Fatigue Life Prediction and Evaluation of Exhaust Manifold Based on Neural Networks

This paper presents fatigue life prediction of an engine exhaust manifold, this approach developed an exhaust manifold prediction model which used neural network technology and finite element analysis model. Firstly, Modern exhaust systems should withstand severe cyclic mechanical and thermal loads throughout the whole life cycle. Thermos-Fluid-Structure coupling analysis established the prestressed conditions and identified the potential fatigue risk area, this area was under cycle fatigue. Secondly, several cycles the fatigue micro cracks would appear in this region. random vibrations were applied in the fatigue life evaluation, and a life prediction model was established and trained to improve the prediction efficiency. Finally, the model showed that the predicted life of parts approached the actual life, the prediction error of fatigue life was controlled within 10%. This method emphasized the advantages of the model which combined neural networks with finite element analysis model to accurately and effectively, and the lifetime of this part can be determined through the modeling analysis instead of only experiment.