摘要:针对柴油机运行时部分零件应力无法直接测量的问题,以柴油机连杆为例,提出一种基于模型降阶的连杆应力实时重构方法。通过采集某型直列四缸柴油机的缸压、转速等可测数据,基于刚柔耦合多体动力学仿真结果,修正曲柄连杆机构动力学公式用于构造机理模型,实现可测数据向连杆载荷数据的转化。利用瞬态动力学仿真获取连杆多节点应力,通过径向基(radial basis function,RBF)神经网络构建连杆多节点应力的降阶数值模型。经机理模型与数值模型相结合实现模型降阶。降阶后的模型计算结果与有限元仿真结果相比较,连杆峰值应力重构误差均小于2%,应力重构时间缩减至0.1 s,满足剩余寿命预测的准确性及实时性要求。该方法可应用于此类应力无法直接测量零件的实时重构,为剩余寿命预测提供基本条件。 |
关键词: 柴油机 连杆 模型降阶 神经网络 |
|
A Real-Time Reconstruction Method of Diesel Connecting Rod Stress Based on Model Reduction |
HE Zhaowei,JIN Jiangshan,QIN Ciwei,KONG Chen |
Shanghai Marine Diesel Engine Research Institute, Shanghai 201108, China |
Abstract:In view of the problem that the stress of some parts in diesel engines can not be obtained by direct measurement when the diesel engine is running,taking the connecting rod of a diesel engine as an example,a real-time reconstruction method of connecting rod stress based on model reduction is proposed.Through the collection of a certain type of inline four-cylinder diesel engine cylinder pressure,rotational speed and other measurable data,and based on the results of rigid-flexible coupled multi-body dynamics simulation,the crank-connecting rod mechanism dynamics equation was modified for the construction of mechanism model to achieve the conversion of measurable data to connecting rod load data.Transient dynamics simulation was used to obtain the connecting rod multi-nodal stress, and the radial basis function(RBF)neural network was used to construct a connecting rod multi-nodal stress reduced-order numerical model. Model reduction was realized by combining the mechanism model with the numerical model.The simulation results of model reduction were compared with the finite element simulation results.It showed that the reconstruction error of the peak stress of the connecting rod was less than 2%,and the stress reconstruction time was reduced to 0.1 s,which satisfied the accuracy and real-time requirements of the remaining useful life prediction.This method can be applied to the real-time reconstruction of the parts whose stress can not be directly measured,which provides basic conditions for remaining life prediction. |
Key words: diesel engine connecting rod model reduction neural network |