Robot Learning for Object Handling in an Unstructured Environment
MetadataShow full metadata
In the oil exploration industry, perforated pipe assembly can be a prolonged process in the manufacturing environment. A pipe keyway must be aligned to successfully assemble the perforated outer and inner pipes. However, the current method uses vision devices to rotate a pipe multiple times, eventually rotating to the angle that meets the requirement, which is time-consuming, leading to a lack of productivity. Therefore, the purpose of conducting this research is to establish an automatic rotating angle correction method such that the keyway can be aligned by only rotating a pipe once.
The system executes a series of processes: recognizing a pipe, picking it up, detecting the keyway, and rotating it to the desired orientation using only a single rotation. A General Regression Neural Network (GRNN) model predicts the actual robot rotation angle needed for correct orientation. The robot will rotate the pipe using the predicted rotation angle. After rotation, the deviation from the desired keyway angle must be less than the given threshold.
This research is of importance in the application of machine vision (MV) in industrial production. In this thesis research, the pipe keyway alignment problem is addressed using a 2D machine vision method as well as the GRNN algorithm. The proposed method is tested using a pipe handling process. A steel pipe with a keyway is to be placed in a random orientation. As the keyway must be aligned in the following manufacturing process, a robot is used to rotate the pipe to the correct orientation by applying the machine learning algorithm. The experiment was set up and used to test the proposed machine learning method. Also, for easier automatic picking up the pipe, we implemented a 3D machine vision recognition procedure. Compared with the current method, the proposed method allows the robot to only needs to rotate a pipe once to align the keyway. Hence the proposed method can greatly increase the manufacturing efficiency and reduce manufacturing cost.
The thesis introduces the experimental system, explains the theories and the methodologies, describes the procedure of the experiment, and arrives at a result. The system uses an industrial robot ABB IRB 4400; Cognex DS1300 3D Displacement Sensor; Cognex In-sight 7000 2D Smart Camera, and a computer with the GRNN algorithm.