ABSCTRACTWear state is an important indicator of machinery operation condition that reveals whether faults are developed and maintenance should be scheduled. Among the available techniques, vision based on-line monitoring of wear particles in the lubricant circuit is preferred, where threedimensional particle characterizations can be obtained for wear mode analysis. This paper presents the application of an imaging system, which captures wear particles in lubricant flow, and the development of image processing procedures for multi-view feature extraction. In particular, a framework including background subtraction, object segmentation, and debris tracking was adopted. Particle features were then used in a comprehensive morphological description of wear debris. Experiments showed that the system is able to produce a feasible and reliable indication of wear debris characteristics for machine condition monitoring.
INTRODUCTIONWear is one of the most inherent and noticeable phenomena in machines having components in contact and relative motion. The health status of a machine can therefore be investigated by monitoring the degree of wear on the components. However, this task often requires shutting down or disassembling the machine, thus imposing additional operation costs as well as loss of production. On the other hand, wear debris particles are produced during the wear process, and these particles can be an indicator of the morphology and degree of wear. Based on the fact that most machines have a lubrication system and wear debris particles are carried through the lubricant circuitry, on-line monitoring of wear particles is possible by securing a representative fluid sample for microscopic analysis, in particular, ferrography. This vision-based technique is very attractive as it is not necessary to interrupt the operation of the machine (Kumar, et al. (1); Wu, et al. (2); Wang and Wang (3)).Ferrography as a vision based technique has been regarded as an effective method for the estimation of wear particle concentration as well as their characteristics (Zhang, et al. (4), Wang, et al. (5), Li, et al. (6)). However, particle overlap is a commonly reported issue with ferrography, making it difficult to examine individual wear particles. Furthermore, since conventional ferrography makes use of only single-view images, three-dimensional (3D) particle features cannot be conveniently provided. Thus, the needed spatial information, such as surface roughness and particle thickness, is not available to fully study wear mechanisms and reveal wear conditions with better precision (Yuan, et al. (7)).Research efforts have also been directed toward extracting particle features in 3D (Stachowiak and Podsiadlo (8); Tian, et al. (9); Yuan, et al. (10); Podsiadlo and Stachowiak (11)).3 Among these attempts, stereo scanning electron microscopy (SEM) (Stachowiak, et al. (12)), laser scanning confocal microscopy (LSCM) (Peng, et al. (13); Peng and Tomovich (14)), and atomic force microscopy (AFM) (Wang, et al. (15)) were developed to...
