Determining parameters of surface defects in the base metal of pipelines using results of complex diagnostics

We discuss issues of determining surface operational defects parameters using results of complex diagnostics including ultrasonic, eddy current and visual non-destructive testing methods. We note that the visual method is implemented using a television camera with a computer vision function and a laser triangulation sensor. The study presents a data set in which the non-destructive testing results are used as input variables, with the depth of planar and volumetric defects and the width of volumetric defects used as target variables. The study also assesses the degree of influence of various non-destructive testing results on the determination of target variables. Models are trained using various algorithms. Finally, the models based on gradient boosting are found to be optimal for all target variables. We propose an algorithm for combined processing of the results of complex diagnostics in which the obtained models are used. The accuracy of the proposed algorithm determined using the RMSE metric is found to be 0.011 mm.