Improvement of Inertial Profiler Measurements of Urban and Low-Speed Roadways

Abstract

Inertial profilers can measure the longitudinal road elevation profile and the International Roughness Index (IRI) accurately when they are operated under favorable conditions. However, their performance deteriorates when they experience disturbances such as lateral and longitudinal accelerations, and when the profiler host vehicle travels very slowly or comes to a stop. Preliminary analytical and experimental work confirmed two major sources of measurement errors. First, slowly varying bias in vertical acceleration measurements causes drift in the signals derived from them after they are integrated twice to form a component of the measured profile. When the profiler host vehicle decelerates to very low speed or comes to a stop, the drift appears directly in the profile and causes a large artificial change in elevation that is concentrated over a small travel distance. Second, the inertial sensors experience dynamic changes in pitch and roll orientation as the profiler host vehicle reacts to driver inputs, such as braking and steering. Tilt of the sensitive axis causes contamination of the accelerometer signal by inputs along other axes. After double integration, this appears as errors in vertical curvature in the profile over the region where the braking or steering occurred. The errors reduce the accuracy of longitudinal profiles on urban road networks and low-speed roadways, and render the measurements of IRI unusable at important locations, such as intersections. This research proposed two solutions for addressing these measurement errors. First, the research investigated data processing algorithms that reduce measurement errors without the need for adding sensors to the typical inertial profiler design. These solutions combine specialized processing algorithms with standard filtering techniques to mitigate artificial roughness caused by drift and misalignment. Since no additional hardware is required, these algorithms offer low-cost options for immediate implementation within the existing fleet. The algorithms do not offer a complete solution, because they mitigate large upward biases in roughness measured at stops at the cost of reducing the validity of measured profile at low speed. Second, the research proposed and tested the use of additional sensors to improve profile measurement at low speed, during braking, and at stops. The augmented system includes inertial and GPS measurement of profiler kinematics. A multi-rate extended Kalman filter combines the inertial sensors with the GPS outputs to reduce drift and errors associated with profiler host-vehicle tilt. Use of a Rauch-Tung-Striebel smoother improves the mitigation of drift. A custom measurement system was designed and built for this research to enable an experimental evaluation of the proposed solutions. The performance of the error suppression algorithms and sensor augmentation proposed in this research was evaluated using the results from several test runs collected under challenging conditions, including operation at low speed, braking, and operation through a stop. For all test runs, performance is quantified using standard measures for accuracy of longitudinal profile and IRI used by road agencies for pavement network quality assurance and pavement network management. Use of inertial measurement in three dimensions and GPS measurement of profiler height and orientation is shown to give the best performance. The recommended processing algorithm integrates an additional mode of operation into the Kalman filter that applies an alternative measurement model at stops. Nearly equivalent performance was observed when the GPS outputs were replaced by artificial signals. This version of the system offers an option for measurement in urban canyons.