Monitoring of friction-related failures using diffusion maps of acoustic time series

Friction-related failures are a frequent cause of catastrophic damage in mechanical systems, resulting in significant repair expenses and/or production delays in industry worldwide. Scuffing is one of the most concerning friction-related failures mechanisms. It is characterized by an abrupt failure mechanism, which makes its prediction unresolved today. Therefore, any in situ and real-time monitoring system of such failures is of great demand and this contribution is supplement to and enrichment of existing studies on this topic. We propose a probabilistic analysis of acoustic signals synchronously recorded with the coefficient of friction during real-life experiments. They, simulate a specific industrial operating condition of a cam, cylinder pairs and gear teeth. The experimental conditions are a reciprocating movement of a stainless steel cylinder (counter-body) against a grey cast iron body. The counter-body was under a constant load and had a line contact with the body. The acoustic signals from one-directional strokes are selected and substituted by their wavelet spectrograms. Then, diffusion maps are computed over several spectrograms. Their corresponding Fourier spectra are used as features of the surface state and so of the upcoming failures. The real surface state and damage were identified from the dynamical behavior of the coefficient of friction. Based on our approach, it was found that the friction behavior can be divided into four confidence intervals, three for the steady-state regime and one for the scuffing regime. The transients between the confidence intervals allow estimating the proximity of scuffing and this is the major achievement of this work. The first transient took place between 5 min 40 s. - 47 min 30 s. prior to scuffing. The second one took place between 50 s. and 13 min 40 s. The proposed in situ and real-time monitoring system has significant industrial potential since it is built as a client - server architecture and requires minimum modifications of existing commercial industrial machines. (C) 2020 Elsevier Ltd. All rights reserved.