Development and calibration of a multi-segment 15-DOF biomechanical model for predicting seat and backrest apparent mass of the seated human body

Backrest inclination significantly influences the biodynamic response of seated occupants under whole-body vibration; however, existing models remain limited in their ability to simultaneously predict apparent mass (APM) responses at both the seat and the backrest across varying postures. This study develops and calibrates a 15-degree-of-freedom (DOF) multibody dynamics (MBD) model of the seated human body to quantify the effect of backrest inclination (0°, 10°, 20°, and 30°) under vertical excitation. The model consists of five rigid segments—thighs, pelvis, lower torso, upper torso, and head–neck—connected through translational and rotational spring–damper joints, with the body-seat and body-backrest interactions represented by translational spring-damper elements at three contact locations. A total of 36 model parameters are identified via a genetic algorithm by minimising the least-squares error between predicted and experimental responses. The model accurately reproduces the vertical apparent mass at the seat across all inclinations, with mean errors below 2.85% and coefficients of determination (R 2 ) exceeding 0.97. The best overall vertical performance occurs at 0°, with a mean error of 1.30%, root-mean-square (RMS) error of 1.70%, and R 2 = 0.997. Strong agreement is also obtained for the normal apparent mass at the backrest, with mean errors below 6.98% and RMS errors below 8.56%. The lowest RMS error and peak error occur at 30°, with RMS error of 6.73% and peak error of 1.15%, while the highest R 2 occurs at 0° with R 2 = 0.972. The fore-aft cross-axis response remains the most difficult to predict, with larger errors than the vertical and normal responses. Its best performance is obtained at 10°, with a mean error of 11.54%, RMS error of 14.70%, and R 2 = 0.947. Across all response directions, the model predicts the primary resonance frequency with errors below 9.65%, indicating reliable resonance prediction over the full range of backrest inclinations. Global sensitivity analyses (GSA) across all four backrest inclinations identified posture-dependent influential parameter subsets. The collective total-order Sobol contribution of the Morris low-sensitivity group remained modest, with a maximum point estimate of 0.091 across the examined responses. Sensitivity-informed recalibration reduced the number of adjustable coefficients from 36 to 13–18, with negligible changes in the objective function relative to the corresponding full 36-parameter calibrations. The model should therefore be interpreted as a condition-specific linear engineering representation calibrated for vertical excitation at 0.5 m/s 2 r.m.s. over 0.5–15 Hz; predictive performance at other excitation magnitudes is not established in the present study.