Fraunhofer Institute for Mechanics of Materials IWMBlasting processes such as shot blasting are essential for the service life of springs, gears, turbine blades, and welded structures—yet for 80 years, quality control has been based on the simple Almen test, which offers little potential for optimization. For the first time, OptiPeen links the entire process chain—from machine parameters to abrasive dynamics and surface layer condition to component service life—using physics-informed machine learning methods. This enables blasting processes to be optimized in real time: either to maximize service life or to minimize energy and resource consumption while maintaining consistent performance. For industry, this means: up to 20% lower energy consumption in blasting processes, up to a 10% reduction in weight for spring products, and—for the first time—a digital twin of blasted components. The methodology can be applied across industries—from gear manufacturing and spring technology to aerospace.
Optimization of energy and resource usage in shot-blasted components through physically informed machine learning methods
Ongoing research project
Project description
Market Needs and Challenges
Approximately 35 million metric tons of steel are processed annually in Germany—the majority of the components manufactured from it (springs, gears, shafts, turbine blades, welded structures) undergo standard shot blasting treatments to increase surface roughness for corrosion protection, remove contaminants, or specifically enhance fatigue strength. Up to 90% of all component failures are due to fatigue damage, with the condition of the surface layer having a significant impact on service life. Despite this enormous economic significance, quality and process control for blasting processes has hardly evolved in the past 80 years—it is still based on the Almen test from 1944. A holistic optimization of the process chain (machine parameters → abrasive dynamics → surface layer condition → component service life) does not exist. The result: significant energy waste due to over-treatment, unnecessary material consumption due to conservative design, and a lack of options for condition-based monitoring. Compressed air for blasting processes alone consumes up to 141,000 m³ per year per plant; at the same time, optimized processes could reduce CO₂ emissions in the German spring industry by over 42,000 metric tons per year.
The Project’s Approach
OptiPeen is developing a data-driven methodology that, for the first time, links all stages of the process chain for blasted components using machine learning. The approach combines three pillars: (1) A systematic experimental database with approximately 250 parameter sets for four high-strength steels, (2) physically informed numerical simulations ( : CFD for abrasive dynamics, FEM for surface layer conditions, fracture mechanics for service life), which enrich the database and ensure physical consistency, and (3) automated machine learning methods that enable both predictions (forward approach: blasting parameters → service life) and optimizations (reverse approach: desired service life → optimal blasting parameters). A demonstrator software tool and an online monitoring system using high-speed camera-based grain velocity measurement are being developed.
Contribution to Addressing the Challenges
For the first time, the project enables a quantitative assessment of the balance between the benefits achieved through blasting and the energy and resource costs involved. In the short term, blasting intensities can be reduced by over 20% (potential energy savings of approximately 35,000 kWh/year per blasting system). In the medium term, ML-based service life prediction enables a weight reduction of 10% on average for blasted components while maintaining the same service life—representing a CO₂ savings potential of 42,000 metric tons per year in the spring industry alone. In the long term, the digital twin of blasted components opens up pathways to a circular economy: reuse, re-blasting, and condition-based maintenance can now be quantitatively planned for the first time. The results are incorporated into national and international standards (DGM, VDFI, SAE) and strengthen the competitiveness of the German metalworking industry.
Fraunhofer IWM’s work packages in the project:
Fraunhofer IWM provides the physical modeling foundation for the project. It creates the materials-mechanics-based database and the fracture mechanics evaluation models upon which all of the project’s data-driven methods are built—thereby providing the machine learning approach with the physical grounding that is essential for reliable transfer to new materials and process conditions.
Numerical Prediction of the Surface Layer State Following Blasting Processes
Fraunhofer IWM is creating a physically consistent, synthetic database comprising 300–400 automated simulations of the ball impact process—for all four steels under investigation (54SiCr6, 51CrV4, 16MnCr5, 18CrNiMo7-6). The result is a complete set of numerically determined surface layer conditions (residual stress depth profiles, hardness distribution, roughness), which specifically complement experimental measurements and serve as reliable training data for the project’s ML algorithms.
Physics-Based Lifespan Model as a “Physics-Informed” Foundation for ML
Fraunhofer IWM is developing a calibrated fracture mechanics service life model that directly links the numerically determined boundary layer state to the component’s service life. This model—implemented in Fraunhofer IWM’s proprietary software VERB—provides quantitative service life predictions for parameter combinations that cannot be tested experimentally, thereby forming the physical backbone of the ML models in AP7.
Component-Specific Design and User Guide
Fraunhofer IWM applies the developed simulation methodology to industrially relevant component geometries (gears, springs, parabolic cams) and develops a practical user guide that makes the project results accessible for industrial use.
Based on the project results, Fraunhofer IWM can offer industrial companies the following research and development services:
1. Numerical prediction of surface layer conditions following blasting processes
Fraunhofer IWM determines the resulting surface layer conditions—residual stress depth profiles, hardness distribution, and surface roughness—for customer-specific materials and blasting parameters based on validated FE simulations.
2. Fracture mechanics-based service life prediction for blasted components
Fraunhofer IWM generates quantitative service life predictions for customer-specific component geometries and loading scenarios, explicitly accounting for the actual surface layer conditions—implemented in Fraunhofer IWM’s proven in-house software, VERB.
3. Physically Consistent Training Data Generation for ML Models
Using automated FE simulation, Fraunhofer IWM generates synthetic, physically consistent datasets for customer-specific materials and process conditions—serving as a robust training basis for machine learning models aimed at process optimization.
4. Material characterization of blasted surface layers using micro-sample technology
Fraunhofer IWM determines the cyclic deformation behavior directly at the blasted surface layer using micro-tensile-compression tests—thereby providing material data that precisely reflects the actual local work-hardened state.
5. Optimization of blasting intensities to avoid over- and under-treatment
Based on the validated simulation chain, Fraunhofer IWM determines the energetic optimum of the blasting treatment for customer-specific components—the point of maximum service life with minimal energy consumption.
6. Service-life evaluation of beam processes for components with complex initial conditions
Fraunhofer IWM quantifies the interaction between beam treatment and the existing surface layer condition for welded, case-hardened, or otherwise pretreated components.
7. Integration of beam processing data into digital twins and remanufacturing concepts
Fraunhofer IWM supports companies in integrating simulation-based surface layer data into digital twins and deriving condition-based maintenance and remanufacturing strategies from this data.
