Prediction model for nonlinear variation of bolt preload with tightening torque based on mechanism and data fusion

Abstract

Abstract In response to the difficulty in predicting the change of bolt preload when using torque method to load bolt, this paper proposes a bolt preload prediction method based on mechanism and data fusion calculation for hexagonal end face bolt, and establishes a tightening prediction model based on machine learning method. Firstly, a tightening mechanism model is established, revealing the reasons why bolt preload is difficult to predict and errors cannot be eliminated. Secondly, sensitivity evaluation indicator is established to conduct parameter sensitivity analysis, and the fusion method of “mechanism model guiding data model to perform feature selection” is determined. Finally, the tightening prediction model based on Gaussian Process Regression is proposed, and corresponding engineering prediction software is established. The experimental results show that this prediction model can not only predict the variation of bolt preload with tightening torque, but also synchronously display the confidence interval of bolt preload fluctuation in a probabilistic sense. Under different operating conditions, the prediction accuracy still remains above 98.18%. The prediction model breaks through the limitation of traditional method, which calculates the torque coefficient and indirectly loads the bolt preload.

Keywords

• Bolted connection
• Mechanism model
• Mechanism and data fusion
• Gaussian process regression
• Prediction model