多重不确定因素下船舶舵机电液伺服系统鲁棒故障检测、定位及自学习方法研究

Due to the complexity and harsh working conditions, a ship rudder electro-hydraulic servo system (RESS) is often influenced by some uncertain factors (such as uncertain parameters, nonlinearities and unknown and time-varying wind/wave disturbances). These uncertainties complicate the design of an effective fault diagnosis system. Thus, this project studies the fault diagnosis strategy for a ship RESS with multiple uncertainties and to build up an online automatic fault diagnosis system, which is also the requirements of ship safe operation. According to the characteristics of different uncertain factors, the disturbance decoupling and suppression strategy is studied, and then an integrated model-based and data-driven robust fault diagnosis method is proposed for the RESS. In actual applications, not only the occurrence of a fault, but also the type and localization of the fault are wanted. In this project, a batch of specific observers is designed and work asynchronously and cooperatively to isolate different faults in the uncertain RESS. Commonly encountered faults including actuator, component and sensor faults in the RESS are all considered. Considering the development and expansion requirements of the fault diagnosis system, online self-learning method is studied. A support vector machine (SVM) based method combined with an incremental learning algorithm is offered. Such studies could be helpful for developing an online automatic fault diagnosis scheme in an actual RESS, which is important for the safety of a ship. The work has important academic value and a broad application prospect.

针对实际船舶舵机电液伺服系统中存在的未知时变风浪流干扰、不确定参数以及非线性等多重内外不确定因素对系统故障准确诊断带来极大挑战的问题，本项目重点研究这些多重不确定因素下系统故障的有效诊断方法，根据系统中各类不确定因素特点，研究分类逐层干扰解耦和抑制综合策略，提出集成基于模型与数据驱动融合方法的鲁棒故障诊断方法；针对目前实际船上不仅想知道系统中有没有故障，更希望知道已发生故障的类型和具体位置等的需求，充分考虑系统中的执行器、元部件和传感器故障，研究不确定条件下基于多观测模型的分步协作式故障隔离方法，并研究制定在线故障隔离定位的流程体系；针对目前船上故障诊断系统可扩展的需求，研究在线自学习方法，提出基于支持向量机的船舶舵机电液伺服系统在线多分类增量式自学习模式。本项目研究成果将有望应用于船舶舵机电液伺服系统的在线故障诊断和维护，研究工作具有重要的学术研究价值和广泛的实际应用前景。

船舶舵机电液伺服系统是一个复杂的机、电、液综合系统，回路错综，耦合性强。该系统内部存在多类不确定参数以及不确定非线性关系，此外，实际船舶舵机在航行过程中会受到未知时变的风、浪、流，造成系统的外干扰力也是未知时变的，这些多重内外不确定因素对系统故障准确诊断带来极大的挑战。本项目重点研究多重不确定因素下系统故障的有效诊断方法，根据系统中各类不确定因素特点，研究干扰逐层解耦和抑制策略，提出了模型数据联合驱动的鲁棒故障诊断方法，可实现系统常见故障的有效检测；针对目前实际船上不仅想知道系统中有无故障，更希望知道已发生故障的类型和位置等需求，充分考虑系统中的执行器、元部件和传感器故障，提出了不确定条件下基于多观测模型的分步协作式故障隔离方法，建立基于神经网络的补偿模型，构建系统早期故障隔离定位体系，可实现系统多类故障的有效辨识定位；对船舶舵机关键元件开展了故障自学习方法研究，研究了元件故障特征的有效提取方法，针对元件早期故障特征不明显难以区分的问题，开展了特征数据的非线性映射研究，制定学习算法，可以较低的复杂度检测到元件早期故障，提高故障诊断的敏感性。本项目研究是对电液伺服系统故障诊断理论的深入和诊断方法的探索创新，研究成果有助于实现船舶舵机电液伺服系统的在线自动故障诊断，提高故障排查效率，保障船舶安全航行，研究工作具有重要的科学意义和工程价值。