基于深度置信网络诊断模型构建及在风电齿轮传动系统状态识别的应用研究

Wind turbine gearbox is composed of planetary gear and ordinary gear. Due to frequently operate under extreme loading conditions and state information coupling serious, the technology for traditional gearbox fault diagnosis fail to solve the difficulty of complex input and complex classification. For deep learning algorithm can effectively deal with complex function and complex classification problems, wind turbine gearbox condition recognition based on deep learning is proposed in this project. The deep belief networks(DBN) diagnosis technology for wind turbine gearbox is builded. The specific contents are as follow: Puts forward the optimal design of new DBN model of wind turbine gearbox by manifold learning and glowworm swarm optimization; Signal decoupling technology based on DBN; In order to eliminate varying operating conditions interfering, this project construct compound DBN model by order analysis and denoising technology. DBN diagnosis technology sharing of wind turbine gearbox is realized by parameter-transfer of transfer learning. Finally, condition recognition, fault source location, and information sharing technology of wind turbine gearbox are realized, which provide the new methods for fault diagnosis and condition monitoring of complex mechanical system as frequently operate under extreme loading conditions and state information coupling serious.

风电复合式传动系统由行星齿轮系+定轴齿轮系组成，系统结构复杂，状态信息耦合严重，且长期处于恶劣工况，传统智能诊断方法难解复杂输入及复杂分类问题。针对该问题，本项目利用深度学习算法能有效处理复杂函数表示及复杂分类问题的优点，通过对深度置信网络算法进行改进，建立风电齿轮传动系统的深度置信网络诊断技术。具体内容包括：利用流形学习及萤火虫寻优方法，提出基于流形学习的深度置信网络诊断模型及参数优化方法，确定与风电齿轮传动系状态样本相匹配的优化模型；利用互相关信息构建基于深度置信网络的信号解耦技术；结合阶次分析及降噪技术，提出基于深度置信网络的复合诊断模型，消除交变工况对识别效果的干扰；融合模型参数的迁移学习方法，构建集成参数迁移学习与深度置信网络的诊断信息共享技术。最终实现风电齿轮传动系统状态判定、故障定位、诊断信息共享等功能，为恶劣环境条件下服役的复杂机械系统故障诊断及状态监测提供新的手段和方法。