Recursive total least squares with improved parameter tracking: Application to model-based vehicle mass estimation

Abstract

Vehicle mass plays an influential role in various dynamical systems for vehicle safety and control. In this work, a novel recursive total least squares (RTLS) solution is presented for the online estimation of gross vehicle mass. The proposed method requires access to engine torque, engine speed, wheel speed, and vehicle IMU acceleration measurements. Different algorithm configurations are considered for mass estimation of internal combustion engine and electric vehicles, with a focused application to passenger cars and light commercial vehicles. The baseline RTLS algorithm is improved by means of regularization, outlier attenuation, parameter projection, and enhanced tracking of jumping parameters, all of which play an important role in optimizing estimator performance for industrial applications. The proposed algorithm is then generalized to account for heterogeneous and heteroscedastic measurement noise with a recursive noise covariance estimation algorithm. The method is tested against two well-known benchmark algorithms from the mass estimation literature with experimental electric vehicle data, and solution sensitivity to model assumptions and model input parameters is discussed. The vehicle experiments show that the proposed method outperforms the benchmark methods in terms of accuracy and convergence characteristics.