Tiny Machine Learning Empowers Climbing Inspection Robots

Tzu-Hsuan Lin, Chien-Ta Chang, and Alan Putranto, Tiny machine learning empowers climbing inspection robots for real-time multiobject bolt-defect detection. Engineering Applications of Artificial Intelligence, 2024. 133: p. 108618. 50 days' free access to our article. Anyone clicking on this link before July 05, 2024 https://reurl.cc/WxxKOZ

Summary and Highlights:

We are excited to introduce our groundbreaking research on real-time bolt defect detection using tiny AI cameras integrated into climbing robots. This innovative system enhances the safety and efficiency of inspecting steel structures by automating the detection of bolt defects with remarkable precision. Below are the key highlights of our research:

Tiny Machine Learning (TinyML): Our system employs the cutting-edge Faster Objects, More Objects (FOMO) algorithm, optimized for edge computing on microcontrollers, enabling efficient real-time processing with minimal hardware requirements.

• Tiny Machine Learning (TinyML): Our system employs the cutting-edge Faster Objects, More Objects (FOMO) algorithm, optimized for edge computing on microcontrollers, enabling efficient real-time processing with minimal hardware requirements.

High Accuracy: The bolt defect detection system achieves an impressive accuracy of 82%, with precision and recall values ranging from 0.57 to 0.89 and 0.67 to 0.87, respectively. This robust performance is validated across diverse environmental conditions.

• High Accuracy: The bolt defect detection system achieves an impressive accuracy of 82%, with precision and recall values ranging from 0.57 to 0.89 and 0.67 to 0.87, respectively. This robust performance is validated across diverse environmental conditions.

Climbing Robots: Our magnetic climbing robots are designed to navigate complex steel structures autonomously, overcoming the limitations of manual inspections and ensuring thorough coverage even in hard-to-reach areas.

• Climbing Robots: Our magnetic climbing robots are designed to navigate complex steel structures autonomously, overcoming the limitations of manual inspections and ensuring thorough coverage even in hard-to-reach areas.

Real-Time Inspection: The integration of FOMO allows our system to identify various bolt conditions, including normal, loose, and missing bolts, with an F1 score of approximately 75%, outperforming other models like MobileNetV2 SSD and YOLOv5.

• Real-Time Inspection: The integration of FOMO allows our system to identify various bolt conditions, including normal, loose, and missing bolts, with an F1 score of approximately 75%, outperforming other models like MobileNetV2 SSD and YOLOv5.

Efficiency and Low Power Consumption: The FOMO model operates efficiently on low-capacity microcontrollers, with less than 0.1 MB of flash memory and 893.8 KB of RAM, making it ideal for real-time applications.

• Efficiency and Low Power Consumption: The FOMO model operates efficiently on low-capacity microcontrollers, with less than 0.1 MB of flash memory and 893.8 KB of RAM, making it ideal for real-time applications.

Scalable and Cost-Effective: Our system offers a scalable and cost-effective solution for enhancing the safety and durability of steel structures, setting new benchmarks for real-time inspection technology.

• Scalable and Cost-Effective: Our system offers a scalable and cost-effective solution for enhancing the safety and durability of steel structures, setting new benchmarks for real-time inspection technology.

This research represents a significant advancement in the field of structural inspection, leveraging the power of TinyML and autonomous robots to ensure the integrity and safety of critical infrastructure. Join us on Hackster.io to explore the full potential of this innovative technology and its applications in real-world scenarios.