Procedural Defect Modeling for Virtual Surface Inspection Environments

Bosnar, Lovro; Hagen, Hans; Gospodnetić, Petra

doi:10.1109/mcg.2023.3243276

Cited by 5 publications

(3 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[ 3 ] used Unity to create synthetic data for spacecraft component detection. In industrial contexts, synthetic data have been employed for tasks including segmenting car parts in the automotive industry [ 4 ], training robots for bin-picking chicken fillets in the food industry [ 5 ], and detecting surface defects on machine parts [ 6 , 7 , 8 ]. The generation of the synthetic images from 3D geometry is often carried out by using Unity [ 3 , 4 ], and sometimes custom solutions are put into place [ 7 , 9 ], whereas Blender is utilized in the majority of the studies [ 1 , 2 , 5 , 6 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 ].…”

Section: Introductionmentioning

confidence: 99%

“…This is based on the assumption that a greater extent of domain randomization leads to enhanced domain adaptation and, thus, improved model performance for real images. While it’s common to alter at least some of the rendering parameters, particularly focusing on camera position and lighting, beyond this assumption, most studies do not delve deeper into the specific impacts of these variations [ 1 , 2 , 3 , 4 , 5 , 6 , 7 , 10 , 12 , 13 , 14 , 15 , 17 , 18 , 19 ].…”

Section: Introductionmentioning

confidence: 99%

“…These range from medical domains, such as Schenkenfelder et al's [1] generation of virtual images for burn wound detection on human skin, to civil infrastructure inspection, such as Howells et al's [2] work that can be used in assessing earthquake damage, and even space technology, where Viggh et al [3] used Unity to create synthetic data for spacecraft component detection. In industrial contexts, synthetic data have been employed for tasks including segmenting car parts in the automotive industry [4], training robots for bin-picking chicken fillets in the food industry [5], and detecting surface defects on machine parts [6][7][8]. The generation of the synthetic images from 3D geometry is often carried out by using Unity [3,4], and sometimes custom solutions are put into place [7,9], whereas Blender is utilized in the majority of the studies [1,2,5,6,[10][11][12][13][14][15][16][17][18][19].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Synthetic Training Data in AI-Driven Quality Inspection: The Significance of Camera, Lighting, and Noise Parameters

Schraml,

Notni

2024

Sensors

View full text Add to dashboard Cite

Industrial-quality inspections, particularly those leveraging AI, require significant amounts of training data. In fields like injection molding, producing a multitude of defective parts for such data poses environmental and financial challenges. Synthetic training data emerge as a potential solution to address these concerns. Although the creation of realistic synthetic 2D images from 3D models of injection-molded parts involves numerous rendering parameters, the current literature on the generation and application of synthetic data in industrial-quality inspection scarcely addresses the impact of these parameters on AI efficacy. In this study, we delve into some of these key parameters, such as camera position, lighting, and computational noise, to gauge their effect on AI performance. By utilizing Blender software, we procedurally introduced the “flash” defect on a 3D model sourced from a CAD file of an injection-molded part. Subsequently, with Blender’s Cycles rendering engine, we produced datasets for each parameter variation. These datasets were then used to train a pre-trained EfficientNet-V2 for the binary classification of the “flash” defect. Our results indicate that while noise is less critical, using a range of noise levels in training can benefit model adaptability and efficiency. Variability in camera positioning and lighting conditions was found to be more significant, enhancing model performance even when real-world conditions mirror the controlled synthetic environment. These findings suggest that incorporating diverse lighting and camera dynamics is beneficial for AI applications, regardless of the consistency in real-world operational settings.

show abstract