Page 487 - AI for Good Innovate for Impact

AI for Good Innovate for Impact 4.5: Manufacturing

Operational Mode

Morphing-BOT operates in semi-autonomous mode, autonomously navigating and mapping
pipelines using sensor fusion (IMU, Lidar, acoustic sensors). Remote control is available for
manual intervention in complex or highly obstructed areas.

Data Utilization

The datasets are utilized to train AI models for anomaly detection (cracks, leaks, deformation),
predictive maintenance prediction, asset condition rating, and navigation optimization (route
mapping and obstacle avoidance).

AI Architecture & On-Device Functions

Our intelligent inspection robot uses multimodal sensor data, including high-resolution
images, vibration signals, acoustic measurements, and IMU data, to detect pipeline anomalies.
To process this data, we employ Convolutional Neural Networks (CNNs) for image-based
defect detection, Recurrent Neural Networks (RNNs) for time-series vibration and acoustic
analysis, and Transformer-based architectures for multimodal feature fusion and anomaly
prediction. The on-device AI facilitates real-time processing and decision-making, performing
immediate obstacle detection and avoidance, defect recognition and classification without
external computing support, route adjustment for unexpected conditions, and energy-efficient
navigation. Deploying AI models directly on the robot hardware ensures low-latency, robust,
and autonomous inspection even with limited communication.

Customer Validation & Market Interest

Our intelligent inspection solution has been successfully validated through pilot projects with
four Korean municipalities: Jindo, Siheung, Jeju, and Jeonju. Municipalities expressed high
satisfaction with the robot's precise navigation, accurate defect detection, and efficient data
acquisition.

These large-scale deployments yielded over 110,000 high-quality multimodal datasets (images,
vibration, acoustic, IMU data), which were independently validated by TTA (Telecommunications
Technology Association of Korea) for accuracy, consistency, and completeness, enabling
advanced AI model development. Due to this proven performance, we are preparing overseas
PoC projects in Canada, the United States, and the Middle East. Singapore's DSTA has requested

451