针对滚动轴承微弱故障特征易被噪声和强故障成分淹没导致漏诊或误诊问题，基于互信息与信息熵构建多目标适应度函数，形成面向故障诊断的自适应变分模态分解算法(Diagnosis Oriented Adaptive Variational Mode Decomposition， DOAVMD)，有效提取信息以传达故障特征且不产生异常模态干扰；并采用NSGAII算法对多目标适应度函数搜寻最优Pareto解集；然后考虑峭度是反应冲突的有效指标，以最大峭度值为目标，筛选解集中最优结果实现DOAVMD参数的确定和特征提取；基于齿轮箱轴承内圈损伤数据验证提出方法的可靠性。结果表明：DOAVMD可剔除含噪分量并保留具有最显著冲击信号的特征，且该特征较传统VMD方法更能凸显故障特征频率。

Aiming at the problem that rolling bearing weak fault characteristics are easily submerged by noise and strong fault components leading to the missed diagnosis or misdiagnosis, based on the mutual information and information entropy, the multiobjective fitness function formed by the diagnosis oriented adaptive variational mode decomposition (DOAVMD) algorithm for fault diagnosis can effectively extract information to convey fault features without generating abnormal modal interference; and the NSGAII algorithm is used to search for the optimal Pareto solution set for the multiobjective fitness function; then, with considering the kurtosis is effective indicator to reflect the conflict, and the maximum kurtosis value is taken as the target to filter the optimal results in the solution set for the determination of DOAVMD parameters and feature extraction; the reliability of the proposed method is verified based on the gearbox bearing inner ring damage data. Results show that DOAVMD can eliminate the noisy components and retain the features with the most significant impact signals, and the features can highlight the fault feature frequency better than the traditional VMD method.

TK730.2

内蒙古电力科学研究院2022年自筹项目（510241220009）；国家自然科学基金（52006148）