With the industry shift toward electrification, autonomy, and software-defined vehicles, the perceptual experience of passengers—particularly ride comfort—has become a core differentiator. While ride comfort and handling stability often exhibit a ...
With the industry shift toward electrification, autonomy, and software-defined vehicles, the perceptual experience of passengers—particularly ride comfort—has become a core differentiator. While ride comfort and handling stability often exhibit a trade-off, this study focuses exclusively on ride comfort and proposes a data-driven framework that links real-world measurements, vehicle dynamics reasoning, and shock-dyno/valve-code information to quantitative design guidance.
Experiments were conducted on a 2019 Kia Carnival Hi-Limousine with three suspension setups (OE, twin-tube, mono-tube). Vertical body acceleration at the seat rail was recorded with an integrated Data Acquisition (DAQ)/Inertial Measurement Unit (IMU) and 3-axis accelerometer. Signals were quantified per ISO 2631-1 using the weighted RMS acceleration (a_w), Vibration Dose Value (VDV), and Welch-PSD–based band energies; bump events were additionally summarized by peak shock and settling time (T_settle). A “One Voice” protocol aligned the subjective scoring from three expert evaluators, enabling consistent labels for model training.
On our dataset, OE tended to yield lower a_w and reduced mid/high-frequency PSD energy during steady-state highway driving (better residual-vibration isolation), whereas the mono-tube showed higher peak shock at bump entry but shorter T_settle(faster post-impact stabilization). The twin-tube generally exhibited balanced behavior between the two. Building on these observations, we developed an interpretable regression pipeline (regularized/agnostic, with cross-validation and simple bias correction) that predicts the consensus subjective score from a compact set of physical features. Model interpretation consistently highlighted residual-vibration magnitude (a_w) and T_settle as dominant determinants across scenarios.
To close the loop to tunable hardware, we assembled an internal valve-code–shock-dyno database and mapped dyno features high-speed compression, low-speed rebound, and compression/rebound ratio—to vehicle-level indices (a_w, T_settle). This enables inverse design: from a target ride envelope (e.g., “a_w low, T_settle short”), the framework proposes actionable valve-code edits (e.g., modest C_disc relaxation, one-step R_disc increase, S_orf hold/micro-adjust), which can be verified with a small follow-up drive.
The contribution is twofold: (i) a compact, reproducible quantification of ride comfort (a_w, VDV/PSD, T_settle) linked to expert scores, and (ii) a practical bridge from those metrics to valve-level tuning guidance via shock-dyno features. This reduces physical iteration and shortens development lead time while improving consistency of perceived ride. Limitations include a single vehicle, controlled routes, and idealized dyno conditions; future work will expand the dataset across segments and integrate road-preview (LiDAR/vision) for real-time semi-active/active suspension control.