Diagnosis of abrasive wear of steels by machine learning in Python

Abstract

Comparative tribotechnical tests of wear-resistant equal-hard steels of the martensitic class Hardox 450, Quard 450 in comparison with ferrite-pearlite steel 09G2S were carried out. The wear resistance “ε” increased in the direction of 1.0 → 7.9 → 18.1 in steels 09G2S → Hardox 450 → Quard 450, respectively. As it turned out, after testing for abrasive wear of samples made of Quard 450 steel with maximum wear resistance, traces of martensitic transformation were found on the friction sites. Using machine learning in Python, graphs of fluctuations of friction moments in time of the studied steels were constructed, their smoothing (averaging) by the Hannah method was carried out. The Fourier transform was performed, the amplitude-frequency characteristics of the friction moments were calculated and frequency spectrograms were constructed. It is shown that a decrease in the amplitude of the sinusoids of the first (main) harmonic occurred in the direction of 3.40 → 2.15 → 1.40 in steels 09G2S → Hardox 450 → Quard 450, respectively. Unlike the 09G2S and Hardox 450 steels, in the Quard 450 steel with maximum wear resistance, the second harmonic was split into two with amplitudes of 0.85 and 0.70. The spectra of two exciting forces (abrasive wear and structural-phase (martensitic) transformation) were found in it, strengthening the surface in the contact zone and increasing its wear resistance.