中文摘要：

针对单一特征在进行故障诊断时准确率不高的问题，提出了一种基于自组织神经网络（SOM）的滚动轴承状态评估方法。该方法首先从原始振动信号中提取出多特征数据，运用主成分分析（PCA）方法对多特征数据进行预处理，采用SOM进行网络训练，构建多特征数据的融合模型，输出竞争神经元层的权值矢量；然后，计算每一个样本到竞争神经元层权值矢量的最小欧氏距离，输出最终的融合指标；最后，通过比较待检测样本与正常样本的最小欧氏距离的差异来判断轴承的状态。将该方法应用于滚动轴承状态评估，试验结果表明：融合指标比单一指标对早期故障更加敏感、更加稳健；同时，融合指标能够定量地描述轴承状态的劣化过程。

英文摘要：

Aiming at the problems that single feature fault diagnosis accuracy was not too high, a rolling bearing condition assessment method was proposed based on SOM herein. Firstly, the multi- dimensional features were extracted from the original vibration signals and preprocessed by PCA, a fusion model was established by training SOM network and weight vectors of competitive neuron were obtained. Secondly, the fusion index, which was the minimum Euclidean distance between every sam- ple values to the competitive neuron weighting vector, was achieved. Finally, the conditions of rolling bearings were classified by comparing the minimum Euclidean distances among the detected samples and the normal samples. The proposed method herein was applied to condition assessment of the roll- ing bearings, and the test results show that the fusion index is more sensitive and robust than that of original single feature during the stages of early faults; meanwhile, the fusion index may reflect the states of rolling bearings more accuratelv.