FOD-S2R: A FOD Dataset for Sim2Real Transfer Learning based Object Detection
Revolutionizing Aviation Safety with Advanced AI Detection
Foreign Object Debris (FOD) within aircraft fuel tanks presents critical safety hazards including fuel contamination, system malfunctions, and increased maintenance costs. Despite the severity of these risks, there is a notable lack of dedicated datasets for the complex, enclosed environments found inside fuel tanks. To bridge this gap, we present a novel dataset, FOD-S2R, composed of real and synthetic images of the FOD within a simulated aircraft fuel tank. Unlike existing datasets that focus on external or open-air environments, our dataset is the first to systematically evaluate the effectiveness of synthetic data in enhancing the real-world FOD detection performance in confined, closed structures. The real-world subset consists of 3,114 high-resolution HD images captured in a controlled fuel tank replica, while the synthetic subset includes 3,137 images generated using Unreal Engine. The dataset is composed of various Field of views (FOV), object distances, lighting conditions, color, and object size. Prior research has demonstrated that synthetic data can reduce reliance on extensive real-world annotations and improve the generalizability of vision models. Thus, we benchmark several state-of-the-art object detection models and demonstrate that introducing synthetic data improves the detection accuracy and generalization to real-world conditions. These experiments demonstrate the effectiveness of synthetic data in enhancing the model performance and narrowing the Sim2Real gap, providing a valuable foundation for developing automated FOD detection systems for aviation maintenance.
Key Executive Impact
Leveraging synthetic data for FOD detection offers significant advancements in aviation maintenance, enhancing safety and operational efficiency.
0
Improvement in detection performance with synthetic data
0
Real-world HD images in dataset
0
Synthetic images in dataset
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Data Generation
This paper introduces FOD-S2R, a novel hybrid dataset for Foreign Object Debris (FOD) detection in aircraft fuel tanks. It combines 3,114 real-world images from a custom-built fuel tank replica and 3,137 synthetic images generated using Unreal Engine. The synthetic data provides diverse variations in lighting, object placement, occlusion, and reflections, which are crucial for enhancing model generalization in complex, confined environments like fuel tanks. The dataset addresses the limitations of existing FOD datasets that primarily focus on open-air environments, lacking the complexities of internal aircraft compartments.
6250
Total High-Resolution Images in FOD-S2R
Enterprise Process Flow
CAD Modeling of Fuel Tank (Blender)
→
Import to Unreal Engine
→
Synthetic Image Generation (Unreal Engine)
→
Real-World Image Capture (Fuel Tank Replica)
→
Automated & Manual Annotation
→
FOD-S2R Dataset
Sim2Real Transfer Learning
The study demonstrates the effectiveness of Sim2Real transfer learning. Models are initially pretrained on the synthetic dataset, which offers wide distribution coverage and scale-invariant localization cues, and then fine-tuned on a limited real-world dataset. This approach significantly improves detection accuracy and generalization to real-world conditions, narrowing the domain gap. The results show that synthetic data pretraining is more effective than reverse fine-tuning (real-first, synthetic-second), which can degrade previously learned domain-specific representations due to differences in high-frequency surface irregularities.
| Strategy | Key Benefits | Observed Performance |
|---|---|---|
| Synthetic Pretraining + Real Fine-tuning |
|
Highest performance (e.g., RF-DETR mAP50=0.931, mAP50:95=0.740, mAPs improved from 0.383 to 0.676) |
| Real Pretraining + Synthetic Fine-tuning |
|
Weaker performance (e.g., mAP50:95 dropping to 0.712) due to degradation from synthetic textures lacking real-world irregularities. |
15%
Improvement in detection performance by integrating synthetic data
Object Detection Performance
The research benchmarks several state-of-the-art object detection models (anchor-based like YOLOv5, Faster R-CNN, RetinaNet; and anchor-free like YOLOv11, YOLOv12, RT-DETR, DDQ) on the FOD-S2R dataset. It reveals a clear domain disparity between synthetic and real conditions, with synthetic data offering uniform lighting and occlusion, while real-world data introduces higher variation. RF-DETR shows strong real-domain performance overall (mAP50=0.930, mAP50:95=0.702), though small-scale performance (mAPs=0.383) is a challenge. Synthetic data pretraining significantly boosts small-object detection (mAPs) from 0.383 to 0.676.
| Model (Real Data) | mAP50 | mAP50:95 | mAPs (Small Objects) |
|---|---|---|---|
| YOLOv11 | 0.929 | 0.715 | 0.743 |
| RF-DETR | 0.930 | 0.702 | 0.383 |
0.676
RF-DETR mAPs (Small Objects) with Sim2Real
Calculate Your Potential ROI
Estimate the significant time and cost savings AI-powered FOD detection can bring to your operations.
Annual Savings
$0
Hours Reclaimed Annually
0
Implementation Roadmap
A phased approach for integrating this AI solution into your enterprise operations.
Phase 1: Environment & Data Synthesis
Establish a high-fidelity simulation environment (e.g., Unreal Engine) mirroring physical fuel tank structures. Generate diverse synthetic image data with controlled variations in lighting, object placement, and occlusion, alongside initial real-world data collection from a replica. Develop automated annotation pipelines.
Phase 2: Model Pretraining & Baseline Evaluation
Pretrain state-of-the-art object detection models on the comprehensive synthetic dataset to leverage its breadth and scale-invariant features. Establish baseline performance metrics on both synthetic and limited real-world data to identify the initial domain gap.
Phase 3: Sim2Real Adaptation & Fine-tuning
Implement Sim2Real transfer learning, fine-tuning the pretrained models on the limited real-world data. Focus on bridging the domain gap by adapting models to real-world textures, lighting inconsistencies, and unique challenges of confined spaces. Benchmark performance improvements on real-world test sets.
Phase 4: Deployment & Continuous Improvement
Deploy the optimized FOD detection models in a controlled aviation maintenance environment. Continuously monitor performance, collect additional real-world edge cases, and iteratively refine models using active learning or further synthetic data augmentation to enhance robustness and generalization.
Ready to Transform Your Operations?
Connect with our AI specialists to explore how FOD-S2R and Sim2Real transfer learning can be tailored to your specific enterprise needs. Discover a clear roadmap to enhanced safety and efficiency.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat