To　addresses　the　challenge　of　underwater　grouting　sleeve　corrosion　affecting　bridge　service　life,　we　propose　an　innovative　inspection　approach　by　integrating　a　modified　ultrasonic　testing　instrument,　ultrasonic　data　augmentation　method,　and　an　improved　ensemble　learning　prediction　model.　Firstly,　the　ultrasonic　data　was　collected　from　grouting　sleeve　specimens　by　the　modified　instrument.　Secondly,　a　data　generation　model,　incorporating　Bray　Curtis　distance　and　rejection　sampling　algorithm,　was　developed　to　generate　samples　and　filter　abnormal　data.　Then,　Circle　Chaotic　Map　improved　Grey　Wolf　Optimizer　was　proposed　to　optimize　parameters　from　the　Random　Forest　(RF)　algorithm　for　establishing　the　prediction　model.　Finally,　SHapley　Additive　exPlanations　(SHAP)　method　was　used　to　interpret　the　model　globally　and　locally.　Experimental　results　validate　the　effectiveness　of　the　data　augmentation　method,　ensuring　high-quality　ultrasonic　data　for　accurate　corrosion　rate　predictions.　The　proposed　inspection　method　provide　technical　support　for　assessing　remaining　bridge　bearing　capacity　and　service　life,　showcasing　high　accuracy,　interpretability,　and　noise　tolerance.