A hybrid strategy of OTPA and machine learning for efficient root-cause analysis of NVH in multi-source systems

• Novel integration of machine learning in Operational Transfer Path Analysis (OTPA). • Random Forest regression preserves critical data, outperforming traditional SVD methods. • Achieved 2.28 % improvement in interior noise levels using ML-optimized paths. • Feature importance analysis identifies dominant noise and vibration sources. • Validated cost-effective modifications enhance NVH characteristics by 7.15 %. The growing awareness of health benefits, along with the competitive emphasis on vehicle comfort, has led automakers to place greater attention on reducing Noise, Vibration, and Harshness (NVH). One of the most beneficial techniques for NVH engineers to identify, rank, and eliminate dominant noise and vibration sources and paths is Transfer Path Analysis (TPA). Unlike traditional TPA, Operational Transfer Path Analysis (OTPA) requires neither the preliminary acquisition of the transfer matrix between excitation and response points nor the measurement of forces transferred through the active and passive side connection points. Although the OTPA method offers significant advantages over classical TPA methods, it still faces challenges such as data loss caused by the pseudo-inversion of the indicator matrix. In this paper, we estimate the transmissibility matrix using a machine learning-based regression algorithm (random forest). We demonstrated that machine learning is an effective alternative to the truncated Singular Value Decomposition (SVD) method for estimating the transmissibility matrix, as it is a swift solution that preserves essential information in the indicator matrix. The efficiency of the method has been verified by a 2.28 % improvement in the Sound Pressure Level (SPL) of the driver’s ear noise of a sedan-type vehicle through the modification of the most critical path found by this approach.