Fault analysis and diagnosis of ZD6 switch machines based on permutation entropy-driven CEEMDAN and multiscale topological dynamical analysis

The safe and reliable operation of railway switch machines is of fundamental importance to railway transportation systems. However, vibration signals collected from these systems typically exhibit strong nonlinearity and non-stationarity, posing significant challenges for fault analysis and diagnosis. This study focuses on fault analysis and diagnosis of ZD6 switch machines using vibration signals collected under different operating conditions. To address this issue, a multiscale topology-driven framework is proposed to characterize the dynamical properties of vibration signals and their relationship with fault conditions. Specifically, the raw vibration signals are first decomposed into intrinsic mode functions (IMFs) using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), in which the noise amplitude is adaptively determined according to permutation entropy (PE). These components are subsequently reconstructed into high-frequency and low-frequency modes, representing distinct dynamical scales of the system. To further reveal the underlying structural characteristics, topological data analysis (TDA) is employed to transform each component into phase-space point clouds and persistence diagrams, enabling multiscale characterization of fault-related topological and dynamical variations. The resulting Betti-curve-based topological descriptors are subsequently utilized for fault characterization and classification. The results demonstrate that the proposed method effectively distinguishes topological characteristics across different frequency scales and achieves an average accuracy of 97.04% ± 1.16% under stratified 5-fold cross-validation. These findings indicate that the proposed approach provides an interpretable framework for topological characterization and fault identification across different operating conditions in ZD6 switch machines.