Optimizing Scalable Additive Repair Using PBF-LB/M with Real-Image-Based Synthetic Data for Defect Detection

Additive repair of parts with powder bed fusion of metals using a laser beam (PBF-LB/M) is a promising approach for extending the service life of high-value parts, particularly in industrial applications where cost-efficiency and sustainability are crucial. However, the process remains time-consuming, especially when repairing parts of varying scales. To reduce manufacturing time and improve process efficiency, a damage analysis method is proposed to predict the position, type, and severity of defects prior to repair. The damage analysis relies on object detection algorithms that require high-resolution images to accurately identify and classify defects. However, the availability of real-world defect images is limited particularly for parts such as turbine blades, which vary widely in size and application. To overcome this training data scarcity, this contribution presents a synthetic image generation approach. This method uses real part images as a foundation to automatically generate and augment realistic defect scenarios based on defined damage parameters and modeling techniques. The results of the model training show that the generated synthetic images are a promising alternative to real-world datasets and can effectively support the object detection process within the additive repair workflow.