Abstract: In tank defect inspection, achieving automated detection holds significant practical importance. Currently, tank bottom plate inspections often require tank opening and cleaning followed by manual operations, which are labor-intensive and expose workers to hazardous internal environments that pose serious health risks. This study designed and developed a self-propelled inspection robot based on remote field eddy current (RFEC) detection technology that overcome interference from non-ferromagnetic impurities on tank bottom, thus achieving automated inspection of tank bottom plate. Simultaneously, to address insufficient data accuracy in remote field eddy current (RFEC) detection of tank bottom plate, an intelligent defect recognition model based on Gradient Boosting Decision Tree (GBDT) was constructed. The model forms a training dataset containing dimensions such as defect size and eddy current phase shift by utilizing geometric parameters of tank bottom plate defects (length, width, depth) and corresponding signal features from Remote Field Eddy Current (RFEC) sensor arrays. This enables the inversion of defect morphological characteristics using RFEC signals. This inspection technology provides an efficient and precise unmanned solution for petrochemical storage tank bottom plates, yielding substantial engineering merits in safeguarding industrial storage and transportation safety.

Key words: roboticsltank floor corrosionlonline inspectionlgradient boosting decision tree