DOI: 10.18245/ijaet.1922382 ISSN: 2146-9067

Confidence aware wheelbase preview MPC with road estimation for half-car active suspension

Haşmet Çağrı Sezgen
Show PDF Cite
The value of wheelbase preview in half-car active suspension depends on how reliably preview information can be reconstructed as road excitation changes. A confidence-aware wheelbase-preview model predictive control strategy is formulated for a four-degree-of-freedom half-car active suspension system subjected to random-road and impact-road inputs. The framework combines augmented state-road estimation with shift-register-based rear-preview generation. Preview usage is then adjusted online through an innovation-based confidence measure, while control increments are regularized within a constrained MPC formulation. Reconstruction quality differs clearly between the two simulated scenarios: road-estimation correlation falls from 0.9652 in the random-road case to 0.6935 in the impact-road case, and the mean confidence drops from 0.968 to 0.919. Preview is therefore useful, but its reliability is not stable enough for fixed use across operating conditions. In random-road simulations, rear acceleration root-mean-square (RMS) falls from 1.594 to 1.159 m/s², rear suspension deflection RMS from 7.82 to 5.93 mm, and rear relative dynamic load RMS from 0.235 to 0.203 relative to the passive baseline. For impact-road excitation, the controller provides the strongest improvements in rear-axle acceleration and dynamic-load response, attaining a rear acceleration RMS of 1.213 m/s², a rear acceleration P2P value of 13.269 m/s², and a rear relative dynamic load P2P value of 2.059 among the compared controllers. Changes at the front axle remain modest, whereas the clearest benefit appears at the rear, where preview is most directly relevant. Ablation results show that warm-up, confidence gating, and control-increment regularization improve different aspects of the response, leaving the full configuration as the most balanced overall trade-off rather than the best result in every metric. Mean solve times remain at 0.16 ms for the random-road case and 0.22 ms for the impact-road case, both below the selected 0.01 s sampling interval. Taken together, these results show that wheelbase preview is most effective when its contribution is adjusted according to estimation reliability rather than treated as uniformly trustworthy information.

More from our Archive