风电传动系统多模态监测信息学习融合及健康预警诊断

Supervisory Control and Data Acquisition (SCADA) and Condition Monitoring System (CMS) has provided massive multimodal monitoring data for the wind turbine drivetrain system. It has become a new challenging issue in the wind industry to extract and mine useful information from multi-source heterogeneous monitoring data for accurately assessing and identifying the health of the wind turbine drivetrain system and then enabling early fault warning and diagnosis. Considering respective heterogeneity and complementarity of both types of monitoring data from SCADA and CMS, this project introduces the deep learning methods into the research field of multimodal information fusion and focuses on the studies from the following aspects. The temporal-spatial modeling and fusing method for heterogeneous monitoring data from SCADA will be first studied to extract the latent correlations hidden in data. Furthermore, the deep representation learning and fusion model for different sensor data including vibration signals, current signals and acoustic emission signals from CMS, will be constructed to reveal the common features among multimodal sensor data. Finally, the decision fusion strategy will be investigated based on feature learning and mining of heterogeneous SCADA and CMS data to establish a multi-level health assessment and diagnosis model. The research results of this project will provide new techniques and methods for monitoring big data processing and health assessment and diagnosis of the wind turbine drivetrain system, and are of great significance for improving the operational stability and reliability of wind turbines.

数据采集与监控系统(SCADA)和状态监测系统(CMS)为风电传动系统提供了海量丰富的多模态监测数据。如何从这些复杂多变、多源异构的监测数据中提取和挖掘有用信息，准确评估和识别传动系统的健康状态，对早期故障进行预警诊断，已成为当前风电领域面临的挑战性问题。本项目将深度学习理论引入多模态信息融合领域，针对SCADA和CMS两类监测数据的多源、异构、互补特点，从以下层面展开研究：研究SCADA多模态数据时空关系建模与融合分析方法，提取潜在的关联特征；构建振动、电流、声发射等CMS不同传感信息的深度特征学习及融合模型，揭示多模态信息间的共有属性；研究基于SCADA和CMS两类异构数据特征学习和挖掘的集成监测策略，建立风电传动系统多级健康评估与诊断模型。本项目研究成果将为风电传动系统的大数据处理及健康评估诊断提供新方法和新技术，对于提高风电装备运行稳定性和安全可靠性具有重要意义。

传动系统是风电机组的重要动力装置，由于长时间运行在变速变载的复杂工况条件下，导致故障时有发生，健康监测与故障预警诊断对保障风电传动系统安全运行至关重要。本项目以“高维时变、多源异构、互补关联”的风电监测数据为基础，研究基于多模态信息融合的风电传动系统健康监测与故障预警诊断方法。针对SCADA运行数据的高维冗余、多变量时空关联等特点，提出了多模态降噪自编码融合网络、小波多尺度时序特征融合网络、时空注意力融合网络等多模态时空特征融合方法，实现了多变量传感器数据特征的高效学习与融合，并构建叶片结冰预测、叶片开裂故障预警、主轴承温度监测、发电机多参数融合监测等故障预警与诊断模型；针对振动与电流等CMS数据的机电耦合特性，提出机电多视图融合的深度学习网络方法，实现了振动与电流信号之间共有故障信息和互补故障信息的学习与提取及齿轮箱齿轮和轴承故障的智能诊断；针对SCADA和CMS两类数据的互补监测特点，从特征层融合角度提出多源监测信息融合分析策略与模型，从SCADA和CMS数据中挖掘时空关联特征，实现了多源互补特征的动态加权融合，提高了风电传动系统健康评估的准确性和可靠性。最后，集成所研究模型与方法，设计开发风电传动系统故障预警诊断软件系统，并结合实验数据和风电场实测数据进行了验证分析。项目研究成果可以实现机组的高精度、高可靠健康监测与数据智能化建模，提升风电监测大数据的利用价值，为保障风电装备安全可靠运行和智能化风电场运维提供技术支持.