Analysis and optimization of assembly accuracy for a specific bolted joint mechanism based on CatBoost and Bayesian algorithm

This study proposes an integrated approach, including rough surface morphology reconstruction, finite element method modeling of threaded connections, proxy model training and assembly parameters optimization, to investigate the alignment precision of the HexaLock Positioning Assembly in precision inertial platforms as affected by bolt preloads and rough contact surface topography. Firstly, the microscopic morphology of cylindrical contact surfaces was reconstructed based on measured roughness parameters by using conjugate gradient method. Then, a finite element model considering both threaded interfaces and rough contact surface was established. A dataset of center deviation was generated based on 50 sets of Latin hypercube sampling. The CatBoost model was trained to build a surrogate model, which exhibited higher prediction accuracy than other six popular algorithms, RF, SVR, GBM, XGBoost, and AdaBoost. The Bayesian optimization was employed to globally optimize the tightening angles of the six bolts with the objective of minimizing the center offset. The optimized tightening scheme achieved center offset of 0.62 μm, which was better than experimental measurements and within-group simulation results, and consequently improved the assembly accuracy of HexaLock Positioning Assembly significantly. This study provides a new data-driven approach for error prediction and preload control in precision assembly of inertial platforms.