Many types of defects may occur in the manufacture of seamless steel tubes. According to the production process, the types of defects will also be different. It is precisely because of these potential defects that the strength of pipelines decreases greatly in the course of use and fails prematurely, even leading to unpredictable losses. Therefore, it is of far-reaching significance to take effective measures to real-time monitor the defects in the production process and identify the types of defects in order to improve product quality, improve process and make rational use of resources.


For a large number of seamless steel tube production lines, it is obviously not in line with the requirements of modern large-scale production to identify defect types by manual intervention. In order to realize on-line automatic detection of defect type recognition, many artificial intelligence algorithms have shown strong vitality in today's manufacturing informatization, such as genetic algorithm, expert system, empirical heuristic algorithm, signal processing and pattern recognition, adaptive learning, artificial neural network, etc. Artificial neural networks such as networks have been highly anticipated. However, the disadvantage of ANN is that it needs a large number of training samples, easy to fall into local minimum point, slow convergence speed and poor generalization ability of small samples, which makes it difficult to obtain a large number of samples (for example, for power plant boilers, nuclear reactor containers, aeroengine rotors, marine turbochargers, etc.). Predicted. As an intelligent optimization algorithm, particle swarm optimization (PSO) has the characteristics of global optimization.


In order to identify the type of seamless steel tube defects by clustering analysis based on particle swarm optimization, the characteristic parameters of defects must be extracted first. Ultrasound nondestructive testing technology can only obtain the information of echo signal, which is very rich. Information. Therefore, feature extraction can be performed based on defect echo signal.