多源信息融合的风电机组（集群）状态异常检测及健康评估方法研究

Aiming at the health maintenance of the wind turbine generator system, through the fusion of vibration - oil - acoustic emission and other multi - source information, we try to explore one condition anomaly detection & assessment method through process of machine - key components – machine – cluster. By studying the simultaneous registration method of multi-source heterogeneous sensing information, the incomplete problem of source data and feature information will be solved. Through the study of acquisition and characterization of multi-dimensional simultaneous feature information method based on brain-like learning, to achieve the whole machine degradation trend tracking and state anomaly detection. Secondly, based on the multi-sensor information of both anomalous and steady state, the multi-level spatial diagnostic model featured by oil-vibration-acoustic emission is constructed by using the overlapping and complementary characteristics of heterogeneous multi-source information to realize the precise fault location of the key component. Basing on the above results, this paper studies the merits and demerits assessment of fault occurred under machine’s running state, along with visualization and comprehensive health evaluation index construction method; obtains the fuzzy mapping relationship between the multi-source information fusion feature and the health state, and constructs the machine failure prediction model. Through correlation, statistics and fuzzy analysis, this paper will explore the internal relationship between cluster of wind turbine and single wind turbine fault, to achieve the health maintenance of wind turbine (cluster). Thus, the theoretical and technical basis for improving the efficiency of operation, reducing costs and achieving scientific maintenance will be established.

以风力发电机组的健康维护为目标，项目拟通过融合振动-油液-声发射等多源信息，探索形成一种从整机->关键部件->整机->风机集群的服役状态异常检测及健康评估方法。通过研究多源异类传感信息的同步配准方法，解决源数据和特征信息不完备问题；研究基于类脑学习的多维同步特征信息获取、表征方法，实现整机劣化趋势跟踪与状态异常检测；其次，基于异常及稳态时刻的多传感信息，研究利用异类多源信息的交叠互补特性，构建振动-油液-声发射特征空间多层次诊断模型，实现关键部件故障精准定位。在此基础上，研究整机运行状态故障特征的优劣评价、可视化表征及综合评价指标构造方法，获得多源信息融合特征与健康状态的模糊映射关系，进一步基于相关、统计、模糊分析，挖掘风机集群与单机故障之间的内在关联关系，研究构建故障预示自适应模型，实现风机（集群）服役状态的健康评估。项目成果将为提高风机运行效率、降低成本及实现科学维修奠定理论技术基础。

风电机组传动系统状态异常检测及故障溯源仍然是当前风机（集群）运维领域面临的关键问题。本项目针对风机（集群）运维过程中涉及的异类多源信息融合方法展开研究，提出了风机传动系统多源异构数据的智能采集与异常检测方法，构建了集合振动-电流-油液-声发射特征空间的多源异构数据融合模型，形成了一整套适用于风机传动系统和风机集群的故障预示自适应评估模型。主要成果如下：.（1）提出了风机传动系统多源异构数据的智能采集与异常检测方法，探索了基于时频融合的振动信号异常值消除与质量保证策略，消除了风机传动系统振动信号中局部极大值的影响，进一步通过稀疏注意力深度自编码器技术，实现了风机传动系统关键部件运行状态异常检测。.（2）提出了基于深度模型的风机传动系统可靠性评估及故障识别模型，实现了复杂工况条件下的特征域适配及故障诊断，进一步研究了基于循环神经网络的滚动轴承寿命预测建模，实现了复杂工况条件下风机传动系统故障的分析和决策，为风机健康状态预示模型的构建奠定了基础。.（3）针对风机传动系统关键部件服役过程中获取的振动-电流-油液-声发射等多源异类信号，研究了单源多通道数据的特征融合及数据特征参数筛选方法，提出了基于不同类型数据的差异性融合方法，实现了多源异构数据下故障信号的多尺度精确表征和科学定量评估。.（4）提出了基于状态转移注意力深度空间模型的寿命预测方法，实现了多源域适配下风机传动系统轴承退化状态在线迁移识别，并集成和设计了风机现场实时数据采集与智能处理系统，以期实现风机传动系统实时监测及故障定性和定量识别。.上述研究成果解决了风机传动系统在服役过程中的信息冗余和存储传输低效的局限，为融合风机传动系统振动-电流-油液-声发射等异类多源信息提供了理论基础和技术支持，对于提高风机运行效率、降低成本及实现科学维修，具有重要的科学意义和工程应用价值。