基于时空图网络的水泵装备小样本故障鲁棒识别OA

Pumps are common equipment in modern industrial and agricultural production,and their operating environments are often characterized by significant noise,which complicates data-driven fault identification.Research indicates that graph neural networks have a distinct advantage in extracting fault features from noisy signals.By transforming one-dimensional signals into graph-structured data,hidden fault information within the signals can be revealed.However,the reliability of fault identification largely depends on the construction strategy of the input graph.In response to this,this paper proposes a robust graph construction strategy and feature extraction method that is resilient to noise.The graph construction phase embeds node information using short-time Fourier transform and establishes edge relationships through cosine similarity,ensuring that the feature space within the samples is adequately described.Next,an optimized graph pruning method is proposed,which enhances the noise robustness of the input graph while also reducing computational pressure.Furthermore,an improved GraphSAGE network model is employed to perform layer-wise feature extraction on the constructed input graph,and a SoftMax classifier is used to assign fault labels to each sample.Data collection and method validation are conducted using an axial flow pump test platform,demonstrating the reliability of the proposed method for multi-component fault identification in noisy environments.