1 of 1

AUTONOMOUS NETWORKED ROBOTS FOR ADVANCED MANUFACTURING

11th Annual COE Graduate Poster Presentation Competition

Student (PhD): M A Muktadir

Advisor: Dr. Sun Yi

Cross-Disciplinary Research Area:

Additive Manufacturing

Recent advance in advanced manufacturing, including 3D printing technologies, enables different types of manufacturing systems (additive and subtractive) to produce complex objects. However, there still exist many challenges in the production processes. For example, one of the main challenges is safely handling products after production without damage and material loss in an autonomous manner in remote environments where human access is limited due to radioactive materials. Robotics manipulators with mobile robots have been the right choice for these complex works. Depending on the situation of the manufacturing processes, the gripper fingers need to be modified, and the robot’s types should vary; it may be a fixed robotic arm, or a robotic arm placed on a mobile ground robot. Such robot systems can minimize human involvement to avoid insecurity and risk of injury.  Also, unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs) have to operate under complex or unorganized places, such as rough surfaces and windy areas. For UGVs, certain surface types of ground must reach their destination. UAVs cannot move in a complex place with limited flying time and stability limitations. SPOT, a robotic dog, would be an excellent alternative for the AM environment, to inspect the production process on time repeatedly without any operator involvement as it has the capabilities to charge itself and complete a mission in a specific way in a timely manner.

ACKNOWLEDGEMENT

This research is funded by the NCDOT.

Abstract

Methodology

Conclusion & Future work

Data Collection

Implementation

• As the defects are not the problem only on pipes, this research is also will be implemented in manufacturing, such as additive manufacturing or 3D printing.
• The implementation stage is yet not finished, and more training needs to be done with more data to get a concrete result.
• In the future, mapping (simultaneous localization and mapping (SLAM)) have to be done to get the exact location of the defect to help the human operator remotely in the modern manufacturing process.

​

Fig1: Flow chart of Methodology.

Postprocessing of data

Training and result

Evaluation Metrics of the trained model:

​

Precision =

True Positive/ (True Positive + False Positive)

​

Recall =

True Positive/ (False Negative + True Positive )

​

F Score =

(2 X Precision X Recall ) / (Precision + Recall)

​

Fig2: Data collection with a 3D scanner.

Fig3: Postprocessing and classifying the different types of defects.

LIDAR CAMERA

Fig4: ML training result.

Fig5: SPOT robot with a LIDAR camera.