Random Matrix-Based Approach for Data Reduction and Sensor Selection with Application to Degradation Index Construction

Togo Jean Yves Kioye, Malika Kharouf, Khac Tuan Huynh

Abstract: This paper relies on the random matrix theory to reduce data dimension and to identify useful data sources in the unsupervised context. A so-called random matrix based principal component analysis algorithm is thus developed and then applied to the well-known 2008 PHM dataset to build efficient but less costly degradation indices. A comparison of the degradation indices obtained with and without sensors selection confirms the performances of our proposed approach.