刮板输送机是一种煤矿井下用的煤炭输送设备，在煤矿生产中具有重要作用。恶劣工作环境及长期使用磨损导致刮板输送机性能逐渐退化，故及时掌握刮板输送机健康状态极为关键。为克服传统方法在刮板输送机整机健康状态识别过程中存在的部件强耦合性关系难以提取融合及健康指标构建人工参与过多、易受异常值影响的问题，提出一种基于多种图结构信息融合的刮板输送机健康状态识别方法。利用自注意力机制（SA）与标准化流（NF）共同优化的变分自编码器（VAE）无监督地自动构建刮板输送机健康指标，降低了刮板输送机健康指标构建中对人工经验的依赖，同时有效拟合了健康指标的隐式分布，克服了监测数据中存在的异常值影响健康指标构建的问题；提出了一种多种图结构信息提取方法，提取刮板输送机先验图结构及相似性度量图结构，全方位显式地表达了多部件之间的耦合关系；提出了一种多种图结构信息融合方法，利用多个图注意力网络（GAT）有效提取并融合刮板输送机的多种图结构信息。在采集的刮板输送机真实状态数据中进行试验，结果表明，模型识别准确率可达98.60%，宏平均F₁（Macro-F₁）值可达96.81%，该方法为刮板输送机的健康状态识别提供了一种新的可行途径。

Scraper conveyors are essential coal transportation equipment in underground coal mines, significantly impacting mine production. However, the harsh working environment and long-term use lead to wear and tear, degrading their performance. Therefore, timely monitoring of scraper conveyor’s health status is extremely critical. To address the limitations of traditional methods, which struggle with strong component coupling and require excessive manual intervention, a novel method for identifying health status of scraper conveyors is proposed. This method utilizes a Variational Autoencoder (VAE) co-optimized with Self-Attention (SA) and Normalizing Flow (NF) mechanisms to automatically construct health indicators without supervision, effectively fitting the implicit distribution of the indicators and overcoming the influence of outliers. Additionally, a method fusing multiple graph structure information is introduced, using multiple Graph Attention Networks (GAT) to extract and integrate this information. Experiments with real-world data from the scraper conveyor show that the model’s indentification accuracy achieve up to 98.60% and macro-average F₁ scores up to 96.81%. This approach offers a novel and feasible solution for health status identification of scraper conveyors, with significant practical value.