WIRELET 3: DEEP LEARNING MODELS FOR THE AUTOMATIC ANALYSIS OF ROPE SURFACE DEFECTS
Wirelets have been previously introduced to efficiently extract rope event anomalies within magnetic signals. Still, the final assessment of related signatures is left to the inspector’s interpretation with underlying time and reliability constraints. Rope nowadays applications become more and more stringent and new technologies like artificial intelligence offer an opportunity to provide a robust and consistent analysis of rope condition. Deep learning is a computational model composed of multiple processing neurons layers aimed to learn representations of signal data with multiple levels of abstraction. Current work introduces and compares several deep convolutional networks aimed to automatically classify rope surface defects from the knowledge of their Wirelet signature. Such deep learning model which is massively parallel is first expected to bring a breakthrough in the tedious task of evaluating rope condition and second to enable real time analysis of magnetic rope signals.
Keywords: NDT, MRT, VI, Wirelet, Deep learning, Wire rope condition analysis.
Author(s): S. Pernot, O. Reinelt, M. Hanimann