Utilizing Zero-Crossing Rate (ZCR) for Acoustic Leak Detection in Pipelines: From Empirical Models to a Physically Grounded DSP Pipeline

This paper presents the physical and mathematical justification for incorporating Zero-Crossing Rate (ZCR) as a spectral verification layer in the RIVIXI AI v1.3 acoustic leak detection pipeline. We analyze the limitations of NLP-inspired text tokenization of acoustic waveforms and propose a rigorous DSP approach grounded in the physics of hydrodynamic noise. Hydrodynamic leak noise exhibits continuous broadband high-frequency turbulent emissions with elevated ZCR, while mechanical impacts produce low-frequency structural resonance with low ZCR. A dual-factor decision logic combines cross-correlation Z-score (Z ≥ 8.5) with ZCR threshold verification (ZCR > 0.06) to distinguish true leaks from false alarms. Evaluated on 113 industrial field recordings (30 leaks, 72 leak-free, 11 noisy), the hybrid ZCR+Z-score pipeline reduced the False Positive Rate from 90.36% to 77.11% (−13.25% absolute), prevented 11 false alarms, while maintaining 93.33% Recall. Reference signal analysis on synthetic benchmarks (leak hiss, hammer strike, speech/music) confirmed correct suppression of all non-leak sources.