Sensor Fault Detection via Reconstructed Data Deviation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Industrial turbines with numerous sensors in harsh environments often experience undetected sensor faults, leading to distorted system status evaluations due to the complexity and high-dimensional data generated, which existing validation methods fail to address effectively.
Innovation Solution
An automatic fault detection method that calculates deviations in sensor data from an investigated sensor compared to reconstructed data from the entire sensor group, using a multiplication matrix to signal faults outside a trusted range, allowing for online and offline validation and dynamic time interval adaptation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of sensors in industrial turbines is increased to improve measurement coverage, then measurement precision is improved, but device complexity increases and sensor faults become more frequent and difficult to detect
Solution Approach 1:
The patent implements a feedback mechanism where sensor data is continuously validated against a reconstructed model of the system. The validation system compares actual sensor readings with expected values derived from physical relationships between sensors, creating a closed-loop feedback system that automatically detects deviations and signals potential faults without requiring manual intervention.
Solution Approach 2:
The patent introduces an intermediate validation layer that acts as a mediator between raw sensor data and system status evaluation. This validation system uses a multiplication matrix and reconstruction algorithms to create an intermediate representation of expected sensor behavior, allowing faults to be detected before they propagate to the final system status assessment.
2Reliability
If rule-based expert systems are used to validate sensor signals, then some sensor faults can be detected, but the validation quality is limited by vague expert knowledge providing only low-dimensional relationships
Solution Approach 1:
The patent transitions from low-dimensional rule-based validation to high-dimensional mathematical validation using a multiplication matrix that captures complex relationships between all sensors simultaneously. This dimensional expansion allows the system to validate sensor data based on comprehensive physical relationships rather than limited expert rules, preserving more information from the high-dimensional sensor data.
Solution Approach 2:
The patent replaces the mechanical expert system approach (rule-based validation) with a mathematical model-based approach. Instead of relying on human expert knowledge encoded in rules, the system uses mathematical relationships (multiplication matrix, reconstruction algorithms) to automatically validate sensor data, eliminating the information loss inherent in simplifying complex physical relationships into rules.
3Measurement precision
If sensor data from hundreds of sensors is processed to provide comprehensive system status evaluation, then measurement precision is improved, but the ability to detect true sensor faults is reduced due to data distortion
Solution Approach 1:
The patent applies preliminary validation action to sensor data before it is used for system status evaluation. By validating each sensor reading against the reconstructed model in advance, the system identifies and flags potential faults before they can distort the overall system status assessment, ensuring that only validated data contributes to the final evaluation.
Solution Approach 2:
The patent segments the validation process into independent steps: data reception, reconstruction calculation, deviation computation, and fault signaling. This segmentation allows the system to process hundreds of sensor readings systematically, validating each one against the physical relationships without overwhelming computational complexity, thereby maintaining both comprehensive coverage and accurate fault detection.
Data Source
AI summary
A method and machine for validating an investigated sensor within a sensor group includes receiving sensor data from sensors of the sensor group measuring the same physical property, calculating a deviation of the sensor data received from the investigated sensor within the sensor group from sensor data reconstructed based on the sensor data received from all other sensors of the sensor group, and signaling a sensor fault of the investigated sensor if the calculated deviation is outside of a trusted range.