Fraunhofer Institute for Mechanics of Materials IWMHow can manufacturing costs for the support structures of offshore wind turbines be reduced? WeldScanPro-LP is developing an AI-based system that automatically digitizes welds on monopiles via 3D scanning and evaluates their local fatigue strength with spatial resolution using artificial neural networks. For the first time, weld geometry, residual stresses, and microstructure are combined in a data-driven model to systematically reduce conservative design margins. The result replaces the previous manual spot-check inspection with a comprehensive, automated inline inspection. Companies benefit from reduced wall thicknesses, less rework, and digitized quality documentation across the entire weld.
Automated inline inspection and spatially resolved fatigue strength assessment of welded joints on offshore monopiles using 3D surface digitization and artificial neural networks, taking into account local notch geometry, residual stresses, and microstructure to reduce manufacturing and maintenance costs
Ongoing research project
Project description
Market Needs and Challenges
The expansion of offshore wind energy is a central pillar of the European energy transition. Approximately 30% of the investment costs for an offshore wind farm are attributable to monopile foundations—tubular steel structures now measuring 7–10 m in diameter and over 100 m in length, connected by circumferential welds. The fatigue strength of these welds is critical to the design for a 25-year service life under wind and wave loads. However, current design practices are extremely conservative: they take into account neither the actual local weld geometry nor the individual residual stress and microstructural conditions. Quality control is performed manually using templates on a small number of samples—a process that is time-consuming, operator-dependent, and not comprehensive. This leads to unnecessarily thick sheet metal, excessive rework, and significant additional costs. At the same time, modern 3D scanners and machine learning methods open up entirely new possibilities for digital weld evaluation.
Project Approach
WeldScanPro-LP is developing an automated, operator-independent system for spatially resolved evaluation of the fatigue strength of welds. High-resolution 3D scans (strip light projection and laser triangulation) digitize the entire weld surface. Geometric parameters and local notch shape factors are automatically derived from the surface models—supported by FE simulations and KNN. In addition, residual stresses (determined radiographically and numerically) as well as microstructural properties (hardness, half-width) are measured with spatial resolution and correlated with the local fatigue life. An artificial neural network is trained using fatigue test data to predict the local fatigue life along the entire weld from a small number of input variables.
Contribution to Addressing the Challenges
The project is expected to enable savings of an estimated 10–20% in manufacturing costs per monopile (€350,000–700,000 per foundation) through: automation of quality control, targeted reduction of conservative design margins, avoidance of unnecessary rework, and comprehensive digital documentation. The methodology is transferable across industries to all welded structures subject to fatigue—from shipbuilding to crane construction to bridge construction. The results strengthen the competitiveness of the German maritime industry and support the expansion of renewable energy through more cost-efficient offshore structures.
Fraunhofer IWM’s work packages in the project:
Fraunhofer IWM provides the experimental, numerical, and data-based foundation for a precise, locally resolved fatigue strength assessment of welded structures. It creates the database, the assessment models, and the physical validation upon which the project’s ML-based methods are built.
Complete experimental characterization of fatigue behavior
Fraunhofer IWM is compiling a comprehensive, high-resolution dataset of local factors influencing fatigue strength for all welded specimens and real monopiles—including residual stress depth profiles, hardness profiles, grain sizes, and dislocation densities. What makes this unique is that residual stresses and microstructural material properties are simultaneously obtained from a single automated X-ray measurement run—a unique selling point that significantly reduces the measurement effort required for ML training compared to conventional approaches.
Fatigue database with in-situ crack initiation data
The Fraunhofer IWM generates a dense, locally resolved fatigue database: For each specimen, multiple crack initiation locations are recorded along with the corresponding local load cycles—which reduces the required number of specimens and test duration by up to 50%. The result is a validated dataset that directly links the failure location, failure time, and local influencing factors, thereby forming the basis for ML training.
Synthetic Training Dataset and Validated Residual Stress Fields for KNN and structural durability
Fraunhofer IWM is providing the project with two key numerical results: first, 5,000–10,000 automatically calculated simulation datasets for notch shape factor determination from 3D surface scans as a training basis for the KNN; second, validated local and global residual stress fields on two length scales—from the weld geometry to the overall structure—including cyclically stabilized residual stresses under real operating loads.
Based on the project results, Fraunhofer IWM can offer industrial companies the following research and development services:
1. Automated X-ray-based residual stress analysis of welded joints
Fraunhofer IWM measures the residual stress state of welds with spatial resolution and quantitatively—even directly on-site on large structures—using a robot-guided diffractometer based on the cos α method.
2. Fatigue testing with in-situ crack detection using Digital Image Correlation (DIC)
Fraunhofer IWM provides a dense, spatially resolved fatigue database with automated detection of crack initiation and crack propagation along the entire weld—multiple independent data points are obtained for each specimen.
3. Non-destructive characterization of local material properties from X-ray diffraction data
Fraunhofer IWM simultaneously derives local material properties such as hardness, dislocation density, and microstructural gradient from the same X-ray measurement data used to determine residual stress—without the need for additional metallographic preparation.
4. Numerical prediction of residual stresses in welded joints using FE simulation
Fraunhofer IWM calculates local and global residual stress fields in multi-pass welded joints and large tubular structures on two length scales—from the weld cross-section to the overall structure—and validates these against experimental measurement data.
5. Automated determination of notch shape factor from 3D weld bead surface scans
Fraunhofer IWM automatically converts real 3D scan data of weld seam surface geometries into FE models and uses them to calculate spatially resolved stress concentrations along the entire length of the weld.
6. AI-Supported Service Life Prediction for Welded Joints
Fraunhofer IWM applies machine learning methods to perform comprehensive, spatially resolved fatigue strength assessments based on 3D scan data and a few supplementary parameters—enabling quality assessments in a matter of seconds directly on the production line.
