Accurate　and　non-destructive　evaluation　of　underwater　concrete　strength　is　crucial　for　the　durability　and　safety　of　hydraulic　structures,　yet　remains　technically　challenging.　This　study　proposes　a　robust　approach　combining　a　custom-developed　ultrasonic-rebound　tester　with　a　Deep　Extreme　Learning　Machine　(Deep　ELM)　prediction　model　optimized　by　the　Rime　Optimization　Algorithm　(RimeOA).　By　leveraging　multi-source　test　data　and　an　enhanced　global　optimization　strategy,　the　model　significantly　improves　prediction　accuracy　and　stability.　Experimental　validation　under　laboratory　conditions　shows　that　the　proposed　model　achieves　a　high　Coefficient　of　Determination　(R2)　of　0.978,　with　low　prediction　errors　(a　Mean　Absolute　Error　(MAE)　of　1.85　MPa,　a　Root　Mean　Square　Error　(RMSE)　of　2.08　MPa,　and　a　Mean　Absolute　Percentage　Error　(MAPE)　of　3.55　%),　significantly　outperforming　conventional　models.　Preliminary　field　tests　further　demonstrate　the　feasibility　of　the　proposed　method　for　in-situ　applications.　These　findings　suggest　that　the　method　provides　a　reliable,　precise,　and　practical　tool　for　assessing　underwater　concrete　strength,　offering　strong　potential　for　intelligent　structural　monitoring　in　complex　service　environments.　©　2025　Elsevier　Ltd