This content is not included in your SAE MOBILUS subscription, or you are not logged in.
Health Assessment of Liquid Cooling System in Aircrafts: Data Visualization, Reduction, Clustering, and Classification
- Shalabh Gupta - University of Connecticut ,
- Nayeff Najjar - Unviersity of Connecticut ,
- Chaitanya Sankavaram - Unviersity of Connecticut ,
- James Hare - Unviersity of Connecticut ,
- Krishna Pattipati - Unviersity of Connecticut ,
- Rhonda Walthall - Hamilton Sundstrand ,
- Paul DOrlando - Hamilton Sundstrand
ISSN: 1946-3855, e-ISSN: 1946-3901
Published October 22, 2012 by SAE International in United States
Citation: Najjar, N., Sankavaram, C., Hare, J., Gupta, S. et al., "Health Assessment of Liquid Cooling System in Aircrafts: Data Visualization, Reduction, Clustering, and Classification," SAE Int. J. Aerosp. 5(1):119-127, 2012, https://doi.org/10.4271/2012-01-2106.
This paper addresses the issues of data reduction, visualization, clustering and classification for fault diagnosis and prognosis of the Liquid Cooling System (LCS) in an aircraft. LCS is a cooling system that consists of a left and a right loop, where each loop is composed of a variety of components including a heat exchanger, source control units, a compressor, and a pump. The LCS data and the fault correlation analysis used in the paper are provided by Hamilton Sundstrand (HS) - A United Technologies Company (UTC). This data set includes a variety of sensor measurements for system parameters including temperatures and pressures of different components, along with liquid levels and valve positions of the pumps and controllers. A graphical user interface (GUI) is developed in Matlab that facilitates extensive plotting of the parameters versus each other, and/or time to observe the trends in the data. The parameters that reflect interesting information are selected by observing the correlations between different parameters. The data are analyzed using the wavelet transform to highlight the interclass separation and subdue the within class differences for more accurate classification. Subsequently, the number of parameters is significantly reduced using the principal component analysis (PCA). PCA-based data reduction resulted in different clusters when applied to the healthy data, and faulty data of the left and right loops. Several classification methods have been tested and their performance was evaluated. The results of this paper will be used for the purpose of fault diagnosis and prognosis in LCS. The methodology used in this paper can also be applied to other HS systems.