Abnormality Detection System Using Latent Variable Models

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for identifying failure causes in machines with multiple sensors require extensive expertise and man-hours, are inefficient, and struggle to accurately detect rare events with overlapping factors, often leading to missed information in abnormality detection.

Innovation Solution

An abnormality detection system that learns latent variable models and joint probability models from sensor data, using VAE, AAE, LVAE, or ADGM to infer and generate restored data, determining normality or abnormality based on deviation, reducing the need for manual feature design and expertise.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual sensor selection and feature design methods are used, then detection accuracy can be maintained for simple systems, but the time required and expertise needed increase significantly with complex systems having multiple sensors

Engineering Contradiction:

Improvedetection accuracyVSAvoidtime required for sensor selection and feature design

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs self-learning by automatically selecting sensors and designing features through training on normal operation data. The learned model autonomously determines which sensors are most informative for detecting abnormalities, eliminating the need for manual sensor selection and feature engineering by experts.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces manual expert judgment and mechanical feature design processes with an automated machine learning system. The neural network model learns optimal sensor combinations and feature representations automatically, substituting human expertise with computational intelligence.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If manual sensor selection is performed, then detection reliability can be maintained for known failure modes, but the system becomes unable to handle rare events and unknown failures effectively

Engineering Contradiction:

Improvedetection reliabilityVSAvoidability to handle rare events and unknown failures

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system performs preliminary learning during normal operation by training on data collected when the system is functioning correctly. This preliminary action establishes a baseline model of normal behavior, which then enables the system to detect deviations indicating abnormalities, including rare events and unknown failure modes that were not explicitly programmed for.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system continuously refines its detection capabilities through feedback from actual abnormal events. When abnormalities are detected, the system can learn from these cases and update its model, improving its ability to handle rare and unknown failures over time.

Inventive Principle:

Principle #23Feedback

3Measurement precision

If custom abnormality score functions are designed for each sensor and application, then detection precision for specific applications improves, but the system complexity and development time increase

Engineering Contradiction:

Improveapplication-specific detection precisionVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent creates a universal detection system that can be applied across multiple applications and sensor types without requiring custom score functions for each. The learned model generalizes the detection capability, allowing the same system to handle different applications by simply providing new training data, thereby reducing overall system complexity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

4Ease of manufacture

If traditional threshold-based detection is used, then implementation simplicity is maintained, but the system fails to capture complex patterns and overlapping factors in rare events

Engineering Contradiction:

Improveimplementation simplicityVSAvoiddetection precision for complex patterns

Core Design Contradiction:

Ease of manufactureVSMeasurement precision

Solution Approach 1:

The system transforms the detection problem from a simple threshold-based one-dimensional approach to a multi-dimensional feature space. The neural network learns complex patterns and relationships between multiple sensor dimensions, enabling detection of subtle abnormalities that would be invisible to simple thresholds while maintaining practical implementation through automated learning.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS11921566B2Abnormality detection system, abnormality detection method, abnormality detection program, and method for generating learned model

Publication Date: 2024.03.05 PREFERRED NETWORKS INC

AI summary

A method and system that efficiently selects sensors without requiring advanced expertise or extensive experience even in a case of new machines and unknown failures. An abnormality detection system includes a storage unit for storing a latent variable model and a joint probability model, an acquisition unit for acquiring sensor data that is output by a sensor, a measurement unit for measuring the probability of the sensor data acquired by the acquisition unit based on the latent variable model and the joint probability model stored by the storage unit, a determination unit for determining whether the sensor data is normal or abnormal based on the probability of the sensor data measured by the measurement unit, and a learning unit for learning the latent variable model and the joint probability model based on the sensor data output by the sensor.