船舶柴油机故障预测与健康管理技术综述

  目的  曲轴是船用柴油机的关键部件，为实时监测柴油机运行状态下的曲轴应力，保障船舶安全可靠运行，提出基于数字孪生的曲轴应力预测方法。  方法  基于数字孪生思想，提出船用柴油机智能运维的构架，以某型直列6缸船用中速柴油机为研究对象，在实机可获得数据的基础上，实现基于数字孪生的曲轴健康状态评估。首先，基于径向基（RBF）神经网络，利用缸盖振动加速度信号对缸内压力进行识别，计算曲柄销载荷，利用有限元法对曲轴进行应力和疲劳分析，获得影响曲轴寿命的关键位置，应用反向传播（BP）神经网络构造曲轴受力和应力的降阶模型。  结果  模型的缸压和曲轴应力预测误差均小于5%，应力预测的时间缩短到秒级，可实现孪生模型的实时/准实时更新。  结论  研究成果可为柴油机零部件的状态实时监测和智能运维提供新思路。

A review: ship diesel engine prognostics and health management technology

  Objectives  A crankshaft is a key component of marine diesel engines. In order to monitor crankshaft stress and ensure the safe and reliable operation of ships, a crankshaft stress monitoring method based on digital twin technology is proposed.  Methods  Based on the digital twin concept, a framework for the intelligent operation and maintenance of marine diesel engines is put forward. Taking a certain type of in-line 6-cylinder marine medium-speed diesel engine as the research object, a digital crankshaft health condition assessment system is developed on the basis of available data from real engines. First, based on an RBF neural network, the cylinder head vibration acceleration signal is used to identify in-cylinder pressure, then the load of the crank pin is calculated. Based on the finite element method, the key positions which affect the lifecycle of the crankshaft are obtained after analyzing its stress and fatigue. A BP neural network is used to evaluate the stress of the crankshaft after a reduced-order model of forces and stresses is used to determine real-time performance. A BP neural network is then used to evaluate the stress of the crankshaft.  Results  The prediction errors of the cylinder pressure and crankshaft stress of the proposed model are both less than 5%, and the stress prediction time is shortened to the second level, realizing the real-time/quasi-real-time update of the twin model.  Conclusions  The results of this study show that it can be used as a new reference for the real-time monitoring and intelligent operation and maintenance of diesel engine components.