LBMcheck (FFG - ASAP 13)

This work was carried out within the framework of the Austrian Research Promotion Agency's (FFG) ASAP program and was funded by the Austrian Federal Government.

As part of the LBMcheck project, the consortium leader FOTEC Forschungs- und Technologietransfer GmbH and the consortium partners Vienna University of Technology and Plasmo Industrietechnik GmbH are working on quality assurance for the LSS process, which has enormous potential. Component shapes are conceivable and feasible that are precisely optimized for the respective load (geometric freedom). Furthermore, clamping problems are eliminated, and the process achieves the mechanical properties of the base material. The process can optimally leverage these strengths in the aerospace sector. In these industries, weight savings offer enormous advantages, which can be significant for additively manufactured components due to the design freedom of the process. However, as long as no functioning quality assurance system is available on the market, end users often show little acceptance for this manufacturing technology because component inspection is very expensive.

The challenge lies in finding suitable algorithms and a reliable measurement system. Although several research groups have been working on this for some time, no product is yet available on the market. Another difficulty is implementing an industrially viable hardware solution in laser stacking (LSS) systems.

All LSS systems share the characteristic that the build process is purely controlled, meaning it operates without signal feedback in the sense of closed-loop control. While some research groups and machine manufacturers have already succeeded in monitoring parts of the process, such as the residual oxygen concentration in the process chamber, the power of the processing laser, the temperature of the build platform, and possibly even the powder application, neither research groups nor machine manufacturers have yet achieved complete control of the entire build process. Machine manufacturers EOS and Concept Laser recently presented initial steps toward a representative quality control system in the form of preliminary process documentation. In summary, it should be noted that while the manufacturing process is logged using sensor data, this data is generally not available for real-time analysis and only reflects process irregularities (anomalies). This means that the relationship between irregularities in the sensor data and the resulting actual component quality (defect catalog) is unknown, and therefore no reliable assessment of a component's quality is possible.

• Research and implementation of algorithms for calculating characteristics that enable the modeling of component quality from sensor data using system identification methods. The algorithms must be implementable in real time and should preferably operate without reference data.
• Development of a classification strategy that allows for the mapping of individual quality requirements. This objective will be achieved using optimization strategies such as genetic algorithms. By employing these strategies and linking the sensor data, the false defect rate should be reduced by 50% compared to using diode-based monitoring alone.
• Verification/demonstration of the developed system using test geometries for two materials (e.g., AlSi10Mg and Inconel). Verification/demonstration of a 30% reduction in scrap, as well as a 30% reduction in existing computed tomography inspections, resulting in a significant reduction in inspection costs and time.

• Specifically, two measurement systems (diode-based and camera-based) were successfully implemented, and proof of concept was provided that characteristics can be calculated from the measurement signals that correlate with process variations and component defects. The corresponding analysis toolbox was developed within the project.
• Samples were successfully created for the assessment of defects; for example, porosity was successfully compared with CT analyses of the components. Further systematic correlations, e.g., regarding the extraction systems of the machines, were identified and demonstrated in the signals.
• Error catalogs were created, but the main challenge in the project was the lack of corresponding product standards. Consequently, there are no universally applicable guidelines for classifying errors as, for example, acceptable or unacceptable. This made a final alignment of the models, as planned in AP4, difficult. Instead, this classification was based on the project team's experience, successfully demonstrating that appropriate modeling is possible. Furthermore, the framework was developed to accommodate previously non-existent customer-specific quality requirements.
• A highlight is the approach to using spatially defined characteristics. The project transitioned from raw signals with high data volumes to 2D layer-based and 3D volume-based digital process twins. In addition to the higher information content, this results in a drastically reduced storage requirement (a reduction in data volume by a factor greater than 1000), thus enabling almost interactive work with the digital twin.
• Another highlight is the automated calculation of camera characteristics on an FPGA. Proof of concept was provided here that key performance indicators (KPIs) can be calculated in real time using more than 20,000 images per second.
• Furthermore, a machine learning-based approach was developed for the automated detection of anomalies within seconds, without requiring any predefined parameters. Using unsupervised learning methods, KPIs from the digital twins were correlated to identify stochastic and systematic process changes and component anomalies.

More information