Sensor Failure Detection via Mahalanobis Space Normalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional failure detection methods for sensors in gas turbine engines struggle to accurately detect small degree sensor drift failures due to environmental changes, which can mask these failures within normal data ranges.
Innovation Solution
A failure detection device that includes modules for inputting and processing sensor data, generating unit and signal spaces using Mahalanobis-Taguchi systems, and comparing distances to threshold values, with features such as correlation equation identification and data nondimensionalization to isolate and adjust sensor values affected by environmental conditions, enhancing detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional failure detection methods (MT system, error variance method, neural network) are used to detect sensor failures, then the detection system can identify obvious failures, but small degree sensor drift failures are difficult to detect because they are buried in the range of normal values affected by environmental changes
Solution Approach 1:
The patent extracts and separates the environmental change factors (flight condition, outside air temperature) from the sensor data analysis process. By identifying which sensors are affected by environmental changes and extracting this influence, the system can isolate the true sensor drift from normal environmental variations, enabling detection of small degree failures that would otherwise be buried in normal value ranges.
Solution Approach 2:
The patent changes the parameters used in failure detection by introducing nondimensionalization of sensor values based on environmental parameters. Instead of using raw sensor values directly in Mahalanobis distance calculation, the system transforms the data by removing environmental influence, thereby changing the detection space to one where small drift failures become distinguishable from normal variations.
2Measurement precision
If sensor values are used as they are in Mahalanobis distance calculation without adjusting for environmental factors, then the calculation is simple, but the detection accuracy decreases because environmental variations increase data variance and mask small drift failures
Solution Approach 1:
The patent applies preliminary action by pre-identifying which sensors are affected by environmental changes and pre-calculating the nondimensionalization factors before performing failure detection. This preliminary processing of sensor data removes environmental influences in advance, so that when Mahalanobis distance calculation is performed, the data is already optimized for drift detection without requiring complex real-time adjustments during the detection process.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Failure detection accuracy is improved by providing: an accumulated data storage unit 30; a unit space generating module 13 which extracts, from the accumulated data storage unit, sensor values for a unit space to be used by an MT system, and which generates the unit space using condition sensor values which influence other sensor values, sensor values which are not influenced by the condition sensor values, and other sensor values which have either been nondimensionalized by module of the condition sensor values or have been adjusted by module of the condition sensor values; a signal space generating module 14 which, when a plurality of sensor values to be detected are input, uses without modification the condition sensor values and the sensor values that are not influenced by the condition sensor values, and uses the other sensor values, which have either been nondimensionalized by module of the condition sensor values or have been adjusted by module of the condition sensor values, to form a signal space; and an assessing module 16 which compares a distance, representing the relationship between the generated unit space and the generated signal space, with a certain threshold, to assess whether or not there is a possibility that any of the sensors have failed.