Towards high-speed elevator fault diagnosis: A ParallelGraphNet driven multi-sensor optimization selection method

Abstract

Due to the complexity of the high-speed elevator operation system, a single sensor may not fully capture its fault characteristics comprehensively. Moreover, manual sensor placement limits fault feature reflection and fails to exploit sensor spatial advantages, hindering further development. To address these limitations, an optimization method for selecting the positions of multiple sensors is proposed by identifying those offering global information gain to make decisions. Additionally, the ParallelGraphNet (Graph Convolutional Network with parallel graph construction layers and parallel pooling layers) is introduced, which parallelly constructs node graphs and segmented graphs, deeply integrating the spatial and temporal characteristics of the collected signals. Moreover, two parallel pooling layers are introduced after the graph convolutional layer to better extract multi-scale features while preserving the original features. Experiments were conducted on the constructed high-speed elevator platform and publicly available gearbox datasets to validate the effectiveness of the proposed method. Comparative experiments were carried out from multiple perspectives, confirming that the proposed method can effectively select sensors to improve sensor acquisition efficiency and fault diagnosis accuracy. The code library is available be https://github.com/FELIZHANG/TWO_LAYER_GRAPH/tree/main.