改进均衡分布适配的滚动轴承寿命阶段识别

Life stage identification of rolling bearing based on improved BDA method

针对不同工况下训练样本与测试样本分布差异导致滚动轴承寿命阶段无法被有效识别的问题,提出改进均衡分布适配的滚动轴承寿命阶段识别方法。采用无重复均匀随机抽样对源域类间样本进行多次均匀随机抽样,得到源域多样本训练集,以减小源域内部样本选择对目标域预测标签的影响;在再生核希尔伯特空间上利用平衡因子μ动态调节边缘分布和条件分布所占权值,并通过迭代的方式不断优化目标域伪标签以减小两域的最大均值差异;利用源域多样本数据集各自的映射矩阵构造多个分类器,经过一致性判别得到目标域样本最终识别结果。在两组滚动轴承寿命阶段数据集上进行实验验证,证明了所提方法的可行性和有效性。

In view of the problem that the distribution differences between training samples and test samples under different working conditions cannot effectively identify the life stage of rolling bearings,an improved method for identifying the life stage of rolling bearings based on balanced distribution is proposed.Firstly,non-repetitive uniform random sampling is used to conduct multiple uniform random sampling of inter-class samples in source domain,the training set of multi-sample in source domain is obtained to reduce the influence of sample selection in source domain on target domain prediction label.Furthermore,the weights of edge distribution and conditional distribution are dynamically adjusted in reproducing kernel Hilbert space by using equilibrium factorμ,the weights of edge distribution and conditional distribution are continuously optimized by iteration.In order to reduce the maximum mean difference between the two domains,pseudo-labels in the target domain are transformed into pseudo-labels.Finally,multiple classifiers are constructed by using the mapping matrices of the source domain data sets,the final recognition results of the target domain samples are obtained by consistency discrimination.Experiments on two sets of data sets of rolling bearing life stages show that the proposed method is feasible and effective.