Operation State Classification Using PCA for Failure Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing techniques for failure detection in devices like elevators and air conditioners face challenges in accurately distinguishing between abnormal device conditions and changes in operation states, leading to inefficient failure detection due to excessive classifications and reliance on additional analysis.

Innovation Solution

An operation state classification apparatus that utilizes principal component analysis to generate classifications based on probability density distributions, incorporating physical quantities for failure determination and supervised machine learning to effectively group sensor data by operation states.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the number of sensor items used for classification is increased, then the accuracy of operation state classification is improved, but the number of classifications becomes excessive and the same operation state does not occur consistently

Engineering Contradiction:
Improveclassification accuracyVSAvoidnumber of classifications
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent transforms the classification problem from using multiple sensor items directly to using principal components as new parameters. By changing the parameter space from raw sensor readings to derived principal components, the system achieves effective classification with fewer parameters, resolving the contradiction between classification accuracy and system complexity.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent combines multiple sensor items into composite principal components through mathematical transformation. These composite parameters integrate information from multiple sensors while reducing redundancy, enabling accurate operation state classification without requiring excessive individual sensor items or classification categories.

Inventive Principle:
Principle #40Composite materials

2Device complexity

If principal component analysis is used for dimension reduction, then the number of classifications is reduced, but detection accuracy of failures cannot be improved

Engineering Contradiction:
Improvenumber of classificationsVSAvoidfailure detection accuracy
Core Design Contradiction:
Device complexityVSReliability

Solution Approach 1:

The patent introduces feedback mechanisms where classification results are used to adjust and refine the principal component analysis model. By continuously learning from classified data and updating the model parameters, the system maintains high failure detection accuracy even with reduced classification categories, resolving the contradiction between simplicity and reliability.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3373089B1Operating state classification device
Publication Date: 2021.03.10 MITSUBISHI ELECTRIC CORP
  • EP3373089B1 patent drawingFigure 1
  • EP3373089B1 patent drawingFigure 2
  • EP3373089B1 patent drawingFigure 3

AI summary

A principal component calculating unit (13) for calculating a principal component of a plurality of sensor data collected from a device, and a physical quantity for failure determination calculating unit (11) for calculating a physical quantity to be used for failure determination on the basis of the plurality of sensor data are provided. Further, an operation state classification unit (14) for generating classification, for each operation state, of the sensor data collected from the device, using the principal component calculated by the principal component calculating unit (13) and the physical quantity calculated by the physical quantity for failure determination calculating unit (11) is provided.