This data descriptor presents a compact acoustic feature dataset derived from an open simulation-based study on electric vehicle gearbox housings with different structural stiffness levels. The dataset contains band-averaged sound pressure level (SPL) features extracted from radiated noise spectra of three housing concepts—flexible, intermediate, and rigid—differing only in ribbing configuration. Frequency-domain SPL spectra in the 1–6 kHz range were partitioned into five one-kilohertz bands, yielding a five-dimensional acoustic feature vector for each housing–microphone combination. In total, twelve feature vectors are provided, accompanied by stiffness labels and metadata describing the underlying simulation context. In addition to the dataset itself, baseline exploratory analyses are reported to illustrate potential reuse scenarios. Principal component analysis and unsupervised clustering demonstrate that mid-frequency bands, particularly between 2 and 4 kHz, exhibit sensitivity to housing stiffness, whereas total integrated spectral energy shows limited discriminative power. These analyses are intended to be illustrative examples rather than predictive models, given the deliberately small dataset size. The dataset is designed for reuse in benchmarking dimensionality reduction methods, clustering algorithms, uncertainty-aware classifications, and educational demonstrations of small-sample NVH data analysis. By providing a transparent and lightweight acoustic feature representation, this contribution supports reproducible research and early-stage comparative studies in drivetrain noise and vibration analysis.
Krisztian Horvath (Thu,) studied this question.