A Health Status Recognition Method for Scraper Conveyors Based on Improved VAE-GCN
The current working environment of scraper conveyors is harsh,and their performance degrades continuously,making it crucial to accurately identify the health status of scraper conveyors in a timely manner.Traditional methods are prone to errors in constructing health indicators for scraper conveyors due to the influence of outliers and excessive manual involvement.Currently,methods such as convolutional neural networks(CNN)fail to deeply extract the correlation between samples,leading to poor accuracy in sample state identification.To address these issues,this paper proposes a health status identification method for scraper conveyors based on an improved variational autoencoder-graph convolutional network(VAE-GCN).Firstly,the raw signals of samples are converted into Gramian Angular Field images to extract sample state information.A CNN and self-attention mechanism(SA)optimized variational autoencoder(VAE)are used to build an unsupervised health indicator construction model,establishing the probability distribution of health indicators,which overcomes the problem of being easily influenced by outliers and excessive manual intervention in health indicator construction.The normalized mutual correlation coefficient is used to calculate the similarity between pairwise sample image information,constructing a sample association graph structure and establishing the correlation between samples.Finally,the graph convolutional network is utilized to complete the extraction of associated information between samples and the identification of the health status of scraper conveyors,overcoming the difficulty in extracting inter-sample correlations that leads to poor accuracy in sample state identification.In conclusion,the experiments and comparative analysis show that the model's health status identifica-tion accuracy can reach 98.20％,verifying the accuracy and effectiveness of the proposed method.The article provides a new method and technology for the health status identification of the entire scraper conveyor,laying a theoretical foundation for subsequent predictive maintenance of scraper conveyors.
scraper conveyorhealth indicator constructionvariational auto-encodergraph convolutional networkhealth state identification
NETL
NSTL
万方数据
