Performance comparison of discrete kalman filter and dynamic programming technique for pavement roughness identification

Roughness profile identification is indispensable in vehicle dynamics research since it is the major source of excitation to the vehicle body. The present paper deals with pavement roughness identification based on the dynamic response of a quarter car vehicle model. The forward problem involves computing the vehicle responses using Newmark’s beta method. Noise in the measured data poses a serious challenge to the roughness identification problem. In the present paper, the computed dynamic responses are contaminated with Gaussian white noise in order to simulate actual field measurements. Two techniques are employed for identification, and their efficiency is compared. Roughness identification based on dynamic programming technique involves the formulation of an objective function that computes the least square error between measured and identified vehicle responses. Tikhonov regularization is incorporated to deal with the ill-posed inverse problem. Bellman’s principle of optimality is employed to minimize the objective function and compute the pavement roughness profile. In case of discrete Kalman filter-based identification, an optimal filter considering roughness as an unknown input rather than as a state variable is adopted. The efficiency of both methods in dealing with measurement error and estimation accuracy is compared. The computational efficiency and the feasibility of the two methods in tackling different scenarios in the identification problem are also demonstrated.