Anomaly Determination Models for Unknown Device Fault Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing anomaly determination methods overlook unknown anomalies in devices, mistakenly determining them as normal due to the lack of corresponding supervised data for machine learning.

Innovation Solution

An anomaly determination device and method using a first determination model to identify predetermined anomalies and a second model to classify device states, outputting unknown anomalies when the anomaly is not predetermined, with a label setting unit adding new classes for unknown anomalies to update the models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning using failure patterns as supervised data is used, then known anomalies can be identified accurately, but unknown anomalies are overlooked and mistakenly determined as normal

Engineering Contradiction:
Improveanomaly identification accuracyVSAvoidunknown anomaly detection capability
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The anomaly determination device segments the anomaly detection task into two distinct models: a first determination model for identifying known predetermined anomalies using supervised learning, and a second determination model for detecting unknown anomalies by classifying device states. This segmentation allows each model to specialize in its respective strength, resolving the contradiction between known anomaly accuracy and unknown anomaly detection.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The second determination model acts as an intermediary between the first determination model and the final anomaly classification. It receives device state data, performs unsupervised classification, and flags states that do not match predetermined anomaly patterns as potential unknown anomalies, thereby preventing them from being mistakenly classified as normal.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Device complexity

If only supervised learning with predetermined anomaly data is used, then the determination model is simple to implement, but it cannot detect anomalies without prior failure patterns

Engineering Contradiction:
Improvedetermination model structureVSAvoidanomaly type coverage
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The anomaly determination device implements a multi-functional determination system where the first determination model handles known anomaly detection through supervised learning, while the second determination model provides universal state classification capability through unsupervised learning. This multi-functionality allows the system to adapt to both predetermined and unknown anomaly types without requiring complete reconfiguration.

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

Solution Approach 2:

The system performs preliminary classification of device states using the second determination model before final anomaly determination. This preliminary unsupervised classification prepares the data structure and identifies potential unknown anomaly patterns in advance, enabling the system to handle diverse anomaly types even before specific failure patterns are fully characterized.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4033219B1Anomaly determination device and anomaly determination method
Publication Date: 2023.10.11 NISSAN MOTOR CO LTD
  • EP4033219B1 patent drawingFigure 1
  • EP4033219B1 patent drawingFigure 2
  • EP4033219B1 patent drawingFigure 3

AI summary

An anomaly determination device and an anomaly determination method determine an anomaly of a device based on state data of the device, by using a first determination model configured to determine whether a predetermined anomaly has occurred in the device, and a second determination model configured to classify state of the device, and output the determined anomaly of the device as an unknown anomaly in a case where the anomaly of the device is not the predetermined anomaly.