Research on the low-dimensional visualization and identification method of the equipment’s conditions by cloud-based screening and hypergraph embedding

•The SKL divergence is proposed as a difference measure, which achieves the selection of sensitive features beneficial for the equipment’s condition monitoring;•The sensitive elements of the equipment’s conditions are mapped into low-dimensional space based on the hypergraph structure for visualization and identification;•The suggested method has good ante-hoc interpretability and robust performance in small sample situations. Currently, data-driven fault diagnosis and condition monitoring (CM) strategies have emerged as a preeminent approach to enhancing the reliability of mechanical equipment. However, amidst the vast industrial data landscape, the collected condition data pertaining to equipment typically encompasses numerous non-pertinent and redundant details, making the effective labeling of operating conditions challenging. Addressing this issue, this study introduces a novel method for monitoring and visualizing the conditions of the mechanical equipment, leveraging hypergraph information embedding based on the pre-screening of the feature cloud model. Initially, the study establishes a comprehensive feature cloud model tailored for capturing the condition signals of the equipment. By leveraging the analysis of inter-cloud similarities, a coarse-grained screening process is meticulously crafted to filter out non-sensitive features. Subsequently, the screening rules are meticulously projected onto the entire sample set to save the sensitive elements, and a low-dimensional projection method that seamlessly integrates hypergraph information is incorporated to visualize and identify the conditions. To validate the efficacy of this method, rigorous experiments are conducted, encompassing loading monitoring of the shearer’s rocker arm and coal-rock cutting tests. The results confirm the practicality and accuracy of the proposed method. The experimental results further demonstrate that the proposed research method successfully visually identifies the shearer’s loading and coal-rock cutting conditions in a low-dimensional space. Notably, the results show remarkable adaptability, particularly when dealing with scenarios characterized by limited sample sizes and imbalanced condition datasets.