In construction environments, modifications to the dimensions, positioning, and trajectory of plumbing infrastructure within edifices are frequently necessitated by on-site conditions and pragmatic installation procedures. Recent advancements in Scan-to-BIM technology have streamlined pipe construction processes by monitoring development through a 3D model. However, existing 3D point cloud processing methods rely heavily on given local geometric information to distinguish pipes from adjacent components. Furthermore, point clouds originating from construction environments are mostly class-imbalanced data which could negatively impact the date-driven approach. This paper proposed a novel framework for segmenting and reconstructing piping systems utilizing raw 3D point cloud data acquired from construction sites, addressing the aforementioned challenges. The data firstly undergoes preprocesssing, including the elimination of redundant points, rotational adjustments, and sampling procedures. Subsequently, a point cloud semantic segmentation network is trained to predict the per-point class labels after adding local features and mitigating the class imbalance issues. Finally, Efficient RANSAC is employed to identify cylinder-shaped pipes based on the prediction outcomes. The proposed framework shows superior performance compared to existing semantic segmentation methods and exhibits considerable promise for piping system reconstruction.