High Accuracy Pipe Keyway Angle Identification Using CNN
Prediction of keyway angle precisely inside a pipe in different environmental conditions is challenging in manufacturing automation. Machine learning is one of the most prospective, efficient solutions to deal with this problem. The main objective of this research is to correctly predict keyway angle (angle error < 1 degree) in different lighting conditions. Previous research utilizing General Regression Neural Network (GRNN) to serve this purpose restricts the keyway angle within 25 degrees and needs multiple steps to align it. Convolutional neural network (CNN) based hyper tuned VGGNet-16 inspired architecture solves this keyway alignment problem. The framework handles pipe keyway alignment precisely without rotation angle limitation and satisfies the required angle error tolerance level in changing lighting conditions. The alignment of pipe keyway plays a vital role in industrial productions to produce a quality product. In the beginning, pipe images are captured at different rotation angles using a machine vision (MV) tool and a Cognex In-sight 7000 2D Smart Camera. Next, the keyway angles at different locations are determined using this tool’s ‘PatMax’ function. After that, image processing and machine learning are implemented to predict the keyway angle errors. A conventional image processing algorithm utilizes Hough Circle Transform theory to find the keyway position in a circle and determine the keyway angle. Then the focus turns to establish a machine learning model to gain a more precise prediction that satisfies the pipe alignment requirement mentioned earlier for the pipe handling process.
keyway, angle, CNN, reflection, distortion
Sarker, S. (2021). High accuracy pipe keyway angle identification using CNN (Unpublished thesis). Texas State University, San Marcos, Texas.