Recognition of defects in hoisting ropes of metallurgical equipment by an optical method using neural networks

Abstract

The article considers the problem of controlling the condition of hoisting ropes. The existing methods and control systems for hoisting ropes have been analyzed, according to the results of which it was found that the magnetic control systems can detect defects and damage only after they have reached the stage of significant damage. The use of optical inspection methods and pattern recognition technology using convolutional pyramidal neural networks to detect defects in hoisting ropes of metallurgical equipment was considered. It describes the layout solution of the control system of hoisting ropes using the optical method with the angle mirror converter, which allows controlling the entire surface of the hoisting rope using a single digital camera, which will reduce the cost of the hardware. Analysis of research into methods of processing visual information in order to detect defects of control objects showed the prospects of using convolutional neural networks. During the study, neural networks were trained on images of ropes with different characteristics based on the software library TensorFlow. The results of assessing the reliability of detecting defects and damage to hoisting ropes noted the influence of lighting and the distance at which the object under test is located on the reliability of recognition of rope defects and faults. An image preprocessing algorithm with removal of the background component, filtering and scaling of the image has been proposed to improve the reliability of hoisting rope defect and damage detection. A qualitative assessment of applying neural networks to determine the type of defects has been obtained, and the possibility of detecting them with an extraction coefficient of 0.80–0.89 using the proposed algorithm has been shown.

Topics & Concepts