不完备信息下风电转盘轴承早期多源故障声发射特征提取研究

In order to solve key issues existed in the feature extraction of Incipient multi-source fault acoustic emission signals for wind turbine slewing bearings, such as incomplete information, decoupling, etc., a novel feature extraction method with incomplete information, which includes weak feature extraction, information completing, feature reduction, decoupling, will be proposed. Based on the analysis of the influencing factors in the aspects of acoustic emission signal generation, propagation and acquisition of acoustic emission signals, the law of energy attenuation in the transmission path is studied, in order to obtain the sensor arrangement and the optimization scheme. With the analysis of the signal noise characteristics of incipient fault acoustic emission signals, the method of noise suppression in the process of feature space transformation is studied, in order to enhance and extract of weak features in strong noise background. A new multi-path loss transfer mapping model with the evaluation of the non discernible relation is established, in order to obtain the incomplete evaluation index in the measured characteristic information. And then the mutual evaluation index of feature information is established. By weakening the indiscernibility reflexology, symmetry and transitivity of Incomplete information, A rough set model with asymmetric similarity relation is established, In order to realize multi channel fuzzy feature self-repairing. By the best estimation of intrinsic dimension, and the establishment of intrinsic dimension estimation method of performance assessment model, it achieves reduction of high dimensional signal feature dimension. Then the spatial coordinate transformation relation between high dimensional coupling feature and low dimensional decoupling feature is revealed, In order to realize the decoupling of different types of early fault features. The research will help to provide strong theoretical basis and technical support for on-line monitoring and intelligent diagnosis of early multi-source fault of wind turbine slewing bearings with high frequency acoustics.

项目针对风电转盘轴承早期多源故障声发射信号特征信息不完备、耦合等难题，在不完备信息下实现早期故障微弱特征提取、特征完备化、特征约简、流形特征解耦。从声发射信号生成、传播、采集等环节影响因素分析入手，研究获取传感器布置以及优化方案；分析早期多源故障声发射信号的信噪特性，研究特征空间转换过程噪声抑制方法，实现强噪声背景下微弱特征的增强提取；通过探寻可评价不可分辨关系的复杂多路径损耗传递映射模型，在实测特征信息指导下反演不完备度评价指标，弱化不完备信息不可分辨关系的自反性、对称性和传递性，实现多通道模糊特征的自主修复；研究内蕴维数的最佳估计方法，并建立内蕴维数估计方法的性能评价模型，实现高维信号特征维数的有效约简；揭示高维耦合特征和低维解耦特征的空间坐标转换关系，最终实现对早期多源故障特征的解耦。将有助于从高频声学角度为风电转盘轴承早期多源故障的在线监测与智能诊断提供有力的理论依据和技术支持。

项目针对风电转盘轴承早期多源故障AE信号特征提取中的关键科学问题，从不完备信息下早期故障微弱信号增强、敏感特征集建立、故障辨识与趋势预测等方面进行了系统的研究。（1）从采集等环节影响因素分析入手，制定详细的转盘轴承状态监测试验方案。研究非线性微弱特征信号的降噪方法，充分考虑风电转盘轴承早期故障AE信号所含冲击成分的特性，提出利用小波Shannon熵作为目标函数，实现早期故障微弱AE信号自适应增强；（2）构成高维故障特征集，对运行状态进行全面、综合地描述，并剔除特征集中的一些干扰特征和噪音特征来选取敏感故障特征，可更加全面综合地表征早期故障。（3）建立多频带多尺度样本熵特征向量构建方法，以及性能评价模型，实现高维信号特征维数的有效约简，完成了风电转盘轴承早期故障的故障辨识模型的建立，通过参数优化提升了故障诊断的识别率以及稳定性。项目为风电转盘轴承早期多源故障的在线监测与智能诊断提供有力的技术支持。