Classifying Wear Particles Based on Texture Analysis

Wear particles are produced in all machines with interacting mechanical parts. These particles amalgamate with the lubrication oil of the machinery and occur in varying quantities, sizes, compositions, and morphology. The generated wear particles when examined and analyzed provide critical information about the machine’s condition. Analysis of wear particles is essential to identify wear failure modes of different components leading to malfunctioning or even breakdown of the machinery. Experts in the field make use of the analyzed information to ensure the safe, efficient, and economic operation of the machinery. The characteristic of wear particles is described by six morphological attributes of color, edge details, shape, size, texture, and thickness ratio. Manual and traditional methods are used for diagnosing wear conditions; however, the use of these six attributes is the basis of constructing an image analysis system to augment wear judgments. The particle shape is an important attribute, which is investigated in this paper. Various parameters are used to extract shape features to identify six types of wear particles that include severe sliding, cutting-edge, non-metallic, fatigue, water, and fiber wear particles. The proposed methods are either statistical or learning-based techniques, which include a gray-level co-occurrence matrix, the histogram of gradient magnitude, local binary pattern, Tamura features, image moments, and a deep learning algorithm. The result of the investigation shows that the deep learning technique performs better compared to other techniques. The deep learning technique gives an average classification rate of 91%, whereas the highest accuracy from statistical methods is approximately 65%.