Fraunhofer Institute for Mechanics of Materials IWMNuclear transport and storage containers and other cast iron components must meet the highest safety requirements—current verification methods are well-established but time-consuming, expensive, and often overly conservative. DaGeMoCast is developing an AI-supported model that predicts fracture behavior with spatial resolution and load dependence based on small-specimen tests, microstructural data, and simulations. For the first time, this approach directly incorporates local microstructural differences and real stress conditions into the safety analysis. For manufacturers and operators, this means more realistic assessments, reduced testing requirements, and quantifiable safety margins. The methodology is transferable to other materials and components in the nuclear industry. The Fraunhofer IWM is collaborating with the TU-Bergakademie Freiberg on a joint research project. The partners from Freiberg are primarily responsible for the numerical simulations.
Data-Driven Safety Verification of Nuclear Safety Components Made of Ductile Iron Based on Spatially Resolved Small-Scale Sample Tests
Ongoing research project
Project description
Market need and societal challenge: Germany faces the task of safely storing highly radioactive waste for significantly extended periods until a final repository becomes available. The transport and storage containers used for this purpose are primarily made of ductile iron. To demonstrate their safety, it must be proven that even in the event of a severe accident—such as a crane collapse—no fracture will occur. Current practice relies on extensive test series using large, expensive standard specimens, the results of which are conservatively applied as a lower bound to the entire component. Local microstructural variations, which inevitably occur in thick-walled castings, and the actual stress conditions at the hypothetical crack are not taken into account, resulting in safety margins that go unused. This leads to a highly conservative approach, significant testing effort, and limited flexibility in the evaluation of new or aged containers.
Project Approach: DaGeMoCast is developing a data-driven model (machine learning) that predicts fracture toughness with spatial resolution based on microstructural parameters, chemical composition, and loading conditions—without the need for additional large-scale specimen testing. To this end, a comprehensive database is being established using existing and supplementary small-specimen tests, literature values, and synthetic data from micromechanical and damage-mechanics simulations. The model is conceptually validated using tests on large specimens. The goal is a more realistic safety assessment that takes into account the actual local material condition, thereby quantifying and specifically reducing conservativities.
Contribution to addressing the challenges: The methodology drastically reduces the experimental effort, enables a microstructure-based, spatially resolved assessment for the first time, and establishes the scientific foundation for a concept with future viability that is also applicable to extended interim storage periods, aged materials, or new container designs.
Fraunhofer IWM’s work packages in the project:
Fraunhofer IWM establishes the experimental, data-analytical, and modeling foundation of the project—from the validated material database through an AI-supported prediction model to a publicly accessible database for standard-compliant fracture toughness properties.
Validated, comprehensive material database for cast materials subjected to dynamic loading
Fraunhofer IWM provides a consolidated, quality-assured database comprising approximately 400 test results—consisting of processed data from the predecessor project ProCast and targeted supplementary experiments conducted under varying microstructural conditions, temperatures, and loading rates. This database forms the foundation for all subsequent modeling and learning steps in the project.
Synthetic Material Data for Microstructural Conditions Not Accessible Experimentally
Using micromechanical simulation, Fraunhofer IWM generates valid synthetic material properties—stress-strain curves and effective fracture parameters—for microstructural states that cannot be covered experimentally or would require a disproportionate amount of effort to do so. The result is a comprehensive, validated database that enables robust predictions across the entire relevant parameter space.
AI Model for Predicting Standard-Compliant Fracture Toughness Properties and a Public Database
Fraunhofer IWM is developing the project’s central data-driven prediction model: a trained ML model that directly predicts standard-compliant material properties from microstructural and loading parameters—including methods for evaluating model quality and reliability, as well as a final validation using large specimens. In addition, Fraunhofer IWM will create and operate the database, which will be publicly accessible upon project completion, and make it permanently available to the professional community.
Based on the project results, Fraunhofer IWM can offer industrial companies the following research and development services:
1. AI-supported prediction of fracture toughness and failure behavior from local microstructural data
Fraunhofer IWM provides spatially resolved fracture toughness parameters for cast iron components based on local microstructural data—without the need for time-consuming large-scale specimen testing.
2. Fracture mechanics-based safety assessment taking into account component-specific constraint effects
Fraunhofer IWM quantifies the extent to which the actual multi-axial stress distribution at a hypothetical component crack differs from that in standard test specimens—and calculates the resulting actual safety margins.
3. Automated deep learning analysis of fracture surfaces
Fraunhofer IWM analyzes fracture surfaces of test specimens or damage cases using trained neural networks and provides an objective, quantitative determination of the ductile and brittle fracture fractions.
4. Development of Customer-Specific Material Databases and Predictive ML Models
Fraunhofer IWM develops customized material databases and machine learning models that precisely map the customer’s specific materials, manufacturing processes, and test conditions and make them permanently usable.
5. Micromechanical simulation and synthetic data generation for hard-to-access microstructural states
Using validated micromechanical models, Fraunhofer IWM generates virtual material properties for microstructural states that are not accessible experimentally or can only be accessed with disproportionate effort—such as new alloy variants, aged, or irradiated materials.
