An Adaptive Refined Composite Multiscale Differential Symbolic Entropy Rooted in LSC-SAO and Its Application in Fault Diagnosis

Accurate fault diagnosis of rotating machinery is critical for ensuring the reliability of the energy, industrial, and transportation sectors. However, conventional methods face significant challenges, including the susceptibility of the Snow Ablation Optimizer (SAO) to local optima, the instability of Multiscale Differential Symbolic Entropy (MDSE) with short time series, and the non-adaptability of Support Vector Machine parameters. To address these issues, this study proposes a parameter-adaptive fault diagnosis framework integrating an improved SAO with Adaptive Refined Composite Multiscale Differential Symbolic Entropy (Adaptive-RCMDSE). First, the Logistic Sine Cosine strategy (LSC) is introduced to enhance SAO’s global search capability, forming the LSC-SAO algorithm. Subsequently, an Adaptive-RCMDSE method is developed wherein LSC-SAO optimizes the control parameter to significantly improve feature stability for short time series. Furthermore, an Adaptive Support Vector Machine (Adaptive-SVM) model is constructed, employing LSC-SAO to automatically tune the penalty factor and kernel parameters for precise fault identification. Finally, validation is performed on gearbox, ball bearing, and axle box bearing datasets. Results indicate that the proposed method achieves superior diagnostic performance, with average accuracies of 99.70%, 99.29%, and 99.28%, respectively, outperforming existing methods. This work provides an effective and robust solution for intelligent health monitoring of rotating machinery.