Full-life dynamic identification of wear state based on on-line wear debris image features

• Wear state was characterized by two indicators extratcted from wear debris images.
• Binary-class model was adopted based on the stage features of wear variation.
• Full-life wear state dynamic identification model was built with SVDD method.

Wear state identification is a bottleneck for the monitoring of engine's condition due to its complex characteristics as system-dependent, time-dependent and physical coupling. Correspondingly, full-life dynamic identification of the wear state of an engine in service was investigated for real-time performance evaluation. As wear information carrier, images of wear debris carried by the cycling lubricant were sampled by an OLVF (On-line Visual Ferrograph) sensor. Two characteristic indexes including IPCA (Index of Particle Coverage Area) and EDLWD (Equivalent Diameter of Large Wear Debris) extracted from the on-line wear images, were adopted to characterize the wear state quantitatively by representing wear rate and mechanisms, respectively. A dynamic feature-matching model for real-time identification was studied comprehensively by referring to the stage features of wear state variation. Furthermore, a one-class model was built using the SVDD (Support Vector Data Description) method for categorizing statistical samples. By integrating the feature-matching and de-noising methods, a good identification was achieved with those samples. On this basis, a stage-based model for real-time wear state monitoring was built and verified with time-sequence monitoring samples from an engine bench test. The method shows potential as a promising on-line wear state evaluation tool, especially for full-life monitoring.