<![CDATA[This paper proposes a novel idea of a fault detection algorithm based on the Regression Tree (RT) algorithm from the decision tree learning and the distance correlation, which is the nonlinear version of Pearson’s correlation, to reduce the number of sensors without significantly decreasing the model predictive accuracy and the fault diagnosis capability. A numerical validation on an experimental dataset provided by the Los Alamos National Laboratory (LANL) with MATLAB software shows that the proposed algorithm has a comparable model predictive accuracy to the classical RT while requiring a smaller number of sensors (5 instead of 24) and more robust in detecting faults with false negative and positive rates < 15%. Furthermore, we demonstrate that our proposed algorithm runs about 4 times faster than the classical RT on an experimental dataset with 4096 samples on an 8-core, 16 GB RAM machine. In a real-life setup, the proposed algorithm can be used to provide a sensor installment plan on a structure. Such that, the user can still monitor the presence of a fault inside a building precisely, but with a cheaper maintenance cost.]]>
Copyrights © 2023
