Simultaneous estimation of state and unknown road roughness input for vehicle suspension control system based on discrete Kalman filter

Abstract

This study presents an improved discrete Kalman filter for simultaneously estimating both all state variables and the unknown road roughness input for a vehicle suspension control system that plays a key role in the ride quality and handling performance while driving the vehicle. The suspension system is influenced by the road roughness input, which causes undesirable vibrations associated with vehicle instability. It is therefore important to estimate the road roughness and state variables information when designing the model-based controller for the vehicle suspension control system associated with the vehicle’s vertical dynamics. However, the implementation of conventional estimation theories for the suspension control system is challenging because the road roughness acts as an unknown input and is difficult to be measured or estimated while driving. This study presents an improved Kalman filter with unknown input, which can simultaneously estimate the state variables and road roughness without any prior information about the vehicle suspension control system. The proposed road roughness input estimator is evaluated by using an in-vehicle test bed with a laser-type profilometer. Finally, the state estimation performance of the proposed estimator for a vehicle suspension control system is validated by using CarSim software.

Deep Learning-Based Estimation of the Unknown Road Profile and State Variables for the Vehicle Suspension System

“…Although the linear quarter car suspension model offers the advantage of low computational cost in KF algorithms, it incorporates model uncertainties as it only considers the vehicle's heave motion. This study validates that our proposed method outperforms the quarter car modelbased EKF-UI algorithm [9] in estimation accuracy and computation time. Further details are provided in Section III-B.…”

“…All data used in this study are acquired via simulation in the CarSim environment [21], as it is reported to produce highly relevant results compared with actual field test experiments [9], [22], [23]. We first design a random F class road profile according to the ISO 8608 standard [24] for the input to the CarSim vehicle model.…”

“…The KF algorithm is based upon the FIGURE 1. Strategies for establishing the observer systems used in this study: (a) KF-based (the EKF-UI [9]), (b) DL-based (our proposed model, the static NARX model [10]). physical modeling of the vehicle suspension system, of which an offline procedure is required beforehand to identify the parameters of the modeled system.…”

“…A schematic of the above procedures is illustrated in Figure 1 (a). While studies have validated its usage with field test results [9], [12], several setbacks are inherent regarding its working principles. First, model uncertainty is evident as previous studies have mainly utilized the simplified linear quarter car suspension model [13], in which the nonlinearity in the suspension system is considered linear, and the pitch and roll motions of the vehicle body are neglected.…”

“…Furthermore, we validate that the interpretation of the attention values is in-line with the suspension system's physical characteristics, thus providing a valid basis for model reliability. We evaluate qualitative and quantitative results with the extended Kalman filter with unknown inputs (EKF-UI) [9] and the static NARX model [10] to examine the estimation capability of our proposed method in comparison with the existing studies. Lastly, we demonstrate the feasibility of sequence estimation, which estimates a sequence of future values at each time step, and has not been proceeded by previous works.…”