AI-assisted Radiographic Defect Recognition
X 光底片与 AI 缺陷自动判读
标签：AI探伤, X光检测, 深度学习, 无损检测
铜合金铸件 X 光探伤按 ASTM E272 透度计评级，人工判片效率低且易漏检；近年采用深度学习模型（U-Net/YOLO）对数字化底片（DR 采集，分辨率 50 µm 以上）自动标注气孔、缩松与夹渣。某铸铜厂用 3000 张历史底片训练后，单张判片时间从 15 min 降到 30 s，漏检率比人工低 40% 以上。注意 AI 结果仍需 II 级无损检测人员复核签发，ASTM E1030 仍以人工评定为最终依据，AI 只作初筛与培训工具。
Copper casting radiography is graded to ASTM E272 image quality; manual film reading is slow and prone to misses. Deep-learning models (U-Net/YOLO) now flag porosity, shrinkage and inclusions automatically on digitized radiographs (DR capture at 50 µm resolution or better). One copper foundry trained on 3,000 archived films: reading time fell from 15 min to 30 s per image with 40%+ fewer misses than manual review. AI output still requires sign-off by a Level II NDT operator — ASTM E1030 keeps human evaluation final, with AI as a screener and training aid.
