Vehicle Model Calibration in the Frequency Domain and its Application to Large-Scale IRI Estimation

Abstract

A smartphone-based Dynamic Response Intelligent Monitoring System (iDRIMS) was developed to conduct road evaluations with high efficiency and reasonable accuracy [1]. iDRIMS estimates the International Roughness Index (IRI) based on vehicle responses measured with an iOS application, which obtains three-axis acceleration, angular velocity, and GPS with accurate sampling timing. However, the robustness and accuracy was limited. In this paper, the iDRIMS was improved mainly by employing frequency domain analysis. The algorithm consists of two steps. First, a half car (HC) model was selected as the vehicle model, and vehicle parameters were identified through driving tests over a portable hump of known size. In contrast to the previous approach of parameter identification in the time domain using Unscented Kalman Filter, the parameters were optimized to minimize the difference between the simulation and measured hump responses in the frequency domain, using a genetic algorithm. Then, IRI was estimated by measuring the vertical acceleration responses of ordinary vehicles. The measured acceleration was converted into the acceleration root mean square (RMS) of the sprung mass of a standard quarter car (QC) by multiplying a transfer function. The transfer function, estimated through the simulation of the identified HC model, as opposed to QC model in previous approaches, reflected the vehicle pitching motions and sensor installation location. The RMS was further converted to IRI based on the correlation between these values. Numerical simulation was conducted to investigate the performance in terms of various driving speeds and sensor locations. The experiment was conducted at a 13 km road by comparing three types of vehicles and a profiler. Inaccurate IRI estimation at the speed change section was experimentally investigated and compensated. Furthermore, the improved method was applied to 72 vehicles that were driven more than 180,000 km per year. A data collection and analysis platform was built, which successfully collected and analyzed large-scale data with high efficiency. The results from both numerical simulation and real case application show that the improved method accurately estimates IRI with high robustness and efficiency.