Asset-agnostic end-of-line NVH diagnostics through cross-power order analysis: A unified framework for automotive component screening

This study presents a unified End-of-Line (EOL) Noise–Vibration–Harshness (NVH) diagnostic framework that integrates three complementary analytical modules—Cross-Power Order Health Index (CPO-HI), Domain-Adaptive Quantile Thresholding (DAQT), and Multi-Reference Synchronous Coherence (MRSC). Unlike amplitude-only or manually tuned spectral methods, the approach consolidates order-domain cross-spectral information, data-driven acceptance bands, and coherence-weighted filtering to deliver objective and cycle-time-neutral screening. The framework is evaluated on two heterogeneous automotive subsystems—alternators and steering racks—using near-field acoustics with electrical and speed references. Explicit mathematical definitions, signal-processing parameters, and statistical analyses are provided to substantiate reproducibility. Defective alternators exhibited CPO-HI values above 1.5 relative to 0.75–0.9 for healthy units, while defective steering racks displayed inner-race order-band energies near 5.5 compared with 1.8 for nominal samples. Quantitative validation demonstrates >53% noise suppression relative to amplitude-based baselines and ROC-AUC values above 0.92 for defect classification. Cross-asset generalization was confirmed through Kolmogorov–Smirnov significance tests. The results establish a transferable and computationally lightweight NVH screening system capable of supporting intelligent, data-driven EOL operations.