The automation of digital twinning for existing bridges from point clouds remains unresolved. Previous research yielded methods that can generate surface primitives combined with rule-based classification to create labelled cuboids and cylinders. While these methods work well in synthetic datasets or simplified cases, they encounter huge challenges when dealing with real-world point clouds. The proposed framework employs bridge engineering knowledge that mimics the intelligence of human modellers to detect and model reinforced concrete bridge objects in imperfect point clouds. Experiments on ten bridge point clouds indicate the framework can achieve high and reliable performance of geometric digital twin generation of existing bridges.