A novel texture-based damage detection method for wire ropes

Abstract

Visual method is a potential way to intuitively grasp the surface damage state of wire ropes to guarantee its reliability and safety. However, the various appearance characteristics of ropes are seriously affected by noise and effective inspection methods are lacking; therefore, efficient preprocessing and damage detection methods are urgently needed. Aiming at the influence of illumination, feature dimension, and the performance of different classification algorithms, this paper presents a novel method for the damage detection of wire ropes based on texture features, called WR-LBPML. To uniform illuminate and highlight textures of the rope images, different homomorphic filtering algorithms were compared and applied. A total of 59 texture features were extracted from rope surface images using uniform local binary patterns (u-LBP) operators. Due to the redundancy of the dataset, the features were processed using principal component analysis. Finally, we conducted a comparative study of three machine learning algorithms. Extensive experiments and comparisons of rope damage detection indicated that the proposed method can achieve optimal performance (accuracy: 93.3%, efficiency: 0.04 ms per sample) using a support vector machine with 10 features extracted by u-LBP operator, in which the images were filtered by a Gaussian low-pass filter with a block size of 1.