Automatic defect detection system for Fused Filament Fabrication process utilizing deep learning models.
Topic(s) : Manufacturing
Co-authors :
Vanessa LISLEVAND (GREECE), Dimitrios KARASAVVAS (BELGIUM), Elias KOUMOULOSAbstract :
A systematic yet parametric workflow is presented ensuring the effectiveness and output quality of an industrial Fused Filament Fabrication (FFF) process, an additive manufacturing (AM) technology providing viable and cost-effective results for prototyping applications and low-volume manufacturing of high-performance functional parts. Despite the numerous advantages of the FFF process, it is prune to defects without relevant process optimization; hence, a systematic and parametric approach to quality control is necessary for industrial upscaling. By employing artificial intelligence-driven computer vision techniques, a flexible framework is introduced, designed for the automatic detection of defects in the FFF process.
Through cutting-edge deep learning models, the surface structure and weld quality of individual thermoplastic strands are real-time monitored, after having established acceptable thresholds for FFF process parameters. The investigation assesses the capabilities of the NVIDIA Jetson Nano, a low-power, high-performance computer and a pre-trained model assessed along with manual configurations to efficiently discern the surface structure of thermoplastic strands. Overall, we propose a methodology that aims to simplify the selection of process parameters and identify the critical defects at an early stage thus increasing the overall manufacturing performance while minimizing any operator interventions required.
Through cutting-edge deep learning models, the surface structure and weld quality of individual thermoplastic strands are real-time monitored, after having established acceptable thresholds for FFF process parameters. The investigation assesses the capabilities of the NVIDIA Jetson Nano, a low-power, high-performance computer and a pre-trained model assessed along with manual configurations to efficiently discern the surface structure of thermoplastic strands. Overall, we propose a methodology that aims to simplify the selection of process parameters and identify the critical defects at an early stage thus increasing the overall manufacturing performance while minimizing any operator interventions required.