Fault Detection of Train Center Plate Bolts Loss Using Modified LBP and Optimization Algorithm

ABSTRACT

This paper presents a novel approach to fault detection of train center plate bolts loss based on Local Binary Patterns (LBP) and Gabor-GA optimization theory. A modified LBP operator including the positive-negative sign and magnitude components of local gray difference is introduced to extract much more texture information. Multi-channel Gabor wavelet with different scales and orientations is applied on the images to create new representations in the spatial domain. Then, the weight of each Gabor channel can be optimized through the Genetic Algorithm (GA) to obtain enhanced features. Finally, the weighted features are concatenated together and delivered into Support Vector Machine (SVM) network for classification. Experimental results show that the new approach can be an effective and reliable measure for monitoring fault.

Keywords:

Fault detection, genetic algorithm, local binary patterns, support vector machine, train center plate bolts loss.