Sensor Fault Detection Using Partial Qualitative Knowledge
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor fault detection methods in complex systems require comprehensive system models, accurate hardware redundancy, or extensive data, which can be costly and impractical, especially in partially observable, time-varying environments like marine vessels, leading to potential false positives or negatives.
Innovation Solution
A method that identifies principles relating sensor outputs through system equations, deduces reliability, and assigns trustworthiness levels without a comprehensive system model, using partial qualitative and quantitative knowledge, and custom representation schemes including causality, correlation, and inequalities, allowing for automatic detection and classification of sensor faults.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If hardware redundancy or analytical redundancy with comprehensive system models is used, then sensor fault detection reliability is improved, but system complexity and cost increase
Solution Approach 1:
The patent extracts and utilizes only the essential local physical principles and relationships relevant to each sensor subsystem, rather than requiring a complete comprehensive system model. This selective extraction of necessary physical relationships reduces model complexity while maintaining adequate fault detection capability.
Solution Approach 2:
The system divides the complex overall system into smaller subsystems, each governed by its own local physical principles. This segmentation allows fault detection to be performed using simplified local models rather than a single complex global model, reducing overall system complexity.
2Measurement precision
If comprehensive system models are used, then measurement precision is improved, but loss of time and computational resources increase
Solution Approach 1:
The patent segments the computational task by dividing it into multiple independent or loosely coupled sub-tasks, each handling a specific sensor or subsystem. This allows parallel processing and reduces the computational burden on any single processing unit, thereby reducing overall computational time while maintaining detection accuracy.
Solution Approach 2:
The system performs partial model evaluations only for the specific subsystems or sensors being monitored, rather than computing complete system models. This partial action approach reduces computational time while maintaining sufficient precision for the specific fault detection task at hand.
3Reliability
If hardware redundancy is implemented, then fault detection capability is improved, but device complexity and cost increase
Solution Approach 1:
The patent creates virtual copies of sensors through mathematical modeling and signal processing techniques. These virtual sensors are generated computationally from existing sensor data and physical relationships, providing redundant measurement capabilities without requiring physical hardware duplication.
Solution Approach 2:
The patent replaces mechanical hardware redundancy with computational and mathematical approaches. Instead of adding physical sensors, the system uses algorithms, physical models, and data processing to create virtual redundancy, thereby maintaining fault detection capability while reducing hardware complexity.
Data Source
Figure 1
Figure 2
Figure 3a
AI summary
The present disclosure provides methods and systems to automatically detect, classify and/or mitigate sensor errors using partial qualitative and quantitative knowledge of the subsystems. In various examples, sensor fault detection is performed with a custom designed representation scheme covering causality, correlation, system of equality and inequalities, and an associated logic. The logic is described by a set of algorithmic steps to iteratively assign trustworthiness level of sensors. Sensor fault classification is performed by combining mathematical and statistical techniques that can be utilized to expose bias, drift, multiplicative calibration error, precision degradation and spike error. Sensor fault mitigation is also performed on identified bias, drift, multiplicative calibration error, precision degradation and spike error.