协同深度学习的风电机组传动系统早期故障有效可靠预示方法研究

The efficient and reliable prediction method for incipient fault of wind turbine transmission system based on collaborative deep learning is proposed under the condition of varying working status, fault coupling, small samples, and so on. The research stratagem of this proposed method is as follows. A novel theory framework of collaborative deep learning, which has incipient faults decoupling, sample-adding, fault recognition, evolution trend prediction and heterogeneous data fusion together, will be proposed to form the efficient and reliable prediction method for incipient fault. The manifold decoupling method for the strong coupling and weak incipient fault feature of wind turbine transmission system is put forward. The deep transfer learning-based incipient fault sample-adding method under varying working status and incipient fault recognition method based on deep learning model fine-tuning are proposed to achieve the incipient fault recognition precisely under the condition of small sample. The deep recurrent neural network-based prediction method is proposed to fulfill evolution trend prediction of incipient fault. The reliable and efficient prediction method of the incipient fault, which is based on multi-modal deep learning for heterogeneous data, is proposed. Finally, the system prototype of incipient fault prediction based on collaborative deep learning will be developed, and system performance will be evaluated. The research on the proposed incipient fault prediction has important theoretical and economic values not only for ensuring wind turbine steady and reliable operation, but also for enrichment and development of the incipient fault prediction theory and technology.

针对在工况交变、故障耦合、样本稀少等情况下风电机组传动系统早期故障不能有效可靠预示的问题，提出协同深度学习的风电机组传动系统早期故障有效可靠预示方法。该方法研究思路为：提出集早期故障解耦、样本添加、故障识别、演变趋势预测及异构数据融合于一体的协同深度学习的早期故障有效可靠预示理论体系，形成早期故障有效可靠预示的协同深度学习新方法；提出风电机组传动系统强耦合早期故障微弱特征流形解耦方法；提出变工况下深度迁移学习的早期故障样本添加方法和深度学习模型优化的早期故障识别方法，实现少样本条件下早期故障准确识别；提出深度循环神经网络预测方法，实现早期故障演变趋势预测；提出异构数据多模态深度学习的早期故障预示方法，实现早期故障有效可靠预示；最后研发协同深度学习的早期故障预示系统原型，进行系统性能评估。该方法对保证现役风电机组的稳定可靠运行，丰富和发展机械早期故障预示理论和技术，具有重要的理论和经济价值。

风电机组传动系统早期故障特征信息微弱、耦合、时变性强、有标记的早期故障样本稀少，导致风电机组传动系统早期故障预示难度更大、要求更高。针对在工况交变、故障耦合、样本稀少、数据异构等情况下风电机组传动系统早期故障预示难题，项目按计划进行了协同深度学习的风电机组传动系统早期故障有效可靠预示方法研究，在协同深度学习框架内研究了风电机组传动系统早期故障解耦、样本添加、故障识别、演变趋势预测及异构数据融合预警等关键科学问题，提出了风电机组传动系统早期故障有效可靠预示的协同深度学习新方法，建立了协同深度学习的风电机组传动系统早期故障预示理论构架。.提出了强噪声干扰下风电机组传动系统早期故障微弱特征加权融合提取方法和强耦合早期故障的深度学习解耦方法，实现强耦合早期故障微弱特征增强与解耦；提出了多密集块中心矩差异的深度领域自适应的和多源深度领域自适应融合多风电机组信息的样本添加方法，实现变工况下早期故障样本添加；提出了少样本下深度平衡领域适应网络、多尺度动态融合原型聚类网络、半监督多关联层网络、自适应损失加权元残差网络的风电机组传动系统早期故障识别方法，实现少样本条件下早期故障准确识别；提出了元学习门限循环单元神经网络的齿轮退化趋势预测、自编码器和门限循环单元神经网络的滚动轴承退化趋势预测、自注意力卷积长短时记忆网络的滚动轴承剩余寿命预测等方法，实现早期故障演变趋势预测；提出了基于长短时记忆网络融合SCADA数据的、深度变分自编码网络融合SCADA数据的、SCADA数据时空特征深度融合的风电机组状态监测方法，实现健康状态监测预警。.研发了基于微服务架构的风电机组传动系统早期故障预示系统原型，利用模型和组件封装技术，将协同深度学习算法与故障预测知识进行深度融合及组件化封装，为实际应用奠定技术基础。该研究丰富和发展了机械早期故障预测理论和技术，提升了风电机组等重大装备安全运行的稳定性和可靠性，具有重要的理论意义和经济价值。