BIM solutions require a digital model as a foundation to optimize processes such as maintenance, infrastructure renovation, or demolition. However, a vast number of analog building plans are archived by public entities managing urban development, and manually converting these plans into digital models, which is prohibitively expensive. To address this gap, the paper introduces an approach for organizations who need to convert large datasets of legacy electrical floorplans into a BIM. The approach leverages a Machine Learning model for instance segmentation to detect electrical features, and the linesegment detection model DeepLSD for extracting cable traces. To support model training, a new dataset, referred as IPVBA-ELEC, is provided. The approach assembles circuits by establishing semantic relationships between circuit components and wires, and store them in an IFC file. Case studies were evaluated using quantitative and qualitative techniques yielding promising results and encouraging further research of additional MEP domains.