Retargeting extreme learning machines for classification and their applications to fault diagnosis of aircraft engine

Abstract

Since the original extreme learning machine (ELM) generates the hidden nodes randomly, it usually needs more hidden nodes to reach the good classification performance. However, more hidden nodes will jeopardize the real time, which limits its applications to the testing time sensitive scenarios. To this end, the commonly-used methods tend to compact its structure via optimizing the number of hidden nodes. Different from this viewpoint of network structure, in this paper two algorithms are proposed to improve the real time performance of ELM from a viewpoint of data structure. Specially, they improve the ELM classification performance by retargeting its label vectors. As thus, they need fewer hidden nodes to reach the same classification performance, which means the better real time. Finally, experimental results on the benchmark data sets validate the effectiveness and feasibility of the presented two algorithms. To be more important, they are applied to the fault diagnosis of aircraft engine and can be developed as its candidate techniques.