An improved K-means algorithm for reciprocating compressor fault diagnosis

In this paper, an improved K-means clustering algorithm is proposed for reciprocating compressor fault diagnosis. Our algorithm makes improvements on the selection of initial cluster centers and the updating of centers, respectively. With respect to the characteristics of manifold distribution of fault data, cosine distance is used to calculate average similarity of each fault data. Based on the average similarity, P groups of initial cluster centers can be obtained and the average similarity of each initial center for each group is quite different. Moreover, the energy function is introduced to calculate and update cluster centers. Experimental results on a real reciprocating compressor fault dataset show that the proposed improved K-means algorithm has a high clustering accuracy and a fast convergence speed. Moreover, experimental results on the real reciprocating compressor fault dataset with noise demonstrate that the proposed algorithm achieves good performance in anti-noise.