一种基于鲁棒FCM算法的故障诊断方法

Fault diagnosis based on robust FCM algorithm

为提高传统FCM算法应用于故障诊断时的容噪能力,通过修正样本点权值,引出了一种替代欧氏距离范数的新距离函数,采用高斯核函数证明了该函数是一种距离尺度,分析了距离范数对噪声数据的表现,建立了基于该距离尺度的鲁棒FCM聚类算法并给出了算法步骤,IRIS数据和样本数日差异数据实验证明了提出的算法较传统算法具有更好的鲁棒性。最后将此算法用于机载武器系统的故障识别实验,结果表明,本文给出的方法较传统FCM方法在故障诊断中能有效克服数据特征影响,具有更高的诊断精度。

A method for improving the ability of noise tolerance of FCM algorithm during fault diagnosis process was proposed. A new metric distance function was introduced to replace the Euclidean norm in fuzzy c-means clustering procedures, and was proved by Gaussian kernel function. Performances of the two norms were compared under noise background. A robust FCM algorithm based on the new metric distance function was presented and then the approach of the algorithm was given. Numerical results show that the proposed algorithm has higher robustness than the FCM algorithm. Finally, a fault diagnosis example of certain airborne weapon system demonstrates the effectiveness of the algorithm. Experiment results show that the proposed algorithm is effective in dealing with the above problem and has higher precision.