Sensitivity-guided modeling of exhaust noise in large-bore marine engines using 1D acoustic simulation

Abstract

Accurate prediction of exhaust noise is increasingly important for integrating acoustic constraints into model-based development of large-bore marine engines. However, one-dimensional acoustic modeling practices established primarily for automotive engines may not transfer directly to large-bore systems. This study investigates the sensitivity of predicted exhaust-pressure spectra to numerical resolution, downstream exhaust-stack representation, turbocharger acoustic modeling, and local duct geometry using a one-dimensional model of a Wärtsilä 4L20 medium-speed engine. A configuration combining the selected refinements is then evaluated against measured in-duct pressure spectra. Numerical resolution produced the strongest sensitivity, with coupled spatial–temporal refinement reducing the spectral RMSE relative to the finest-mesh reference from 10.7 to 2.3 dB below 600 Hz. Explicit exhaust-stack representation substantially redistributed spectral energy, while an acoustic turbine representation introduced frequency-dependent transmission loss rising to 15–17 dB near 1 kHz. The combined refined configuration produced closer frequency-resolved agreement immediately downstream of the turbine, where the narrowband RMSE decreased from 11.42 to 9.88 dB and the mean absolute error from 9.66 to 7.57 dB, although the mean-level deviation increased from 2.42 to 3.05 dB. Within the plane-wave-valid range, one-third-octave errors also decreased, and a separate comparison at a farther-downstream station gave a mean absolute error of 7.30 dB, although residual narrowband discrepancies of 9–10 dB remained. The results demonstrate that reliable one-dimensional exhaust-acoustic prediction for the investigated large-bore engine requires adequate numerical wave resolution together with representation of the dominant propagation paths and passive turbine acoustics. However, the remaining spectral errors indicate that further model development and validation are required. The quantitative resolution thresholds and relative importance of individual model components remain platform-specific and require validation across additional engines and operating conditions.