Neural network analysis of ultrasonic flaw detection data of main gas pipelines using a hardware and software DVU system

The paper considers the problem of determining the zones exposed to corrosion, obtained during non-destructive testing using ultrasonic flaw detection methods of main gas pipelines and the hardware and software system of the DVU (in-tube ultrasonic flaw detector). It is proposed that the processing of a large amount of data obtained as a result of control using the DVU system is carried out using a neural network model of data analysis. A family of convolutional neural networks was chosen as the main neural network. The processed data were divided into training, validation and test sets. The data within the samples were mixed, which allowed the learning process to be made more qualitative. As the testing analysis showed, the proposed approach makes it possible to determine the defect zones of gas pipelines with sufficient accuracy.