Adaptive Hybrid Model-Based and Data-Driven Clamping Force Estimation for Electromechanical Brake Systems

Accurate clamping force estimation is essential for reliable and efficient control of electromechanical brake systems (EMBs). Existing model-based estimators suffer from degraded robustness under modeling uncertainties, while data-driven approaches often lack generalization when facing time-varying nonlinearities and limited training data. To overcome these limitations, this article proposes a novel adaptive hybrid clamping force estimation framework that tightly integrates an extended Kalman filter (EKF) with a lightweight temporal convolutional network (TCN). Unlike previous approaches, the proposed method fully exploits the complementary strengths of both paradigms: the EKF provides physically grounded estimation with strong robustness to disturbances, whereas TCN rapidly adapts to unmodeled dynamics and nonlinear variations. Furthermore, a new adaptive weighted fusion mechanism is designed to adjust the contribution of the two estimators in real time based on signal characteristics and estimation errors, enabling improved stability and convergence across diverse operating conditions. Offline evaluations on representative hardware-in-the-loop (HIL) datasets demonstrate that the proposed method achieves superior accuracy and robustness compared with state-of-the-art baselines.