A Condition-Aware Data-Driven Framework for Tire–Road Noise Estimation Under Limited Data Conditions

Tire–road noise is a major contributor to traffic noise and is strongly influenced by tire properties, road surface characteristics, and driving conditions. Accurate estimation of tire–road noise remains challenging, especially under limited data availability and varying operating conditions. This study proposes a condition-aware data-driven framework for tire–road noise estimation based on multi-sensor measurements. The proposed approach integrates On-Board Sound Intensity (OBSI), tire cavity noise, and vehicle CAN bus signals using a multi-scale feature extraction strategy. One-third-octave Sound Pressure Level (SPL) representations are employed to capture the dominant spectral characteristics of tire–road interactions. To improve model stability and reduce overfitting under limited-data conditions, an ensemble learning framework is adopted. In addition, a Leave-One-Group-Out (LOGO) validation strategy is used to ensure realistic performance evaluation across different tires, road surfaces, speeds, and acceleration scenarios. Multi-task learning is applied to jointly estimate OBSI and tire cavity noise. Experiments are conducted using four instrumented vehicles and multiple tire types under standardized road conditions. ... mehrThe results indicate stable and consistent performance across diverse operating conditions, particularly within the 400–5000 Hz frequency range where tire–road noise energy is most dominant. Overall, this work provides a condition-aware and reproducible data-driven methodology for tire–road noise estimation and provides a basis for future large-scale monitoring and data-driven noise analysis applications.