关键词: [K]均值聚类, 似然概率, 涡扇发动机, 气路故障, 模式识别

Abstract:

K-means clustering algorithm is one of the most well-known methods in the field of clustering. However, k-means clustering completely relies on Euclidean distance for clustering and ignores the influence of sample feature dispersion on clustering results. As a result, samples at the clustering edge are easy to be misclustered, the algorithm is easy to fall into local convergence, and the clustering accuracy is low. Aiming at the shortcomings of the traditional k-means clustering algorithm, this paper proposes the likelihood k-means clustering algorithm. Considering the discrete degree information of sample characteristics in each dimension for all samples of each cluster, the likelihood probability of samples belonging to a certain cluster is calculated separately, which can effectively improve the clustering accuracy. In this paper, the superiority of the likelihood k-means clustering algorithm is verified in the artificial data set and benchmark data set, and then it is applied to the pattern recognition for gas-path component failure and sensor failure of a turbofan engine, where the practicability and effectiveness of the proposed algorithm for the failure diagnostics of turbofan engine are verified.

Key words: K-means clustering, likelihood probability, turbofan engine, gas path failure, pattern recognition