A　method　has　been　developed　to　quantify　the　bearing　surface　friction　coefficient　in　bolted　joints　based　on　the　mechanics　of　the　tightening　process.　By　measuring　the　surface　topography　under　different　tightening　torques,　a　strong　correlation　between　the　variation　of　the　microscopic　topography　and　the　friction　coefficient　was　revealed.　The　friction　coefficient　initially　decreases　and　subsequently　increases　due　to　surface　wear　and　microstructural　transformations.　A　genetic　algorithm　(GA)　and　backpropagation　neural　network　(BP)　were　employed　to　create　a　predictive　model,　optimizing　network　parameters　for　enhanced　learning　efficiency　and　accuracy.　The　experimentally　validated　GA-BP　prediction　model　forecasts　friction　behavior　with　a　maximum　relative　error　of　3.57　%,　offering　valuable　insights　for　the　design　and　optimization　of　bolted　joints.