Time-shift multiscale slope Lempel-Ziv complexity: a novel framework for feature extraction and fault diagnosis in pump valve signals

This study addresses the challenges of fault diagnosis in industrial pumps by introducing a novel framework that synergistically combines advanced feature extraction with efficient multi-class classification. We propose time-shift multiscale slope Lempel-Ziv complexity (TSMSlopLZC) to overcome key limitations of existing methods: it integrates the amplitude-based symbolic dynamics of slope entropy (SlopEn) into Lempel-Ziv complexity (LZC) to eliminate its reliance on binary mapping, while a novel time-shift multiscale coarse-graining process ensures feature stability and mitigates information loss. For classification, we develop a binary tree least squares twin support vector machines (BT-LSTSVM) that achieves efficient multi-class fault diagnosis through an optimal binary tree structure, significantly reducing computational complexity while maintaining high accuracy. Experimental results on centrifugal pump and diaphragm pump valves demonstrate the superiority of our approach, showing accuracy improvements of 10.83% and 31.94% respectively over leading methods. The proposed framework offers a robust and efficient solution for high-precision fault diagnosis in complex industrial environments.