A quantitative recognition method based on VMD,CNN,and SVM for detecting wire rope breakage damage inside elevator steel belts

Abstract

Breakage of the wire rope inside the belt, a very common damage, affects the performance of the belt and the safety of the elevator. Additionally, serious challenges remain in reducing the noise of damage signals, selecting effective features, and quantitatively detecting damage. Based on variational modal decomposition (VMD), convolution neural network (CNN), and support vector machine (SVM), a new method is proposed for diagnosing steel belt wire rope damage. First, a low-pass noise reduction algorithm is constructed using VMD for reducing the noise of the damaged signals. Then, the model of particle swarm optimization (PSO) of CNN parameters is established for effective extraction of features. Next, based on the pre-extracted features as data samples, a multiparameter optimization model of particle swarm combined with SVM is developed for quantitatively recognizing different damage degrees of wire ropes inside steel belts. Finally, the effectiveness of the method is verified by the constructed wire rope damage simulation test bench, and the recognition accuracy reached 98.89%. The results verify the feasibility of the method, which outperforms the traditional methods of noise reduction, feature extraction, and recognition in the diagnosis of broken wires.