A DATA-KNOWLEDGE INTEGRATED APPROACH FOR PREDICTING FRETTING FATIGUE LIFE OF FRICTION-TYPE CONNECTIONS WITH HIGH-STRENGTH BOLTS

Interfacial fretting fatigue in high-strength bolted friction-type connections constitutes a critical limiting factor affecting their long-term performance and structural safety. To enable accurate prediction of service life, this study first establishes a theoretical solution for the mechanical response within the fretting contact region of the interface by employing the modified Hertz contact theory and the Boussinesq-Cerruti displacement solution. Accounting for the wear-fatigue coupling effect, a damage parameter correction model is proposed by incorporating relative slip displacement and surface roughness parameters, thereby improving the prediction accuracy of the SWT and KBM models-with an average increase of approximately 20% in the proportion of predictions falling within a 1.5-fold error band. Both a purely data-driven PSO-BP neural network model and a data-knowledge-integrated neural network model are developed to predict the fretting fatigue life under multi-factor coupling conditions. Results demonstrate that the proposed data- and knowledge-driven approach achieves high predictive accuracy, with 95.2% of predictions from the integrated model in the test set lying within the 2-fold error band. This work presents a novel framework for evaluating the fretting fatigue performance of bolted connections under complex and coupled influencing factors.