Predictive Modeling of Bolt Preload Using Multilayer Perceptron Neural Networks

Predicting the preload change in bolted joints under transverse vibration loading is challenging due to its complex behavior. This paper proposes a deep learning-based method to predict preload loss in bolted joints when they become loose. A theoretical model of bolted joint behavior was first developed to generate preload curves under various operating conditions, which were then used to create the dataset for training the deep learning model. A deep learning neural network model was subsequently designed, and k-fold cross-validation was employed to optimize the model's hyperparameters, ensuring accurate predictions. Finally, the predictions of the deep learning model were compared to those obtained from the theoretical model. The results demonstrate that, with a network architecture of 4 hidden layers of 512 neurons each, a learning rate of 0.0005, a batch size of 128, and 200 epochs, the two approaches exhibit consistent trends.