Investigation of the relationship between changes in the stress-strain state and structural damage of metals and changes in the parameters of electromagnetic-acoustic signal

Problem. Ensuring the technosphere safety of industrial facilities in the oil and gas sector is complicated by
processes of hidden microstructural degradation of metal, which initiates more than 31 % of emergency situations. Traditional
nondestructive testing methods are significantly limited by the need for thorough surface preparation and ensuring stable liquid acoustic contact, which hinders the operational monitoring of pipelines and pressure vessels through insulation layers.
Aim. The aim of this study is to establish quantitative relationships between the spectral parameters of the electromagneticacoustic (EMA) signal and the stages of stress-strain state and damage of structural steels. Methods. A series of experiments
was performed on St3sp, 09G2S, and 12Kh18N10T steels under static tension on a servohydraulic system. Signal registration
was carried out using an EMA thickness gauge. The analysis is based on dynamic approach: the control zone was identified as
a linear system, the mathematical reference of which is a second-order transfer function W(p) reconstructed by the method of
moments. To automate state recognition, a transition to frequency models in the form of hodographs of amplitude-phasefrequency characteristics (APFC) was implemented. An artificial neural network based on the ML.NET library was trained
on 3 samples of 720 measurements, using the amplitudes and phases of the first 15 signal harmonics as input features.
Results. It was experimentally proved that the accumulation of structural damage causes a regular migration of the roots of
the characteristic polynomial on the complex plane and a qualitative transformation of the shape of the APFC hodographs.
The developed intelligent classifier provided an identification accuracy of the stress-strain state (SSS) stages at the level of
92.16 %, while the error in determining the numerical stress values did not exceed 3.5 %. Conclusions. The spectral parameters of the EMA signal in combination with machine learning technologies make it possible to identify reliably pre-defect
states of metal. The proposed method minimises the influence of the human factor and ensures the detection of zones of critical stress concentration before the initiation of macroscopic cracks.