Influence factors prediction of thread looseness based on Bayesian optimization neural network
In the process of thread loosening,there are many influencing factors that have typical nonlinear characteris-tics,and the attenuation of pre tightening force cannot be predicted.Aiming at these problems,a method for predicting the residual pre tightening force of bolts based on mechanism model and test data was proposed.A dynamic model of thread loosening was built,the response surface method was used to quantitatively analyze the influence of factors on the residual preload,and the initial preload and amplitude were determined as the two most sensitive factors affecting looseness.Further,Bayesian optimization algorithm was used to establish a prediction model of bolt residual preload based on neural network,which could accurately predict the bolt residual preload,and the model was verified.The re-sults showed that the mean square error of the neural network prediction model based on Bayesian optimization was the smallest,and the R2 coefficient was the closest to 1,which was superior to the three-layer neural network,Gaussian process regression and support vector machine models.The experimental verification showed that the error between the predicted value of bolt residual preload and the actual test value was within 7％,which verified the effectiveness and re-liability of the model.It laid a foundation for the reliability design of bolts.
NETL
NSTL
万方数据
