This conference paper was submitted for presentation at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

Manufacturers may suffer considerable product quality losses due to defects during production. The manual inspection used in current inspection techniques can be time-consuming, expensive, and inconsistent. This paper addresses a hybrid framework for defect detection in manufacturing. It combines synthetically generated 2D data with 3D Point Cloud representations to improve the accuracy and efficiency of quality control processes. The approach discussed in this research discusses defect patterns such as scratches, porosity, and gaps using AI-based approaches to create diverse training datasets, addressing the challenge of obtaining labelled actual defect data. In order to analyze the defects that occur during a manufacturing process, this research uses a 3D scanning tool like the Faro scanning arm to collect point cloud data of the formed part. The 2D defects are generated synthetically and integrated with the spatial richness of 3D Point Cloud data. It enables a dual-layer validation approach where defect predictions are verified and validated using actual images, improving reliability. The hybrid framework uses new methods to combine the strengths of transfer-learning CNNs for analyzing 2D data and PointNet-based architectures for 3D analysis. This helps achieve accurate classification of defects. New methods for modelling defects, like using GANs to create synthetic data, help detect complex defects effectively. This research tested this framework on industrial datasets. It connects synthetic and real-world applications and offers a scalable, semi-supervised process for detecting defects in real time. This hybrid framework reduces the need for manual inspections, improves production efficiency, reduces waste, and ensures better product quality. It is especially suitable for high-precision aerospace, automotive, and electronics manufacturing industries. This research addresses critical challenges in automated quality control by combining AI, synthetic data creation, and real-world testing. It provides a solid and adaptable solution for the future of manufacturing. This study introduces a new way to measure defects in produced parts using 2D and 3D data. The new approach offers an accurate, reliable, and efficient approach to identifying defects. It has been tested through experiments and compared with other advanced methods, such as feature extraction image analysis, showing its ability to handle large amounts of data. This method can be easily added to current manufacturing processes, potentially leading to better-quality parts and lower production costs.