Learning local discriminative representations via extreme learning machine for machine fault diagnosis

摘要

Recently, learning data representations have been investigated to reduce the dependences of human intervention and improve the performance of machine fault diagnosis. However, most of the representation learning methods are computationally intensive due to complex training procedures. Extreme learning machine is well-known for its fast training speed and strong generalization ability. It also has been applied to learn data representations for clustering and classification tasks. In this paper, a local discriminant preserving extreme learning machine autoencoder (LDELM-AE) is proposed to learn data representations with the local geometry and local discriminant exploited from the input data. Specifically, LDELM-AE utilizes two graphs to enhance the within-class compactness and between-class separability, respectively. Furthermore, the hierarchical representations can be obtained by stacking several LDELM-AEs. On several benchmark datasets, the proposed method demonstrates better classification accuracies than the state-of-the-art methods. Moreover, the proposed method has been used to diagnostic the rotary machine faults and achieves the diagnostic accuracy of 99.96%, which proves the proposed method is an efficient tool to diagnose machine faults. (C) 2020 Elsevier B.V. All rights reserved.