Reliable detection and classification of defects in manufactured objects continues to pose a challenge in industrial quality control, where even minor anomalies can lead to significant economic losses, safety hazards, or downstream process failures. Conventional inspection systems that rely solely on visual or geometric data often struggle to handle subtle or structurally complex defects. In response, we propose a supervised deep learning framework that employs both spatial (RGB) and geometric (depth) information to improve the classification of defective components. Our approach integrates these complementary modalities through a VGG16-based architecture adapted to jointly process 2D and 3D features. This multimodal fusion enables the model to capture richer representations of industrial objects and leads to improved classification performance. Despite the limited availability of annotated data, the method demonstrates high effectiveness.