Abnormality Detection Control with Dynamic Algorithm Switching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current algorithms for detecting abnormalities in machines using machine learning do not provide an environment for switching between different algorithms to meet varying needs such as high accuracy and short processing time, which are essential in production sites for predictive maintenance.

Innovation Solution

A control device that includes a feature extraction part, processing part, determination part, and switching part, allowing for the dynamic switching of algorithms based on predetermined conditions such as processing time, determination result, and production process, while inheriting a common learning model and parameters for feature quantities.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a single machine learning algorithm is used for abnormality detection, then the system structure is simple, but it cannot meet varying needs for high accuracy and short processing time in different production scenarios

Engineering Contradiction:
Improvealgorithm switching capabilityVSAvoidsystem structure
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The control device is designed to execute multiple types of machine learning algorithms (supervised learning, unsupervised learning, reinforcement learning) through a unified architecture. The processing part can dynamically select and switch between different algorithms based on production needs, allowing the same hardware platform to serve multiple detection scenarios with varying accuracy and speed requirements without requiring separate dedicated systems for each algorithm type.

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

2Measurement precision

If an algorithm with high processing accuracy is selected, then abnormality detection precision is improved, but processing time increases

Engineering Contradiction:
Improveabnormality detection accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system implements dynamic algorithm selection where the processing part can switch between different machine learning algorithms based on real-time production conditions. When high detection accuracy is prioritized, more computationally intensive algorithms with higher precision can be selected. When rapid response is needed, lighter algorithms with shorter processing times are chosen. This dynamic adaptability allows the system to optimize the accuracy-time tradeoff according to specific production scenarios.

Inventive Principle:
Principle #15Dynamics

3Adaptability or versatility

If multiple algorithms are prepared to meet different needs, then processing flexibility is improved, but system complexity and difficulty of operation increase

Engineering Contradiction:
Improvealgorithm selection flexibilityVSAvoidalgorithm management
Core Design Contradiction:
Adaptability or versatilityVSEase of operation

Solution Approach 1:

The control device incorporates an automatic algorithm selection mechanism where the processing part autonomously determines which machine learning algorithm to execute based on pre-set conditions and production parameters. This self-service capability eliminates the need for operators to manually configure and manage multiple algorithms, reducing operational complexity while maintaining the flexibility to switch between different algorithm types based on detection requirements.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3901721B1Control device and program
Publication Date: 2024.09.11 OMRON CORP
  • EP3901721B1 patent drawingFigure 1
  • EP3901721B1 patent drawingFigure 2
  • EP3901721B1 patent drawingFigure 3

AI summary

The purpose of the present invention is to provide an environment in which it is possible to switch an algorithm involved in an abnormality sensing process. A control device: calculates a feature quantity from a state value acquired from a monitored object; uses a learning model on the basis of the calculated feature quantity to execute one of a plurality of types of algorithms for calculating a value indicating the probability that an abnormality is occurring in the monitored object; determines, on the basis of the calculated value, whether the abnormality is occurring; and switches, in accordance with a condition defined in advance, the one algorithm that is executed.