Predictive Fault Diagnosis for Physical Systems With Sensor Fault Isolation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Complex physical systems pose challenges for supervision systems due to unknown mathematical equations, making it difficult to implement analytical redundancy for fault detection and isolation, especially when sensors can also be faulty, leading to increased costs and complexity.

Innovation Solution

A method using predictive models trained with sensor measurements and a signature matrix to generate fault indicators, where minimally overdetermined subgraphs and Dulmage-Mendelsohn decomposition are employed to determine the signature matrix, allowing for diagnosis independent of sensor faults.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If hardware redundancy (multiple sensors) is used to detect sensor failures, then reliability of fault detection is improved, but device complexity and cost increase significantly

Engineering Contradiction:
Improvefault detection reliabilityVSAvoidnumber of sensors
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent creates virtual copies of sensor measurements through predictive models. Instead of physically duplicating sensors, the system uses machine learning models to generate predicted measurements that serve as virtual sensor copies, enabling fault detection without additional hardware

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent replaces the mechanical/hardware redundancy approach with an information-processing approach using predictive models. The system substitutes physical sensor duplication with computational prediction, replacing mechanical systems with intelligent algorithms

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Device complexity

If analytical redundancy based on mathematical equations is used for fault detection, then device complexity is reduced, but reliability decreases when equations are unknown or inaccurate

Engineering Contradiction:
Improvesystem complexityVSAvoidfault detection accuracy
Core Design Contradiction:
Device complexityVSReliability

Solution Approach 1:

The patent changes the parameters used for fault detection from fixed mathematical equations to adaptive predictive models. The system transitions from static equation-based methods to dynamic machine learning models that can adapt to system variations and unknown relationships

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent performs preliminary training of predictive models using normal operating data before actual fault detection. This preliminary action allows the models to learn system behavior patterns in advance, improving their reliability when detecting actual faults

Inventive Principle:
Principle #10Preliminary action

3Reliability

If multiple sensors are deployed for hardware redundancy, then fault detection capability is improved, but loss of time for system installation and maintenance increases

Engineering Contradiction:
Improvesensor failure detectionVSAvoidinstallation and maintenance time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent uses virtual copying through predictive models to eliminate the need for physical sensor duplication, thereby avoiding the installation and maintenance time associated with additional hardware sensors

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4400929A1Diagnosis of a physical system by predicting fault indicators by machine learning
Publication Date: 2024.07.17 ATOS FRANCE
  • EP4400929A1 patent drawingFigure 1~2
  • EP4400929A1 patent drawingFigure 3~4
  • EP4400929A1 patent drawing

AI summary

The invention relates to a diagnostic system (20) for a physical system (10) comprising means for acquiring measurements (m1, m2, ..., ms) provided by sensors (11, 12, ..., 1S), for providing said measurements as input to a set (22) of predictive models (M1, M2, M3, ..., MN) each generating a fault indicator (r1, r2, r3, ..., rN), and then for determining a diagnosis of said physical system from said fault indicators and a signature matrix (24) associating diagnoses and fault indicators (r1, r2, r3, ..., rN).