The construction industry in Germany is undergoing a transformation: The focus is shifting from new construction to the modernization of existing buildings. One challenge in this regard is the often inadequate building documentation. The planned innovation—a novel, rod-shaped locating device—is designed to simplify room surveying and provide detailed information about wall composition, materials, and the routing of utilities (electricity, water, gas). Using sensors and AI, data is generated that is directly integrated into digital building models. This accelerates the planning process while reducing planning uncertainties, which regularly cause major delays on construction sites.

Of particular note is the real-time detection of anomalies within the building, which further enhances operational safety, as well as the visualization of the recorded information using augmented reality. A previous successful feasibility study has already examined the practicality and market potential of this technology.

OFFIS is focusing on the implementation of the subproject “Development of Pipeline Detection and Classification.” This subproject addresses the development of the AI value chain planned for the project: from model selection and design, through the creation of a quality-assured, versioned training database, to the implementation of a real-time-capable analysis pipeline, as well as production-ready classification methods for pipelines and materials. Additionally, it supports system integration and the final validation of the AI.

The innovative core of this subproject is cross-sensor, real-time AI classification with robust measures of uncertainty for field deployment. This includes: 

(a) the combination of modern model families with signal-type-specific preprocessing for radar or capacitive sensor solutions,

(b) robust training and validation protocols, including confidence calibration for reliable decisions,

(c) a latency-optimized edge/cloud pipeline with monitoring for continuous quality assurance, and

(d) clearly defined, machine-readable interfaces for 3D modeling, sensor fusion, and AR visualization.

The results are precise pipeline, material, and anomaly classes, complete with quality indicators, that can be immediately utilized in planning and execution
.