In mine hoisting systems, fault diagnosis of rigid cage guides is of great importance for ensuring equipment operation safety. Traditional methods based on one-dimensional vibration signals face challenges such as insufficient feature extraction ability, loss of spatial structural information, and low diagnostic accuracy when dealing with nonlinear, high-dimensional, and noisy signals. To address these issues, a fault diagnosis method based on planar kernel perception and multi-scale structural embedding (PKP-MSSE) was proposed. The method introduced kernel-based similarity modeling and a multi-scale structural embedding mechanism. A multi-scale kernel space structural representation was constructed to capture both local and global geometric relationships of samples. Discriminative features were extracted in two directions of planar images to preserve spatial structural information and avoid the loss caused by vectorization in traditional approaches. Local preservation and global divergence objective functions were jointly optimized to achieve effective dimensionality reduction and improve fault identification accuracy. Experimental results show that PKP-MSSE achieves excellent recognition rate and robustness on a rigid cage guide simulation dataset. The method demonstrates strong noise resistance and enhances class separability of fault features. It provides a reliable approach for fault identification of rigid cage guides.