Reliable in-situ monitoring of Laser Powder Bed Fusion (LPBF) remains challenging due to data drift arising from variations in material composition, scan parameters, and sensor conditions. To address this, we propose an explainable learnable wavelet scattering (LWS) framework that enhances the information richness of acoustic emission (AE) signals and ensures cross-domain generalization through zero-shot transfer learning. The model employs a trainable Morlet wavelet bank that adaptively refines its center frequencies and bandwidths to capture process-specific spectral patterns. The resulting multi-scale scattering features are projected into a low-dimensional latent space and classified into melt-pool regimes—Lack of Fusion(LoF), conduction, and keyhole. Bayesian optimization yielded an optimal parameter configuration achieving over ~97% validation accuracy with stable convergence. Causal frequency perturbation analysis revealed that keyhole-mode dynamics predominantly manifest in low-frequency bands, whereas conductionand LoFsignatures occupy mid-and higher-frequency regions. This frequency-resolved explainability provides practical guidance for sensor selection, identifying the spectral bands that carry the most discriminative information and enabling targeted tuning of acoustic sensors for improved sensitivity and robustness. The learned filters exhibited strong physical interpretability, converging toward frequency ranges most responsive to melt-pool transitions. Zero-shot transfer testing on an unseendataset confirmed high domain-invariant performance without retraining. Overall, this study establishes LWSas a physics-consistent and interpretable approach for AE-based monitoring of LPBF, capable of maintaining robustness under data drift while enabling generalizable, explainable, and sensor-informed process-state recognition