex real-time monitoring throughout the system.

Inventive Principle:
Principle #10Preliminary action

2Loss of information

If operation information includes detailed sensitive data, then complete operation records are maintained, but construction and running costs increase due to security requirements

Engineering Contradiction:
Improveoperation information completenessVSAvoidconstruction cost
Core Design Contradiction:
Loss of informationVSEase of manufacture

Solution Approach 1:

The system extracts and removes sensitive information from operation information before storing it on the operation server. Only non-sensitive operation data is transmitted and stored, maintaining operational monitoring capabilities while eliminating the need for expensive security infrastructure to protect sensitive data.

Inventive Principle:
Principle #2Taking out (Extraction)

3Reliability

If permission verification is not implemented, then the robot controller can freely use trained models, but unauthorized copying and use occurs leading to loss of business opportunities

Engineering Contradiction:
Improvemodel usage securityVSAvoidoperation convenience
Core Design Contradiction:
ReliabilityVSEase of operation

Solution Approach 1:

The robot controller automatically inquires the operation server for permission before using a trained model and receives an immediate response. This feedback mechanism ensures security through permission verification while maintaining ease of operation, as the process is automated and requires no manual intervention from the user.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3932626B1Machine-learning model operation management system and machine-learning model operation management method
Publication Date: 2026.03.25 KAWASAKI JUKOGYO KK
  • EP3932626B1 patent drawingFigure 1
  • EP3932626B1 patent drawingFigure 2
  • EP3932626B1 patent drawingFigure 3

AI summary

A machine learning model operation management system includes a model building server and an operation server. The model building server builds a trained machine learning model based on received training data. When the trained machine learning model stored in a robot controller operates for determining the operation of a robot, the operation server receives operation information generated by the robot controller. Data of the trained machine learning model built by the model building server is assigned with model identification information that uniquely identifies the trained machine learning model. The robot controller makes an external inquiry as to whether or not it has permission to use the trained machine learning model stored in itself, and if it has the permission to use, it makes the trained machine learning model available for use. The operation server stores the operation information in association with the model identification information.