A novel vision-based method for loosening detection of marked T-junction pipe fittings integrating GAN-based segmentation and SVM-based classification algorithms

Pipes connected by threaded joints are widely applied to transmit fluid and gas in many industries. Loosening in threaded joints causing the problem of fluid or gas leakage may induce disastrous consequences. Regular loosening detection of threaded pipe fittings cannot be overemphasized. In engineering applications, marked bars are drawn on the threaded pipe fittings to indicate loosening/tightening state. Traditional visual inspection requires laborious workloads. An automated method for loosening detection of marked threaded pipe fittings is still lacking. In this paper, a T-junction threaded pipe fitting was chosen as the research object. We proposed a novel vision-based method to conduct the loosening detection of three threaded joints in a T-junction pipe fitting for the first time. Our method contains three integrated modules. A new generative adversarial network-based segmentation module is constructed to accurately segment marked bars first. Then skeleton algorithm is used to extract the center lines of segmented marked bars and three sensitive angle features for loosening detection are constructed. Last, these features are fed into support vector machine-based classification module to differentiate the loosening state from tightening state. The experimental results indicated that the average segmentation accuracy denoted by dice similarity coefficient was 0.96 and the average detection accuracy was 94.7% based on our method. Moreover, our proposed method has been validated having a strong loosening detection ability in different environments, and great potentials in engineering applications.